Chromium Code Reviews
chromiumcodereview-hr@appspot.gserviceaccount.com (chromiumcodereview-hr) | Please choose your nickname with Settings | Help | Chromium Project | Gerrit Changes | Sign out
(378)

Unified Diff: webrtc/media/webrtc/webrtcvoiceengine.cc

Issue 1684163002: Rename webrtc/media/webrtc -> webrtc/media/engine (Closed) Base URL: https://chromium.googlesource.com/external/webrtc.git@master
Patch Set: Rebase f396f6085f9e4f16f37471a7828e3e31308c0d52 #11590 Created 4 years, 10 months ago
Use n/p to move between diff chunks; N/P to move between comments. Draft comments are only viewable by you.
Jump to:
View side-by-side diff with in-line comments
Download patch
« no previous file with comments | « webrtc/media/webrtc/webrtcvoiceengine.h ('k') | webrtc/media/webrtc/webrtcvoiceengine_unittest.cc » ('j') | no next file with comments »
Expand Comments ('e') | Collapse Comments ('c') | Show Comments Hide Comments ('s')
Index: webrtc/media/webrtc/webrtcvoiceengine.cc
diff --git a/webrtc/media/webrtc/webrtcvoiceengine.cc b/webrtc/media/webrtc/webrtcvoiceengine.cc
deleted file mode 100644
index 82aa12ef0671919b9b5884401e1a18b97c68a434..0000000000000000000000000000000000000000
--- a/webrtc/media/webrtc/webrtcvoiceengine.cc
+++ /dev/null
@@ -1,2562 +0,0 @@
-/*
- * Copyright (c) 2004 The WebRTC project authors. All Rights Reserved.
- *
- * Use of this source code is governed by a BSD-style license
- * that can be found in the LICENSE file in the root of the source
- * tree. An additional intellectual property rights grant can be found
- * in the file PATENTS. All contributing project authors may
- * be found in the AUTHORS file in the root of the source tree.
- */
-
-#ifdef HAVE_CONFIG_H
-#include <config.h>
-#endif
-
-#ifdef HAVE_WEBRTC_VOICE
-
-#include "webrtc/media/webrtc/webrtcvoiceengine.h"
-
-#include <algorithm>
-#include <cstdio>
-#include <string>
-#include <vector>
-
-#include "webrtc/audio/audio_sink.h"
-#include "webrtc/base/arraysize.h"
-#include "webrtc/base/base64.h"
-#include "webrtc/base/byteorder.h"
-#include "webrtc/base/common.h"
-#include "webrtc/base/helpers.h"
-#include "webrtc/base/logging.h"
-#include "webrtc/base/stringencode.h"
-#include "webrtc/base/stringutils.h"
-#include "webrtc/call/rtc_event_log.h"
-#include "webrtc/common.h"
-#include "webrtc/media/base/audioframe.h"
-#include "webrtc/media/base/audiorenderer.h"
-#include "webrtc/media/base/constants.h"
-#include "webrtc/media/base/streamparams.h"
-#include "webrtc/media/webrtc/webrtcmediaengine.h"
-#include "webrtc/media/webrtc/webrtcvoe.h"
-#include "webrtc/modules/audio_coding/acm2/rent_a_codec.h"
-#include "webrtc/modules/audio_processing/include/audio_processing.h"
-#include "webrtc/system_wrappers/include/field_trial.h"
-#include "webrtc/system_wrappers/include/trace.h"
-
-namespace cricket {
-namespace {
-
-const int kDefaultTraceFilter = webrtc::kTraceNone | webrtc::kTraceTerseInfo |
- webrtc::kTraceWarning | webrtc::kTraceError |
- webrtc::kTraceCritical;
-const int kElevatedTraceFilter = kDefaultTraceFilter | webrtc::kTraceStateInfo |
- webrtc::kTraceInfo;
-
-// On Windows Vista and newer, Microsoft introduced the concept of "Default
-// Communications Device". This means that there are two types of default
-// devices (old Wave Audio style default and Default Communications Device).
-//
-// On Windows systems which only support Wave Audio style default, uses either
-// -1 or 0 to select the default device.
-#ifdef WIN32
-const int kDefaultAudioDeviceId = -1;
-#else
-const int kDefaultAudioDeviceId = 0;
-#endif
-
-// Parameter used for NACK.
-// This value is equivalent to 5 seconds of audio data at 20 ms per packet.
-const int kNackMaxPackets = 250;
-
-// Codec parameters for Opus.
-// draft-spittka-payload-rtp-opus-03
-
-// Recommended bitrates:
-// 8-12 kb/s for NB speech,
-// 16-20 kb/s for WB speech,
-// 28-40 kb/s for FB speech,
-// 48-64 kb/s for FB mono music, and
-// 64-128 kb/s for FB stereo music.
-// The current implementation applies the following values to mono signals,
-// and multiplies them by 2 for stereo.
-const int kOpusBitrateNb = 12000;
-const int kOpusBitrateWb = 20000;
-const int kOpusBitrateFb = 32000;
-
-// Opus bitrate should be in the range between 6000 and 510000.
-const int kOpusMinBitrate = 6000;
-const int kOpusMaxBitrate = 510000;
-
-// Default audio dscp value.
-// See http://tools.ietf.org/html/rfc2474 for details.
-// See also http://tools.ietf.org/html/draft-jennings-rtcweb-qos-00
-const rtc::DiffServCodePoint kAudioDscpValue = rtc::DSCP_EF;
-
-// Ensure we open the file in a writeable path on ChromeOS and Android. This
-// workaround can be removed when it's possible to specify a filename for audio
-// option based AEC dumps.
-//
-// TODO(grunell): Use a string in the options instead of hardcoding it here
-// and let the embedder choose the filename (crbug.com/264223).
-//
-// NOTE(ajm): Don't use hardcoded paths on platforms not explicitly specified
-// below.
-#if defined(CHROMEOS)
-const char kAecDumpByAudioOptionFilename[] = "/tmp/audio.aecdump";
-#elif defined(ANDROID)
-const char kAecDumpByAudioOptionFilename[] = "/sdcard/audio.aecdump";
-#else
-const char kAecDumpByAudioOptionFilename[] = "audio.aecdump";
-#endif
-
-// Constants from voice_engine_defines.h.
-const int kMinTelephoneEventCode = 0; // RFC4733 (Section 2.3.1)
-const int kMaxTelephoneEventCode = 255;
-const int kMinTelephoneEventDuration = 100;
-const int kMaxTelephoneEventDuration = 60000; // Actual limit is 2^16
-
-class ProxySink : public webrtc::AudioSinkInterface {
- public:
- ProxySink(AudioSinkInterface* sink) : sink_(sink) { RTC_DCHECK(sink); }
-
- void OnData(const Data& audio) override { sink_->OnData(audio); }
-
- private:
- webrtc::AudioSinkInterface* sink_;
-};
-
-bool ValidateStreamParams(const StreamParams& sp) {
- if (sp.ssrcs.empty()) {
- LOG(LS_ERROR) << "No SSRCs in stream parameters: " << sp.ToString();
- return false;
- }
- if (sp.ssrcs.size() > 1) {
- LOG(LS_ERROR) << "Multiple SSRCs in stream parameters: " << sp.ToString();
- return false;
- }
- return true;
-}
-
-// Dumps an AudioCodec in RFC 2327-ish format.
-std::string ToString(const AudioCodec& codec) {
- std::stringstream ss;
- ss << codec.name << "/" << codec.clockrate << "/" << codec.channels
- << " (" << codec.id << ")";
- return ss.str();
-}
-
-std::string ToString(const webrtc::CodecInst& codec) {
- std::stringstream ss;
- ss << codec.plname << "/" << codec.plfreq << "/" << codec.channels
- << " (" << codec.pltype << ")";
- return ss.str();
-}
-
-bool IsCodec(const AudioCodec& codec, const char* ref_name) {
- return (_stricmp(codec.name.c_str(), ref_name) == 0);
-}
-
-bool IsCodec(const webrtc::CodecInst& codec, const char* ref_name) {
- return (_stricmp(codec.plname, ref_name) == 0);
-}
-
-bool FindCodec(const std::vector<AudioCodec>& codecs,
- const AudioCodec& codec,
- AudioCodec* found_codec) {
- for (const AudioCodec& c : codecs) {
- if (c.Matches(codec)) {
- if (found_codec != NULL) {
- *found_codec = c;
- }
- return true;
- }
- }
- return false;
-}
-
-bool VerifyUniquePayloadTypes(const std::vector<AudioCodec>& codecs) {
- if (codecs.empty()) {
- return true;
- }
- std::vector<int> payload_types;
- for (const AudioCodec& codec : codecs) {
- payload_types.push_back(codec.id);
- }
- std::sort(payload_types.begin(), payload_types.end());
- auto it = std::unique(payload_types.begin(), payload_types.end());
- return it == payload_types.end();
-}
-
-// Return true if codec.params[feature] == "1", false otherwise.
-bool IsCodecFeatureEnabled(const AudioCodec& codec, const char* feature) {
- int value;
- return codec.GetParam(feature, &value) && value == 1;
-}
-
-// Use params[kCodecParamMaxAverageBitrate] if it is defined, use codec.bitrate
-// otherwise. If the value (either from params or codec.bitrate) <=0, use the
-// default configuration. If the value is beyond feasible bit rate of Opus,
-// clamp it. Returns the Opus bit rate for operation.
-int GetOpusBitrate(const AudioCodec& codec, int max_playback_rate) {
- int bitrate = 0;
- bool use_param = true;
- if (!codec.GetParam(kCodecParamMaxAverageBitrate, &bitrate)) {
- bitrate = codec.bitrate;
- use_param = false;
- }
- if (bitrate <= 0) {
- if (max_playback_rate <= 8000) {
- bitrate = kOpusBitrateNb;
- } else if (max_playback_rate <= 16000) {
- bitrate = kOpusBitrateWb;
- } else {
- bitrate = kOpusBitrateFb;
- }
-
- if (IsCodecFeatureEnabled(codec, kCodecParamStereo)) {
- bitrate *= 2;
- }
- } else if (bitrate < kOpusMinBitrate || bitrate > kOpusMaxBitrate) {
- bitrate = (bitrate < kOpusMinBitrate) ? kOpusMinBitrate : kOpusMaxBitrate;
- std::string rate_source =
- use_param ? "Codec parameter \"maxaveragebitrate\"" :
- "Supplied Opus bitrate";
- LOG(LS_WARNING) << rate_source
- << " is invalid and is replaced by: "
- << bitrate;
- }
- return bitrate;
-}
-
-// Returns kOpusDefaultPlaybackRate if params[kCodecParamMaxPlaybackRate] is not
-// defined. Returns the value of params[kCodecParamMaxPlaybackRate] otherwise.
-int GetOpusMaxPlaybackRate(const AudioCodec& codec) {
- int value;
- if (codec.GetParam(kCodecParamMaxPlaybackRate, &value)) {
- return value;
- }
- return kOpusDefaultMaxPlaybackRate;
-}
-
-void GetOpusConfig(const AudioCodec& codec, webrtc::CodecInst* voe_codec,
- bool* enable_codec_fec, int* max_playback_rate,
- bool* enable_codec_dtx) {
- *enable_codec_fec = IsCodecFeatureEnabled(codec, kCodecParamUseInbandFec);
- *enable_codec_dtx = IsCodecFeatureEnabled(codec, kCodecParamUseDtx);
- *max_playback_rate = GetOpusMaxPlaybackRate(codec);
-
- // If OPUS, change what we send according to the "stereo" codec
- // parameter, and not the "channels" parameter. We set
- // voe_codec.channels to 2 if "stereo=1" and 1 otherwise. If
- // the bitrate is not specified, i.e. is <= zero, we set it to the
- // appropriate default value for mono or stereo Opus.
-
- voe_codec->channels = IsCodecFeatureEnabled(codec, kCodecParamStereo) ? 2 : 1;
- voe_codec->rate = GetOpusBitrate(codec, *max_playback_rate);
-}
-
-webrtc::AudioState::Config MakeAudioStateConfig(VoEWrapper* voe_wrapper) {
- webrtc::AudioState::Config config;
- config.voice_engine = voe_wrapper->engine();
- return config;
-}
-
-class WebRtcVoiceCodecs final {
- public:
- // TODO(solenberg): Do this filtering once off-line, add a simple AudioCodec
- // list and add a test which verifies VoE supports the listed codecs.
- static std::vector<AudioCodec> SupportedCodecs() {
- LOG(LS_INFO) << "WebRtc VoiceEngine codecs:";
- std::vector<AudioCodec> result;
- for (webrtc::CodecInst voe_codec : webrtc::acm2::RentACodec::Database()) {
- // Change the sample rate of G722 to 8000 to match SDP.
- MaybeFixupG722(&voe_codec, 8000);
- // Skip uncompressed formats.
- if (IsCodec(voe_codec, kL16CodecName)) {
- continue;
- }
-
- const CodecPref* pref = NULL;
- for (size_t j = 0; j < arraysize(kCodecPrefs); ++j) {
- if (IsCodec(voe_codec, kCodecPrefs[j].name) &&
- kCodecPrefs[j].clockrate == voe_codec.plfreq &&
- kCodecPrefs[j].channels == voe_codec.channels) {
- pref = &kCodecPrefs[j];
- break;
- }
- }
-
- if (pref) {
- // Use the payload type that we've configured in our pref table;
- // use the offset in our pref table to determine the sort order.
- AudioCodec codec(
- pref->payload_type, voe_codec.plname, voe_codec.plfreq,
- voe_codec.rate, voe_codec.channels,
- static_cast<int>(arraysize(kCodecPrefs)) - (pref - kCodecPrefs));
- LOG(LS_INFO) << ToString(codec);
- if (IsCodec(codec, kIsacCodecName)) {
- // Indicate auto-bitrate in signaling.
- codec.bitrate = 0;
- }
- if (IsCodec(codec, kOpusCodecName)) {
- // Only add fmtp parameters that differ from the spec.
- if (kPreferredMinPTime != kOpusDefaultMinPTime) {
- codec.params[kCodecParamMinPTime] =
- rtc::ToString(kPreferredMinPTime);
- }
- if (kPreferredMaxPTime != kOpusDefaultMaxPTime) {
- codec.params[kCodecParamMaxPTime] =
- rtc::ToString(kPreferredMaxPTime);
- }
- codec.SetParam(kCodecParamUseInbandFec, 1);
- codec.AddFeedbackParam(
- FeedbackParam(kRtcpFbParamTransportCc, kParamValueEmpty));
-
- // TODO(hellner): Add ptime, sprop-stereo, and stereo
- // when they can be set to values other than the default.
- }
- result.push_back(codec);
- } else {
- LOG(LS_WARNING) << "Unexpected codec: " << ToString(voe_codec);
- }
- }
- // Make sure they are in local preference order.
- std::sort(result.begin(), result.end(), &AudioCodec::Preferable);
- return result;
- }
-
- static bool ToCodecInst(const AudioCodec& in,
- webrtc::CodecInst* out) {
- for (webrtc::CodecInst voe_codec : webrtc::acm2::RentACodec::Database()) {
- // Change the sample rate of G722 to 8000 to match SDP.
- MaybeFixupG722(&voe_codec, 8000);
- AudioCodec codec(voe_codec.pltype, voe_codec.plname, voe_codec.plfreq,
- voe_codec.rate, voe_codec.channels, 0);
- bool multi_rate = IsCodecMultiRate(voe_codec);
- // Allow arbitrary rates for ISAC to be specified.
- if (multi_rate) {
- // Set codec.bitrate to 0 so the check for codec.Matches() passes.
- codec.bitrate = 0;
- }
- if (codec.Matches(in)) {
- if (out) {
- // Fixup the payload type.
- voe_codec.pltype = in.id;
-
- // Set bitrate if specified.
- if (multi_rate && in.bitrate != 0) {
- voe_codec.rate = in.bitrate;
- }
-
- // Reset G722 sample rate to 16000 to match WebRTC.
- MaybeFixupG722(&voe_codec, 16000);
-
- // Apply codec-specific settings.
- if (IsCodec(codec, kIsacCodecName)) {
- // If ISAC and an explicit bitrate is not specified,
- // enable auto bitrate adjustment.
- voe_codec.rate = (in.bitrate > 0) ? in.bitrate : -1;
- }
- *out = voe_codec;
- }
- return true;
- }
- }
- return false;
- }
-
- static bool IsCodecMultiRate(const webrtc::CodecInst& codec) {
- for (size_t i = 0; i < arraysize(kCodecPrefs); ++i) {
- if (IsCodec(codec, kCodecPrefs[i].name) &&
- kCodecPrefs[i].clockrate == codec.plfreq) {
- return kCodecPrefs[i].is_multi_rate;
- }
- }
- return false;
- }
-
- // If the AudioCodec param kCodecParamPTime is set, then we will set it to
- // codec pacsize if it's valid, or we will pick the next smallest value we
- // support.
- // TODO(Brave): Query supported packet sizes from ACM when the API is ready.
- static bool SetPTimeAsPacketSize(webrtc::CodecInst* codec, int ptime_ms) {
- for (const CodecPref& codec_pref : kCodecPrefs) {
- if ((IsCodec(*codec, codec_pref.name) &&
- codec_pref.clockrate == codec->plfreq) ||
- IsCodec(*codec, kG722CodecName)) {
- int packet_size_ms = SelectPacketSize(codec_pref, ptime_ms);
- if (packet_size_ms) {
- // Convert unit from milli-seconds to samples.
- codec->pacsize = (codec->plfreq / 1000) * packet_size_ms;
- return true;
- }
- }
- }
- return false;
- }
-
- static const AudioCodec* GetPreferredCodec(
- const std::vector<AudioCodec>& codecs,
- webrtc::CodecInst* voe_codec,
- int* red_payload_type) {
- RTC_DCHECK(voe_codec);
- RTC_DCHECK(red_payload_type);
- // Select the preferred send codec (the first non-telephone-event/CN codec).
- for (const AudioCodec& codec : codecs) {
- *red_payload_type = -1;
- if (IsCodec(codec, kDtmfCodecName) || IsCodec(codec, kCnCodecName)) {
- // Skip telephone-event/CN codec, which will be handled later.
- continue;
- }
-
- // We'll use the first codec in the list to actually send audio data.
- // Be sure to use the payload type requested by the remote side.
- // "red", for RED audio, is a special case where the actual codec to be
- // used is specified in params.
- const AudioCodec* found_codec = &codec;
- if (IsCodec(*found_codec, kRedCodecName)) {
- // Parse out the RED parameters. If we fail, just ignore RED;
- // we don't support all possible params/usage scenarios.
- *red_payload_type = codec.id;
- found_codec = GetRedSendCodec(*found_codec, codecs);
- if (!found_codec) {
- continue;
- }
- }
- // Ignore codecs we don't know about. The negotiation step should prevent
- // this, but double-check to be sure.
- if (!ToCodecInst(*found_codec, voe_codec)) {
- LOG(LS_WARNING) << "Unknown codec " << ToString(*found_codec);
- continue;
- }
- return found_codec;
- }
- return nullptr;
- }
-
- private:
- static const int kMaxNumPacketSize = 6;
- struct CodecPref {
- const char* name;
- int clockrate;
- size_t channels;
- int payload_type;
- bool is_multi_rate;
- int packet_sizes_ms[kMaxNumPacketSize];
- };
- // Note: keep the supported packet sizes in ascending order.
- static const CodecPref kCodecPrefs[12];
-
- static int SelectPacketSize(const CodecPref& codec_pref, int ptime_ms) {
- int selected_packet_size_ms = codec_pref.packet_sizes_ms[0];
- for (int packet_size_ms : codec_pref.packet_sizes_ms) {
- if (packet_size_ms && packet_size_ms <= ptime_ms) {
- selected_packet_size_ms = packet_size_ms;
- }
- }
- return selected_packet_size_ms;
- }
-
- // Changes RTP timestamp rate of G722. This is due to the "bug" in the RFC
- // which says that G722 should be advertised as 8 kHz although it is a 16 kHz
- // codec.
- static void MaybeFixupG722(webrtc::CodecInst* voe_codec, int new_plfreq) {
- if (IsCodec(*voe_codec, kG722CodecName)) {
- // If the ASSERT triggers, the codec definition in WebRTC VoiceEngine
- // has changed, and this special case is no longer needed.
- RTC_DCHECK(voe_codec->plfreq != new_plfreq);
- voe_codec->plfreq = new_plfreq;
- }
- }
-
- static const AudioCodec* GetRedSendCodec(
- const AudioCodec& red_codec,
- const std::vector<AudioCodec>& all_codecs) {
- // Get the RED encodings from the parameter with no name. This may
- // change based on what is discussed on the Jingle list.
- // The encoding parameter is of the form "a/b"; we only support where
- // a == b. Verify this and parse out the value into red_pt.
- // If the parameter value is absent (as it will be until we wire up the
- // signaling of this message), use the second codec specified (i.e. the
- // one after "red") as the encoding parameter.
- int red_pt = -1;
- std::string red_params;
- CodecParameterMap::const_iterator it = red_codec.params.find("");
- if (it != red_codec.params.end()) {
- red_params = it->second;
- std::vector<std::string> red_pts;
- if (rtc::split(red_params, '/', &red_pts) != 2 ||
- red_pts[0] != red_pts[1] || !rtc::FromString(red_pts[0], &red_pt)) {
- LOG(LS_WARNING) << "RED params " << red_params << " not supported.";
- return nullptr;
- }
- } else if (red_codec.params.empty()) {
- LOG(LS_WARNING) << "RED params not present, using defaults";
- if (all_codecs.size() > 1) {
- red_pt = all_codecs[1].id;
- }
- }
-
- // Try to find red_pt in |codecs|.
- for (const AudioCodec& codec : all_codecs) {
- if (codec.id == red_pt) {
- return &codec;
- }
- }
- LOG(LS_WARNING) << "RED params " << red_params << " are invalid.";
- return nullptr;
- }
-};
-
-const WebRtcVoiceCodecs::CodecPref WebRtcVoiceCodecs::kCodecPrefs[12] = {
- { kOpusCodecName, 48000, 2, 111, true, { 10, 20, 40, 60 } },
- { kIsacCodecName, 16000, 1, 103, true, { 30, 60 } },
- { kIsacCodecName, 32000, 1, 104, true, { 30 } },
- // G722 should be advertised as 8000 Hz because of the RFC "bug".
- { kG722CodecName, 8000, 1, 9, false, { 10, 20, 30, 40, 50, 60 } },
- { kIlbcCodecName, 8000, 1, 102, false, { 20, 30, 40, 60 } },
- { kPcmuCodecName, 8000, 1, 0, false, { 10, 20, 30, 40, 50, 60 } },
- { kPcmaCodecName, 8000, 1, 8, false, { 10, 20, 30, 40, 50, 60 } },
- { kCnCodecName, 32000, 1, 106, false, { } },
- { kCnCodecName, 16000, 1, 105, false, { } },
- { kCnCodecName, 8000, 1, 13, false, { } },
- { kRedCodecName, 8000, 1, 127, false, { } },
- { kDtmfCodecName, 8000, 1, 126, false, { } },
-};
-} // namespace {
-
-bool WebRtcVoiceEngine::ToCodecInst(const AudioCodec& in,
- webrtc::CodecInst* out) {
- return WebRtcVoiceCodecs::ToCodecInst(in, out);
-}
-
-WebRtcVoiceEngine::WebRtcVoiceEngine()
- : voe_wrapper_(new VoEWrapper()),
- audio_state_(webrtc::AudioState::Create(MakeAudioStateConfig(voe()))) {
- Construct();
-}
-
-WebRtcVoiceEngine::WebRtcVoiceEngine(VoEWrapper* voe_wrapper)
- : voe_wrapper_(voe_wrapper) {
- Construct();
-}
-
-void WebRtcVoiceEngine::Construct() {
- RTC_DCHECK(worker_thread_checker_.CalledOnValidThread());
- LOG(LS_VERBOSE) << "WebRtcVoiceEngine::WebRtcVoiceEngine";
-
- signal_thread_checker_.DetachFromThread();
- std::memset(&default_agc_config_, 0, sizeof(default_agc_config_));
- voe_config_.Set<webrtc::VoicePacing>(new webrtc::VoicePacing(true));
-
- webrtc::Trace::set_level_filter(kDefaultTraceFilter);
- webrtc::Trace::SetTraceCallback(this);
-
- // Load our audio codec list.
- codecs_ = WebRtcVoiceCodecs::SupportedCodecs();
-}
-
-WebRtcVoiceEngine::~WebRtcVoiceEngine() {
- RTC_DCHECK(worker_thread_checker_.CalledOnValidThread());
- LOG(LS_VERBOSE) << "WebRtcVoiceEngine::~WebRtcVoiceEngine";
- if (adm_) {
- voe_wrapper_.reset();
- adm_->Release();
- adm_ = NULL;
- }
- webrtc::Trace::SetTraceCallback(nullptr);
-}
-
-bool WebRtcVoiceEngine::Init(rtc::Thread* worker_thread) {
- RTC_DCHECK(worker_thread_checker_.CalledOnValidThread());
- RTC_DCHECK(worker_thread == rtc::Thread::Current());
- LOG(LS_INFO) << "WebRtcVoiceEngine::Init";
- bool res = InitInternal();
- if (res) {
- LOG(LS_INFO) << "WebRtcVoiceEngine::Init Done!";
- } else {
- LOG(LS_ERROR) << "WebRtcVoiceEngine::Init failed";
- Terminate();
- }
- return res;
-}
-
-bool WebRtcVoiceEngine::InitInternal() {
- RTC_DCHECK(worker_thread_checker_.CalledOnValidThread());
- // Temporarily turn logging level up for the Init call
- webrtc::Trace::set_level_filter(kElevatedTraceFilter);
- LOG(LS_INFO) << webrtc::VoiceEngine::GetVersionString();
- if (voe_wrapper_->base()->Init(adm_) == -1) {
- LOG_RTCERR0_EX(Init, voe_wrapper_->error());
- return false;
- }
- webrtc::Trace::set_level_filter(kDefaultTraceFilter);
-
- // Save the default AGC configuration settings. This must happen before
- // calling ApplyOptions or the default will be overwritten.
- if (voe_wrapper_->processing()->GetAgcConfig(default_agc_config_) == -1) {
- LOG_RTCERR0(GetAgcConfig);
- return false;
- }
-
- // Set default engine options.
- {
- AudioOptions options;
- options.echo_cancellation = rtc::Optional<bool>(true);
- options.auto_gain_control = rtc::Optional<bool>(true);
- options.noise_suppression = rtc::Optional<bool>(true);
- options.highpass_filter = rtc::Optional<bool>(true);
- options.stereo_swapping = rtc::Optional<bool>(false);
- options.audio_jitter_buffer_max_packets = rtc::Optional<int>(50);
- options.audio_jitter_buffer_fast_accelerate = rtc::Optional<bool>(false);
- options.typing_detection = rtc::Optional<bool>(true);
- options.adjust_agc_delta = rtc::Optional<int>(0);
- options.experimental_agc = rtc::Optional<bool>(false);
- options.extended_filter_aec = rtc::Optional<bool>(false);
- options.delay_agnostic_aec = rtc::Optional<bool>(false);
- options.experimental_ns = rtc::Optional<bool>(false);
- options.aec_dump = rtc::Optional<bool>(false);
- if (!ApplyOptions(options)) {
- return false;
- }
- }
-
- // Print our codec list again for the call diagnostic log
- LOG(LS_INFO) << "WebRtc VoiceEngine codecs:";
- for (const AudioCodec& codec : codecs_) {
- LOG(LS_INFO) << ToString(codec);
- }
-
- SetDefaultDevices();
-
- initialized_ = true;
- return true;
-}
-
-void WebRtcVoiceEngine::Terminate() {
- RTC_DCHECK(worker_thread_checker_.CalledOnValidThread());
- LOG(LS_INFO) << "WebRtcVoiceEngine::Terminate";
- initialized_ = false;
-
- StopAecDump();
-
- voe_wrapper_->base()->Terminate();
-}
-
-rtc::scoped_refptr<webrtc::AudioState>
- WebRtcVoiceEngine::GetAudioState() const {
- RTC_DCHECK(worker_thread_checker_.CalledOnValidThread());
- return audio_state_;
-}
-
-VoiceMediaChannel* WebRtcVoiceEngine::CreateChannel(webrtc::Call* call,
- const AudioOptions& options) {
- RTC_DCHECK(worker_thread_checker_.CalledOnValidThread());
- return new WebRtcVoiceMediaChannel(this, options, call);
-}
-
-bool WebRtcVoiceEngine::ApplyOptions(const AudioOptions& options_in) {
- RTC_DCHECK(worker_thread_checker_.CalledOnValidThread());
- LOG(LS_INFO) << "ApplyOptions: " << options_in.ToString();
- AudioOptions options = options_in; // The options are modified below.
-
- // kEcConference is AEC with high suppression.
- webrtc::EcModes ec_mode = webrtc::kEcConference;
- webrtc::AecmModes aecm_mode = webrtc::kAecmSpeakerphone;
- webrtc::AgcModes agc_mode = webrtc::kAgcAdaptiveAnalog;
- webrtc::NsModes ns_mode = webrtc::kNsHighSuppression;
- if (options.aecm_generate_comfort_noise) {
- LOG(LS_VERBOSE) << "Comfort noise explicitly set to "
- << *options.aecm_generate_comfort_noise
- << " (default is false).";
- }
-
-#if defined(WEBRTC_IOS)
- // On iOS, VPIO provides built-in EC and AGC.
- options.echo_cancellation = rtc::Optional<bool>(false);
- options.auto_gain_control = rtc::Optional<bool>(false);
- LOG(LS_INFO) << "Always disable AEC and AGC on iOS. Use built-in instead.";
-#elif defined(ANDROID)
- ec_mode = webrtc::kEcAecm;
-#endif
-
-#if defined(WEBRTC_IOS) || defined(ANDROID)
- // Set the AGC mode for iOS as well despite disabling it above, to avoid
- // unsupported configuration errors from webrtc.
- agc_mode = webrtc::kAgcFixedDigital;
- options.typing_detection = rtc::Optional<bool>(false);
- options.experimental_agc = rtc::Optional<bool>(false);
- options.extended_filter_aec = rtc::Optional<bool>(false);
- options.experimental_ns = rtc::Optional<bool>(false);
-#endif
-
- // Delay Agnostic AEC automatically turns on EC if not set except on iOS
- // where the feature is not supported.
- bool use_delay_agnostic_aec = false;
-#if !defined(WEBRTC_IOS)
- if (options.delay_agnostic_aec) {
- use_delay_agnostic_aec = *options.delay_agnostic_aec;
- if (use_delay_agnostic_aec) {
- options.echo_cancellation = rtc::Optional<bool>(true);
- options.extended_filter_aec = rtc::Optional<bool>(true);
- ec_mode = webrtc::kEcConference;
- }
- }
-#endif
-
- webrtc::VoEAudioProcessing* voep = voe_wrapper_->processing();
-
- if (options.echo_cancellation) {
- // Check if platform supports built-in EC. Currently only supported on
- // Android and in combination with Java based audio layer.
- // TODO(henrika): investigate possibility to support built-in EC also
- // in combination with Open SL ES audio.
- const bool built_in_aec = voe_wrapper_->hw()->BuiltInAECIsAvailable();
- if (built_in_aec) {
- // Built-in EC exists on this device and use_delay_agnostic_aec is not
- // overriding it. Enable/Disable it according to the echo_cancellation
- // audio option.
- const bool enable_built_in_aec =
- *options.echo_cancellation && !use_delay_agnostic_aec;
- if (voe_wrapper_->hw()->EnableBuiltInAEC(enable_built_in_aec) == 0 &&
- enable_built_in_aec) {
- // Disable internal software EC if built-in EC is enabled,
- // i.e., replace the software EC with the built-in EC.
- options.echo_cancellation = rtc::Optional<bool>(false);
- LOG(LS_INFO) << "Disabling EC since built-in EC will be used instead";
- }
- }
- if (voep->SetEcStatus(*options.echo_cancellation, ec_mode) == -1) {
- LOG_RTCERR2(SetEcStatus, *options.echo_cancellation, ec_mode);
- return false;
- } else {
- LOG(LS_INFO) << "Echo control set to " << *options.echo_cancellation
- << " with mode " << ec_mode;
- }
-#if !defined(ANDROID)
- // TODO(ajm): Remove the error return on Android from webrtc.
- if (voep->SetEcMetricsStatus(*options.echo_cancellation) == -1) {
- LOG_RTCERR1(SetEcMetricsStatus, *options.echo_cancellation);
- return false;
- }
-#endif
- if (ec_mode == webrtc::kEcAecm) {
- bool cn = options.aecm_generate_comfort_noise.value_or(false);
- if (voep->SetAecmMode(aecm_mode, cn) != 0) {
- LOG_RTCERR2(SetAecmMode, aecm_mode, cn);
- return false;
- }
- }
- }
-
- if (options.auto_gain_control) {
- const bool built_in_agc = voe_wrapper_->hw()->BuiltInAGCIsAvailable();
- if (built_in_agc) {
- if (voe_wrapper_->hw()->EnableBuiltInAGC(*options.auto_gain_control) ==
- 0 &&
- *options.auto_gain_control) {
- // Disable internal software AGC if built-in AGC is enabled,
- // i.e., replace the software AGC with the built-in AGC.
- options.auto_gain_control = rtc::Optional<bool>(false);
- LOG(LS_INFO) << "Disabling AGC since built-in AGC will be used instead";
- }
- }
- if (voep->SetAgcStatus(*options.auto_gain_control, agc_mode) == -1) {
- LOG_RTCERR2(SetAgcStatus, *options.auto_gain_control, agc_mode);
- return false;
- } else {
- LOG(LS_INFO) << "Auto gain set to " << *options.auto_gain_control
- << " with mode " << agc_mode;
- }
- }
-
- if (options.tx_agc_target_dbov || options.tx_agc_digital_compression_gain ||
- options.tx_agc_limiter) {
- // Override default_agc_config_. Generally, an unset option means "leave
- // the VoE bits alone" in this function, so we want whatever is set to be
- // stored as the new "default". If we didn't, then setting e.g.
- // tx_agc_target_dbov would reset digital compression gain and limiter
- // settings.
- // Also, if we don't update default_agc_config_, then adjust_agc_delta
- // would be an offset from the original values, and not whatever was set
- // explicitly.
- default_agc_config_.targetLeveldBOv = options.tx_agc_target_dbov.value_or(
- default_agc_config_.targetLeveldBOv);
- default_agc_config_.digitalCompressionGaindB =
- options.tx_agc_digital_compression_gain.value_or(
- default_agc_config_.digitalCompressionGaindB);
- default_agc_config_.limiterEnable =
- options.tx_agc_limiter.value_or(default_agc_config_.limiterEnable);
- if (voe_wrapper_->processing()->SetAgcConfig(default_agc_config_) == -1) {
- LOG_RTCERR3(SetAgcConfig,
- default_agc_config_.targetLeveldBOv,
- default_agc_config_.digitalCompressionGaindB,
- default_agc_config_.limiterEnable);
- return false;
- }
- }
-
- if (options.noise_suppression) {
- const bool built_in_ns = voe_wrapper_->hw()->BuiltInNSIsAvailable();
- if (built_in_ns) {
- if (voe_wrapper_->hw()->EnableBuiltInNS(*options.noise_suppression) ==
- 0 &&
- *options.noise_suppression) {
- // Disable internal software NS if built-in NS is enabled,
- // i.e., replace the software NS with the built-in NS.
- options.noise_suppression = rtc::Optional<bool>(false);
- LOG(LS_INFO) << "Disabling NS since built-in NS will be used instead";
- }
- }
- if (voep->SetNsStatus(*options.noise_suppression, ns_mode) == -1) {
- LOG_RTCERR2(SetNsStatus, *options.noise_suppression, ns_mode);
- return false;
- } else {
- LOG(LS_INFO) << "Noise suppression set to " << *options.noise_suppression
- << " with mode " << ns_mode;
- }
- }
-
- if (options.highpass_filter) {
- LOG(LS_INFO) << "High pass filter enabled? " << *options.highpass_filter;
- if (voep->EnableHighPassFilter(*options.highpass_filter) == -1) {
- LOG_RTCERR1(SetHighpassFilterStatus, *options.highpass_filter);
- return false;
- }
- }
-
- if (options.stereo_swapping) {
- LOG(LS_INFO) << "Stereo swapping enabled? " << *options.stereo_swapping;
- voep->EnableStereoChannelSwapping(*options.stereo_swapping);
- if (voep->IsStereoChannelSwappingEnabled() != *options.stereo_swapping) {
- LOG_RTCERR1(EnableStereoChannelSwapping, *options.stereo_swapping);
- return false;
- }
- }
-
- if (options.audio_jitter_buffer_max_packets) {
- LOG(LS_INFO) << "NetEq capacity is "
- << *options.audio_jitter_buffer_max_packets;
- voe_config_.Set<webrtc::NetEqCapacityConfig>(
- new webrtc::NetEqCapacityConfig(
- *options.audio_jitter_buffer_max_packets));
- }
-
- if (options.audio_jitter_buffer_fast_accelerate) {
- LOG(LS_INFO) << "NetEq fast mode? "
- << *options.audio_jitter_buffer_fast_accelerate;
- voe_config_.Set<webrtc::NetEqFastAccelerate>(
- new webrtc::NetEqFastAccelerate(
- *options.audio_jitter_buffer_fast_accelerate));
- }
-
- if (options.typing_detection) {
- LOG(LS_INFO) << "Typing detection is enabled? "
- << *options.typing_detection;
- if (voep->SetTypingDetectionStatus(*options.typing_detection) == -1) {
- // In case of error, log the info and continue
- LOG_RTCERR1(SetTypingDetectionStatus, *options.typing_detection);
- }
- }
-
- if (options.adjust_agc_delta) {
- LOG(LS_INFO) << "Adjust agc delta is " << *options.adjust_agc_delta;
- if (!AdjustAgcLevel(*options.adjust_agc_delta)) {
- return false;
- }
- }
-
- if (options.aec_dump) {
- LOG(LS_INFO) << "Aec dump is enabled? " << *options.aec_dump;
- if (*options.aec_dump)
- StartAecDump(kAecDumpByAudioOptionFilename);
- else
- StopAecDump();
- }
-
- webrtc::Config config;
-
- if (options.delay_agnostic_aec)
- delay_agnostic_aec_ = options.delay_agnostic_aec;
- if (delay_agnostic_aec_) {
- LOG(LS_INFO) << "Delay agnostic aec is enabled? " << *delay_agnostic_aec_;
- config.Set<webrtc::DelayAgnostic>(
- new webrtc::DelayAgnostic(*delay_agnostic_aec_));
- }
-
- if (options.extended_filter_aec) {
- extended_filter_aec_ = options.extended_filter_aec;
- }
- if (extended_filter_aec_) {
- LOG(LS_INFO) << "Extended filter aec is enabled? " << *extended_filter_aec_;
- config.Set<webrtc::ExtendedFilter>(
- new webrtc::ExtendedFilter(*extended_filter_aec_));
- }
-
- if (options.experimental_ns) {
- experimental_ns_ = options.experimental_ns;
- }
- if (experimental_ns_) {
- LOG(LS_INFO) << "Experimental ns is enabled? " << *experimental_ns_;
- config.Set<webrtc::ExperimentalNs>(
- new webrtc::ExperimentalNs(*experimental_ns_));
- }
-
- // We check audioproc for the benefit of tests, since FakeWebRtcVoiceEngine
- // returns NULL on audio_processing().
- webrtc::AudioProcessing* audioproc = voe_wrapper_->base()->audio_processing();
- if (audioproc) {
- audioproc->SetExtraOptions(config);
- }
-
- if (options.recording_sample_rate) {
- LOG(LS_INFO) << "Recording sample rate is "
- << *options.recording_sample_rate;
- if (voe_wrapper_->hw()->SetRecordingSampleRate(
- *options.recording_sample_rate)) {
- LOG_RTCERR1(SetRecordingSampleRate, *options.recording_sample_rate);
- }
- }
-
- if (options.playout_sample_rate) {
- LOG(LS_INFO) << "Playout sample rate is " << *options.playout_sample_rate;
- if (voe_wrapper_->hw()->SetPlayoutSampleRate(
- *options.playout_sample_rate)) {
- LOG_RTCERR1(SetPlayoutSampleRate, *options.playout_sample_rate);
- }
- }
-
- return true;
-}
-
-void WebRtcVoiceEngine::SetDefaultDevices() {
- RTC_DCHECK(worker_thread_checker_.CalledOnValidThread());
-#if !defined(WEBRTC_IOS)
- int in_id = kDefaultAudioDeviceId;
- int out_id = kDefaultAudioDeviceId;
- LOG(LS_INFO) << "Setting microphone to (id=" << in_id
- << ") and speaker to (id=" << out_id << ")";
-
- bool ret = true;
- if (voe_wrapper_->hw()->SetRecordingDevice(in_id) == -1) {
- LOG_RTCERR1(SetRecordingDevice, in_id);
- ret = false;
- }
- webrtc::AudioProcessing* ap = voe()->base()->audio_processing();
- if (ap) {
- ap->Initialize();
- }
-
- if (voe_wrapper_->hw()->SetPlayoutDevice(out_id) == -1) {
- LOG_RTCERR1(SetPlayoutDevice, out_id);
- ret = false;
- }
-
- if (ret) {
- LOG(LS_INFO) << "Set microphone to (id=" << in_id
- << ") and speaker to (id=" << out_id << ")";
- }
-#endif // !WEBRTC_IOS
-}
-
-bool WebRtcVoiceEngine::GetOutputVolume(int* level) {
- RTC_DCHECK(worker_thread_checker_.CalledOnValidThread());
- unsigned int ulevel;
- if (voe_wrapper_->volume()->GetSpeakerVolume(ulevel) == -1) {
- LOG_RTCERR1(GetSpeakerVolume, level);
- return false;
- }
- *level = ulevel;
- return true;
-}
-
-bool WebRtcVoiceEngine::SetOutputVolume(int level) {
- RTC_DCHECK(worker_thread_checker_.CalledOnValidThread());
- RTC_DCHECK(level >= 0 && level <= 255);
- if (voe_wrapper_->volume()->SetSpeakerVolume(level) == -1) {
- LOG_RTCERR1(SetSpeakerVolume, level);
- return false;
- }
- return true;
-}
-
-int WebRtcVoiceEngine::GetInputLevel() {
- RTC_DCHECK(worker_thread_checker_.CalledOnValidThread());
- unsigned int ulevel;
- return (voe_wrapper_->volume()->GetSpeechInputLevel(ulevel) != -1) ?
- static_cast<int>(ulevel) : -1;
-}
-
-const std::vector<AudioCodec>& WebRtcVoiceEngine::codecs() {
- RTC_DCHECK(signal_thread_checker_.CalledOnValidThread());
- return codecs_;
-}
-
-RtpCapabilities WebRtcVoiceEngine::GetCapabilities() const {
- RTC_DCHECK(signal_thread_checker_.CalledOnValidThread());
- RtpCapabilities capabilities;
- capabilities.header_extensions.push_back(RtpHeaderExtension(
- kRtpAudioLevelHeaderExtension, kRtpAudioLevelHeaderExtensionDefaultId));
- capabilities.header_extensions.push_back(
- RtpHeaderExtension(kRtpAbsoluteSenderTimeHeaderExtension,
- kRtpAbsoluteSenderTimeHeaderExtensionDefaultId));
- if (webrtc::field_trial::FindFullName("WebRTC-Audio-SendSideBwe") ==
- "Enabled") {
- capabilities.header_extensions.push_back(RtpHeaderExtension(
- kRtpTransportSequenceNumberHeaderExtension,
- kRtpTransportSequenceNumberHeaderExtensionDefaultId));
- }
- return capabilities;
-}
-
-int WebRtcVoiceEngine::GetLastEngineError() {
- RTC_DCHECK(worker_thread_checker_.CalledOnValidThread());
- return voe_wrapper_->error();
-}
-
-void WebRtcVoiceEngine::Print(webrtc::TraceLevel level, const char* trace,
- int length) {
- // Note: This callback can happen on any thread!
- rtc::LoggingSeverity sev = rtc::LS_VERBOSE;
- if (level == webrtc::kTraceError || level == webrtc::kTraceCritical)
- sev = rtc::LS_ERROR;
- else if (level == webrtc::kTraceWarning)
- sev = rtc::LS_WARNING;
- else if (level == webrtc::kTraceStateInfo || level == webrtc::kTraceInfo)
- sev = rtc::LS_INFO;
- else if (level == webrtc::kTraceTerseInfo)
- sev = rtc::LS_INFO;
-
- // Skip past boilerplate prefix text
- if (length < 72) {
- std::string msg(trace, length);
- LOG(LS_ERROR) << "Malformed webrtc log message: ";
- LOG_V(sev) << msg;
- } else {
- std::string msg(trace + 71, length - 72);
- LOG_V(sev) << "webrtc: " << msg;
- }
-}
-
-void WebRtcVoiceEngine::RegisterChannel(WebRtcVoiceMediaChannel* channel) {
- RTC_DCHECK(worker_thread_checker_.CalledOnValidThread());
- RTC_DCHECK(channel);
- channels_.push_back(channel);
-}
-
-void WebRtcVoiceEngine::UnregisterChannel(WebRtcVoiceMediaChannel* channel) {
- RTC_DCHECK(worker_thread_checker_.CalledOnValidThread());
- auto it = std::find(channels_.begin(), channels_.end(), channel);
- RTC_DCHECK(it != channels_.end());
- channels_.erase(it);
-}
-
-// Adjusts the default AGC target level by the specified delta.
-// NB: If we start messing with other config fields, we'll want
-// to save the current webrtc::AgcConfig as well.
-bool WebRtcVoiceEngine::AdjustAgcLevel(int delta) {
- RTC_DCHECK(worker_thread_checker_.CalledOnValidThread());
- webrtc::AgcConfig config = default_agc_config_;
- config.targetLeveldBOv -= delta;
-
- LOG(LS_INFO) << "Adjusting AGC level from default -"
- << default_agc_config_.targetLeveldBOv << "dB to -"
- << config.targetLeveldBOv << "dB";
-
- if (voe_wrapper_->processing()->SetAgcConfig(config) == -1) {
- LOG_RTCERR1(SetAgcConfig, config.targetLeveldBOv);
- return false;
- }
- return true;
-}
-
-bool WebRtcVoiceEngine::SetAudioDeviceModule(webrtc::AudioDeviceModule* adm) {
- RTC_DCHECK(worker_thread_checker_.CalledOnValidThread());
- if (initialized_) {
- LOG(LS_WARNING) << "SetAudioDeviceModule can not be called after Init.";
- return false;
- }
- if (adm_) {
- adm_->Release();
- adm_ = NULL;
- }
- if (adm) {
- adm_ = adm;
- adm_->AddRef();
- }
- return true;
-}
-
-bool WebRtcVoiceEngine::StartAecDump(rtc::PlatformFile file,
- int64_t max_size_bytes) {
- RTC_DCHECK(worker_thread_checker_.CalledOnValidThread());
- FILE* aec_dump_file_stream = rtc::FdopenPlatformFileForWriting(file);
- if (!aec_dump_file_stream) {
- LOG(LS_ERROR) << "Could not open AEC dump file stream.";
- if (!rtc::ClosePlatformFile(file))
- LOG(LS_WARNING) << "Could not close file.";
- return false;
- }
- StopAecDump();
- if (voe_wrapper_->base()->audio_processing()->StartDebugRecording(
- aec_dump_file_stream, max_size_bytes) !=
- webrtc::AudioProcessing::kNoError) {
- LOG_RTCERR0(StartDebugRecording);
- fclose(aec_dump_file_stream);
- return false;
- }
- is_dumping_aec_ = true;
- return true;
-}
-
-void WebRtcVoiceEngine::StartAecDump(const std::string& filename) {
- RTC_DCHECK(worker_thread_checker_.CalledOnValidThread());
- if (!is_dumping_aec_) {
- // Start dumping AEC when we are not dumping.
- if (voe_wrapper_->base()->audio_processing()->StartDebugRecording(
- filename.c_str(), -1) != webrtc::AudioProcessing::kNoError) {
- LOG_RTCERR1(StartDebugRecording, filename.c_str());
- } else {
- is_dumping_aec_ = true;
- }
- }
-}
-
-void WebRtcVoiceEngine::StopAecDump() {
- RTC_DCHECK(worker_thread_checker_.CalledOnValidThread());
- if (is_dumping_aec_) {
- // Stop dumping AEC when we are dumping.
- if (voe_wrapper_->base()->audio_processing()->StopDebugRecording() !=
- webrtc::AudioProcessing::kNoError) {
- LOG_RTCERR0(StopDebugRecording);
- }
- is_dumping_aec_ = false;
- }
-}
-
-bool WebRtcVoiceEngine::StartRtcEventLog(rtc::PlatformFile file) {
- RTC_DCHECK(worker_thread_checker_.CalledOnValidThread());
- webrtc::RtcEventLog* event_log = voe_wrapper_->codec()->GetEventLog();
- if (event_log) {
- return event_log->StartLogging(file);
- }
- LOG_RTCERR0(StartRtcEventLog);
- return false;
-}
-
-void WebRtcVoiceEngine::StopRtcEventLog() {
- RTC_DCHECK(worker_thread_checker_.CalledOnValidThread());
- webrtc::RtcEventLog* event_log = voe_wrapper_->codec()->GetEventLog();
- if (event_log) {
- event_log->StopLogging();
- return;
- }
- LOG_RTCERR0(StopRtcEventLog);
-}
-
-int WebRtcVoiceEngine::CreateVoEChannel() {
- RTC_DCHECK(worker_thread_checker_.CalledOnValidThread());
- return voe_wrapper_->base()->CreateChannel(voe_config_);
-}
-
-class WebRtcVoiceMediaChannel::WebRtcAudioSendStream
- : public AudioRenderer::Sink {
- public:
- WebRtcAudioSendStream(int ch, webrtc::AudioTransport* voe_audio_transport,
- uint32_t ssrc, const std::string& c_name,
- const std::vector<webrtc::RtpExtension>& extensions,
- webrtc::Call* call)
- : voe_audio_transport_(voe_audio_transport),
- call_(call),
- config_(nullptr) {
- RTC_DCHECK_GE(ch, 0);
- // TODO(solenberg): Once we're not using FakeWebRtcVoiceEngine anymore:
- // RTC_DCHECK(voe_audio_transport);
- RTC_DCHECK(call);
- audio_capture_thread_checker_.DetachFromThread();
- config_.rtp.ssrc = ssrc;
- config_.rtp.c_name = c_name;
- config_.voe_channel_id = ch;
- RecreateAudioSendStream(extensions);
- }
-
- ~WebRtcAudioSendStream() override {
- RTC_DCHECK(worker_thread_checker_.CalledOnValidThread());
- Stop();
- call_->DestroyAudioSendStream(stream_);
- }
-
- void RecreateAudioSendStream(
- const std::vector<webrtc::RtpExtension>& extensions) {
- RTC_DCHECK(worker_thread_checker_.CalledOnValidThread());
- if (stream_) {
- call_->DestroyAudioSendStream(stream_);
- stream_ = nullptr;
- }
- config_.rtp.extensions = extensions;
- RTC_DCHECK(!stream_);
- stream_ = call_->CreateAudioSendStream(config_);
- RTC_CHECK(stream_);
- }
-
- bool SendTelephoneEvent(int payload_type, uint8_t event,
- uint32_t duration_ms) {
- RTC_DCHECK(worker_thread_checker_.CalledOnValidThread());
- RTC_DCHECK(stream_);
- return stream_->SendTelephoneEvent(payload_type, event, duration_ms);
- }
-
- webrtc::AudioSendStream::Stats GetStats() const {
- RTC_DCHECK(worker_thread_checker_.CalledOnValidThread());
- RTC_DCHECK(stream_);
- return stream_->GetStats();
- }
-
- // Starts the rendering by setting a sink to the renderer to get data
- // callback.
- // This method is called on the libjingle worker thread.
- // TODO(xians): Make sure Start() is called only once.
- void Start(AudioRenderer* renderer) {
- RTC_DCHECK(worker_thread_checker_.CalledOnValidThread());
- RTC_DCHECK(renderer);
- if (renderer_) {
- RTC_DCHECK(renderer_ == renderer);
- return;
- }
- renderer->SetSink(this);
- renderer_ = renderer;
- }
-
- // Stops rendering by setting the sink of the renderer to nullptr. No data
- // callback will be received after this method.
- // This method is called on the libjingle worker thread.
- void Stop() {
- RTC_DCHECK(worker_thread_checker_.CalledOnValidThread());
- if (renderer_) {
- renderer_->SetSink(nullptr);
- renderer_ = nullptr;
- }
- }
-
- // AudioRenderer::Sink implementation.
- // This method is called on the audio thread.
- void OnData(const void* audio_data,
- int bits_per_sample,
- int sample_rate,
- size_t number_of_channels,
- size_t number_of_frames) override {
- RTC_DCHECK(!worker_thread_checker_.CalledOnValidThread());
- RTC_DCHECK(audio_capture_thread_checker_.CalledOnValidThread());
- RTC_DCHECK(voe_audio_transport_);
- voe_audio_transport_->OnData(config_.voe_channel_id,
- audio_data,
- bits_per_sample,
- sample_rate,
- number_of_channels,
- number_of_frames);
- }
-
- // Callback from the |renderer_| when it is going away. In case Start() has
- // never been called, this callback won't be triggered.
- void OnClose() override {
- RTC_DCHECK(worker_thread_checker_.CalledOnValidThread());
- // Set |renderer_| to nullptr to make sure no more callback will get into
- // the renderer.
- renderer_ = nullptr;
- }
-
- // Accessor to the VoE channel ID.
- int channel() const {
- RTC_DCHECK(worker_thread_checker_.CalledOnValidThread());
- return config_.voe_channel_id;
- }
-
- private:
- rtc::ThreadChecker worker_thread_checker_;
- rtc::ThreadChecker audio_capture_thread_checker_;
- webrtc::AudioTransport* const voe_audio_transport_ = nullptr;
- webrtc::Call* call_ = nullptr;
- webrtc::AudioSendStream::Config config_;
- // The stream is owned by WebRtcAudioSendStream and may be reallocated if
- // configuration changes.
- webrtc::AudioSendStream* stream_ = nullptr;
-
- // Raw pointer to AudioRenderer owned by LocalAudioTrackHandler.
- // PeerConnection will make sure invalidating the pointer before the object
- // goes away.
- AudioRenderer* renderer_ = nullptr;
-
- RTC_DISALLOW_IMPLICIT_CONSTRUCTORS(WebRtcAudioSendStream);
-};
-
-class WebRtcVoiceMediaChannel::WebRtcAudioReceiveStream {
- public:
- WebRtcAudioReceiveStream(int ch,
- uint32_t remote_ssrc,
- uint32_t local_ssrc,
- bool use_transport_cc,
- const std::string& sync_group,
- const std::vector<webrtc::RtpExtension>& extensions,
- webrtc::Call* call)
- : call_(call), config_() {
- RTC_DCHECK_GE(ch, 0);
- RTC_DCHECK(call);
- config_.rtp.remote_ssrc = remote_ssrc;
- config_.rtp.local_ssrc = local_ssrc;
- config_.voe_channel_id = ch;
- config_.sync_group = sync_group;
- RecreateAudioReceiveStream(use_transport_cc, extensions);
- }
-
- ~WebRtcAudioReceiveStream() {
- RTC_DCHECK(worker_thread_checker_.CalledOnValidThread());
- call_->DestroyAudioReceiveStream(stream_);
- }
-
- void RecreateAudioReceiveStream(
- const std::vector<webrtc::RtpExtension>& extensions) {
- RTC_DCHECK(worker_thread_checker_.CalledOnValidThread());
- RecreateAudioReceiveStream(config_.rtp.transport_cc, extensions);
- }
- void RecreateAudioReceiveStream(bool use_transport_cc) {
- RTC_DCHECK(worker_thread_checker_.CalledOnValidThread());
- RecreateAudioReceiveStream(use_transport_cc, config_.rtp.extensions);
- }
-
- webrtc::AudioReceiveStream::Stats GetStats() const {
- RTC_DCHECK(worker_thread_checker_.CalledOnValidThread());
- RTC_DCHECK(stream_);
- return stream_->GetStats();
- }
-
- int channel() const {
- RTC_DCHECK(worker_thread_checker_.CalledOnValidThread());
- return config_.voe_channel_id;
- }
-
- void SetRawAudioSink(rtc::scoped_ptr<webrtc::AudioSinkInterface> sink) {
- RTC_DCHECK(worker_thread_checker_.CalledOnValidThread());
- stream_->SetSink(std::move(sink));
- }
-
- private:
- void RecreateAudioReceiveStream(
- bool use_transport_cc,
- const std::vector<webrtc::RtpExtension>& extensions) {
- RTC_DCHECK(worker_thread_checker_.CalledOnValidThread());
- if (stream_) {
- call_->DestroyAudioReceiveStream(stream_);
- stream_ = nullptr;
- }
- config_.rtp.extensions = extensions;
- config_.rtp.transport_cc = use_transport_cc;
- RTC_DCHECK(!stream_);
- stream_ = call_->CreateAudioReceiveStream(config_);
- RTC_CHECK(stream_);
- }
-
- rtc::ThreadChecker worker_thread_checker_;
- webrtc::Call* call_ = nullptr;
- webrtc::AudioReceiveStream::Config config_;
- // The stream is owned by WebRtcAudioReceiveStream and may be reallocated if
- // configuration changes.
- webrtc::AudioReceiveStream* stream_ = nullptr;
-
- RTC_DISALLOW_IMPLICIT_CONSTRUCTORS(WebRtcAudioReceiveStream);
-};
-
-WebRtcVoiceMediaChannel::WebRtcVoiceMediaChannel(WebRtcVoiceEngine* engine,
- const AudioOptions& options,
- webrtc::Call* call)
- : engine_(engine), call_(call) {
- LOG(LS_VERBOSE) << "WebRtcVoiceMediaChannel::WebRtcVoiceMediaChannel";
- RTC_DCHECK(call);
- engine->RegisterChannel(this);
- SetOptions(options);
-}
-
-WebRtcVoiceMediaChannel::~WebRtcVoiceMediaChannel() {
- RTC_DCHECK(worker_thread_checker_.CalledOnValidThread());
- LOG(LS_VERBOSE) << "WebRtcVoiceMediaChannel::~WebRtcVoiceMediaChannel";
- // TODO(solenberg): Should be able to delete the streams directly, without
- // going through RemoveNnStream(), once stream objects handle
- // all (de)configuration.
- while (!send_streams_.empty()) {
- RemoveSendStream(send_streams_.begin()->first);
- }
- while (!recv_streams_.empty()) {
- RemoveRecvStream(recv_streams_.begin()->first);
- }
- engine()->UnregisterChannel(this);
-}
-
-bool WebRtcVoiceMediaChannel::SetSendParameters(
- const AudioSendParameters& params) {
- RTC_DCHECK(worker_thread_checker_.CalledOnValidThread());
- LOG(LS_INFO) << "WebRtcVoiceMediaChannel::SetSendParameters: "
- << params.ToString();
- // TODO(pthatcher): Refactor this to be more clean now that we have
- // all the information at once.
-
- if (!SetSendCodecs(params.codecs)) {
- return false;
- }
-
- if (!ValidateRtpExtensions(params.extensions)) {
- return false;
- }
- std::vector<webrtc::RtpExtension> filtered_extensions =
- FilterRtpExtensions(params.extensions,
- webrtc::RtpExtension::IsSupportedForAudio, true);
- if (send_rtp_extensions_ != filtered_extensions) {
- send_rtp_extensions_.swap(filtered_extensions);
- for (auto& it : send_streams_) {
- it.second->RecreateAudioSendStream(send_rtp_extensions_);
- }
- }
-
- if (!SetMaxSendBandwidth(params.max_bandwidth_bps)) {
- return false;
- }
- return SetOptions(params.options);
-}
-
-bool WebRtcVoiceMediaChannel::SetRecvParameters(
- const AudioRecvParameters& params) {
- RTC_DCHECK(worker_thread_checker_.CalledOnValidThread());
- LOG(LS_INFO) << "WebRtcVoiceMediaChannel::SetRecvParameters: "
- << params.ToString();
- // TODO(pthatcher): Refactor this to be more clean now that we have
- // all the information at once.
-
- if (!SetRecvCodecs(params.codecs)) {
- return false;
- }
-
- if (!ValidateRtpExtensions(params.extensions)) {
- return false;
- }
- std::vector<webrtc::RtpExtension> filtered_extensions =
- FilterRtpExtensions(params.extensions,
- webrtc::RtpExtension::IsSupportedForAudio, false);
- if (recv_rtp_extensions_ != filtered_extensions) {
- recv_rtp_extensions_.swap(filtered_extensions);
- for (auto& it : recv_streams_) {
- it.second->RecreateAudioReceiveStream(recv_rtp_extensions_);
- }
- }
- return true;
-}
-
-bool WebRtcVoiceMediaChannel::SetOptions(const AudioOptions& options) {
- RTC_DCHECK(worker_thread_checker_.CalledOnValidThread());
- LOG(LS_INFO) << "Setting voice channel options: "
- << options.ToString();
-
- // Check if DSCP value is changed from previous.
- bool dscp_option_changed = (options_.dscp != options.dscp);
-
- // We retain all of the existing options, and apply the given ones
- // on top. This means there is no way to "clear" options such that
- // they go back to the engine default.
- options_.SetAll(options);
- if (!engine()->ApplyOptions(options_)) {
- LOG(LS_WARNING) <<
- "Failed to apply engine options during channel SetOptions.";
- return false;
- }
-
- if (dscp_option_changed) {
- rtc::DiffServCodePoint dscp = rtc::DSCP_DEFAULT;
- if (options_.dscp.value_or(false)) {
- dscp = kAudioDscpValue;
- }
- if (MediaChannel::SetDscp(dscp) != 0) {
- LOG(LS_WARNING) << "Failed to set DSCP settings for audio channel";
- }
- }
-
- LOG(LS_INFO) << "Set voice channel options. Current options: "
- << options_.ToString();
- return true;
-}
-
-bool WebRtcVoiceMediaChannel::SetRecvCodecs(
- const std::vector<AudioCodec>& codecs) {
- RTC_DCHECK(worker_thread_checker_.CalledOnValidThread());
-
- // Set the payload types to be used for incoming media.
- LOG(LS_INFO) << "Setting receive voice codecs.";
-
- if (!VerifyUniquePayloadTypes(codecs)) {
- LOG(LS_ERROR) << "Codec payload types overlap.";
- return false;
- }
-
- std::vector<AudioCodec> new_codecs;
- // Find all new codecs. We allow adding new codecs but don't allow changing
- // the payload type of codecs that is already configured since we might
- // already be receiving packets with that payload type.
- for (const AudioCodec& codec : codecs) {
- AudioCodec old_codec;
- if (FindCodec(recv_codecs_, codec, &old_codec)) {
- if (old_codec.id != codec.id) {
- LOG(LS_ERROR) << codec.name << " payload type changed.";
- return false;
- }
- } else {
- new_codecs.push_back(codec);
- }
- }
- if (new_codecs.empty()) {
- // There are no new codecs to configure. Already configured codecs are
- // never removed.
- return true;
- }
-
- if (playout_) {
- // Receive codecs can not be changed while playing. So we temporarily
- // pause playout.
- PausePlayout();
- }
-
- bool result = true;
- for (const AudioCodec& codec : new_codecs) {
- webrtc::CodecInst voe_codec;
- if (WebRtcVoiceEngine::ToCodecInst(codec, &voe_codec)) {
- LOG(LS_INFO) << ToString(codec);
- voe_codec.pltype = codec.id;
- for (const auto& ch : recv_streams_) {
- if (engine()->voe()->codec()->SetRecPayloadType(
- ch.second->channel(), voe_codec) == -1) {
- LOG_RTCERR2(SetRecPayloadType, ch.second->channel(),
- ToString(voe_codec));
- result = false;
- }
- }
- } else {
- LOG(LS_WARNING) << "Unknown codec " << ToString(codec);
- result = false;
- break;
- }
- }
- if (result) {
- recv_codecs_ = codecs;
- }
-
- if (desired_playout_ && !playout_) {
- ResumePlayout();
- }
- return result;
-}
-
-bool WebRtcVoiceMediaChannel::SetSendCodecs(
- int channel, const std::vector<AudioCodec>& codecs) {
- // Disable VAD, FEC, and RED unless we know the other side wants them.
- engine()->voe()->codec()->SetVADStatus(channel, false);
- engine()->voe()->rtp()->SetNACKStatus(channel, false, 0);
- engine()->voe()->rtp()->SetREDStatus(channel, false);
- engine()->voe()->codec()->SetFECStatus(channel, false);
-
- // Scan through the list to figure out the codec to use for sending, along
- // with the proper configuration for VAD.
- webrtc::CodecInst send_codec;
- memset(&send_codec, 0, sizeof(send_codec));
-
- bool nack_enabled = nack_enabled_;
- bool enable_codec_fec = false;
- bool enable_opus_dtx = false;
- int opus_max_playback_rate = 0;
- int red_payload_type = -1;
-
- // Set send codec (the first non-telephone-event/CN codec)
- const AudioCodec* codec = WebRtcVoiceCodecs::GetPreferredCodec(
- codecs, &send_codec, &red_payload_type);
- if (codec) {
- if (red_payload_type != -1) {
- // Enable redundant encoding of the specified codec. Treat any
- // failure as a fatal internal error.
- LOG(LS_INFO) << "Enabling RED on channel " << channel;
- if (engine()->voe()->rtp()->SetREDStatus(channel, true,
- red_payload_type) == -1) {
- LOG_RTCERR3(SetREDStatus, channel, true, red_payload_type);
- return false;
- }
- } else {
- nack_enabled = HasNack(*codec);
- // For Opus as the send codec, we are to determine inband FEC, maximum
- // playback rate, and opus internal dtx.
- if (IsCodec(*codec, kOpusCodecName)) {
- GetOpusConfig(*codec, &send_codec, &enable_codec_fec,
- &opus_max_playback_rate, &enable_opus_dtx);
- }
-
- // Set packet size if the AudioCodec param kCodecParamPTime is set.
- int ptime_ms = 0;
- if (codec->GetParam(kCodecParamPTime, &ptime_ms)) {
- if (!WebRtcVoiceCodecs::SetPTimeAsPacketSize(&send_codec, ptime_ms)) {
- LOG(LS_WARNING) << "Failed to set packet size for codec "
- << send_codec.plname;
- return false;
- }
- }
- }
- }
-
- if (nack_enabled_ != nack_enabled) {
- SetNack(channel, nack_enabled);
- nack_enabled_ = nack_enabled;
- }
- if (!codec) {
- LOG(LS_WARNING) << "Received empty list of codecs.";
- return false;
- }
-
- // Set the codec immediately, since SetVADStatus() depends on whether
- // the current codec is mono or stereo.
- if (!SetSendCodec(channel, send_codec))
- return false;
-
- // FEC should be enabled after SetSendCodec.
- if (enable_codec_fec) {
- LOG(LS_INFO) << "Attempt to enable codec internal FEC on channel "
- << channel;
- if (engine()->voe()->codec()->SetFECStatus(channel, true) == -1) {
- // Enable codec internal FEC. Treat any failure as fatal internal error.
- LOG_RTCERR2(SetFECStatus, channel, true);
- return false;
- }
- }
-
- if (IsCodec(send_codec, kOpusCodecName)) {
- // DTX and maxplaybackrate should be set after SetSendCodec. Because current
- // send codec has to be Opus.
-
- // Set Opus internal DTX.
- LOG(LS_INFO) << "Attempt to "
- << (enable_opus_dtx ? "enable" : "disable")
- << " Opus DTX on channel "
- << channel;
- if (engine()->voe()->codec()->SetOpusDtx(channel, enable_opus_dtx)) {
- LOG_RTCERR2(SetOpusDtx, channel, enable_opus_dtx);
- return false;
- }
-
- // If opus_max_playback_rate <= 0, the default maximum playback rate
- // (48 kHz) will be used.
- if (opus_max_playback_rate > 0) {
- LOG(LS_INFO) << "Attempt to set maximum playback rate to "
- << opus_max_playback_rate
- << " Hz on channel "
- << channel;
- if (engine()->voe()->codec()->SetOpusMaxPlaybackRate(
- channel, opus_max_playback_rate) == -1) {
- LOG_RTCERR2(SetOpusMaxPlaybackRate, channel, opus_max_playback_rate);
- return false;
- }
- }
- }
-
- // Always update the |send_codec_| to the currently set send codec.
- send_codec_.reset(new webrtc::CodecInst(send_codec));
-
- if (send_bitrate_setting_) {
- SetSendBitrateInternal(send_bitrate_bps_);
- }
-
- // Loop through the codecs list again to config the CN codec.
- for (const AudioCodec& codec : codecs) {
- // Ignore codecs we don't know about. The negotiation step should prevent
- // this, but double-check to be sure.
- webrtc::CodecInst voe_codec;
- if (!WebRtcVoiceEngine::ToCodecInst(codec, &voe_codec)) {
- LOG(LS_WARNING) << "Unknown codec " << ToString(codec);
- continue;
- }
-
- if (IsCodec(codec, kCnCodecName)) {
- // Turn voice activity detection/comfort noise on if supported.
- // Set the wideband CN payload type appropriately.
- // (narrowband always uses the static payload type 13).
- webrtc::PayloadFrequencies cn_freq;
- switch (codec.clockrate) {
- case 8000:
- cn_freq = webrtc::kFreq8000Hz;
- break;
- case 16000:
- cn_freq = webrtc::kFreq16000Hz;
- break;
- case 32000:
- cn_freq = webrtc::kFreq32000Hz;
- break;
- default:
- LOG(LS_WARNING) << "CN frequency " << codec.clockrate
- << " not supported.";
- continue;
- }
- // Set the CN payloadtype and the VAD status.
- // The CN payload type for 8000 Hz clockrate is fixed at 13.
- if (cn_freq != webrtc::kFreq8000Hz) {
- if (engine()->voe()->codec()->SetSendCNPayloadType(
- channel, codec.id, cn_freq) == -1) {
- LOG_RTCERR3(SetSendCNPayloadType, channel, codec.id, cn_freq);
- // TODO(ajm): This failure condition will be removed from VoE.
- // Restore the return here when we update to a new enough webrtc.
- //
- // Not returning false because the SetSendCNPayloadType will fail if
- // the channel is already sending.
- // This can happen if the remote description is applied twice, for
- // example in the case of ROAP on top of JSEP, where both side will
- // send the offer.
- }
- }
- // Only turn on VAD if we have a CN payload type that matches the
- // clockrate for the codec we are going to use.
- if (codec.clockrate == send_codec.plfreq && send_codec.channels != 2) {
- // TODO(minyue): If CN frequency == 48000 Hz is allowed, consider the
- // interaction between VAD and Opus FEC.
- LOG(LS_INFO) << "Enabling VAD";
- if (engine()->voe()->codec()->SetVADStatus(channel, true) == -1) {
- LOG_RTCERR2(SetVADStatus, channel, true);
- return false;
- }
- }
- }
- }
- return true;
-}
-
-bool WebRtcVoiceMediaChannel::SetSendCodecs(
- const std::vector<AudioCodec>& codecs) {
- RTC_DCHECK(worker_thread_checker_.CalledOnValidThread());
- // TODO(solenberg): Validate input - that payload types don't overlap, are
- // within range, filter out codecs we don't support,
- // redundant codecs etc.
-
- // Find the DTMF telephone event "codec" payload type.
- dtmf_payload_type_ = rtc::Optional<int>();
- for (const AudioCodec& codec : codecs) {
- if (IsCodec(codec, kDtmfCodecName)) {
- dtmf_payload_type_ = rtc::Optional<int>(codec.id);
- break;
- }
- }
-
- // Cache the codecs in order to configure the channel created later.
- send_codecs_ = codecs;
- for (const auto& ch : send_streams_) {
- if (!SetSendCodecs(ch.second->channel(), codecs)) {
- return false;
- }
- }
-
- // Set nack status on receive channels and update |nack_enabled_|.
- for (const auto& ch : recv_streams_) {
- SetNack(ch.second->channel(), nack_enabled_);
- }
-
- // Check if the transport cc feedback has changed on the preferred send codec,
- // and in that case reconfigure all receive streams.
- webrtc::CodecInst voe_codec;
- int red_payload_type;
- const AudioCodec* send_codec = WebRtcVoiceCodecs::GetPreferredCodec(
- send_codecs_, &voe_codec, &red_payload_type);
- if (send_codec) {
- bool transport_cc = HasTransportCc(*send_codec);
- if (transport_cc_enabled_ != transport_cc) {
- LOG(LS_INFO) << "Recreate all the receive streams because the send "
- "codec has changed.";
- transport_cc_enabled_ = transport_cc;
- for (auto& kv : recv_streams_) {
- RTC_DCHECK(kv.second != nullptr);
- kv.second->RecreateAudioReceiveStream(transport_cc_enabled_);
- }
- }
- }
-
- return true;
-}
-
-void WebRtcVoiceMediaChannel::SetNack(int channel, bool nack_enabled) {
- if (nack_enabled) {
- LOG(LS_INFO) << "Enabling NACK for channel " << channel;
- engine()->voe()->rtp()->SetNACKStatus(channel, true, kNackMaxPackets);
- } else {
- LOG(LS_INFO) << "Disabling NACK for channel " << channel;
- engine()->voe()->rtp()->SetNACKStatus(channel, false, 0);
- }
-}
-
-bool WebRtcVoiceMediaChannel::SetSendCodec(
- int channel, const webrtc::CodecInst& send_codec) {
- LOG(LS_INFO) << "Send channel " << channel << " selected voice codec "
- << ToString(send_codec) << ", bitrate=" << send_codec.rate;
-
- webrtc::CodecInst current_codec;
- if (engine()->voe()->codec()->GetSendCodec(channel, current_codec) == 0 &&
- (send_codec == current_codec)) {
- // Codec is already configured, we can return without setting it again.
- return true;
- }
-
- if (engine()->voe()->codec()->SetSendCodec(channel, send_codec) == -1) {
- LOG_RTCERR2(SetSendCodec, channel, ToString(send_codec));
- return false;
- }
- return true;
-}
-
-bool WebRtcVoiceMediaChannel::SetPlayout(bool playout) {
- desired_playout_ = playout;
- return ChangePlayout(desired_playout_);
-}
-
-bool WebRtcVoiceMediaChannel::PausePlayout() {
- return ChangePlayout(false);
-}
-
-bool WebRtcVoiceMediaChannel::ResumePlayout() {
- return ChangePlayout(desired_playout_);
-}
-
-bool WebRtcVoiceMediaChannel::ChangePlayout(bool playout) {
- RTC_DCHECK(worker_thread_checker_.CalledOnValidThread());
- if (playout_ == playout) {
- return true;
- }
-
- for (const auto& ch : recv_streams_) {
- if (!SetPlayout(ch.second->channel(), playout)) {
- LOG(LS_ERROR) << "SetPlayout " << playout << " on channel "
- << ch.second->channel() << " failed";
- return false;
- }
- }
- playout_ = playout;
- return true;
-}
-
-bool WebRtcVoiceMediaChannel::SetSend(SendFlags send) {
- desired_send_ = send;
- if (!send_streams_.empty()) {
- return ChangeSend(desired_send_);
- }
- return true;
-}
-
-bool WebRtcVoiceMediaChannel::PauseSend() {
- return ChangeSend(SEND_NOTHING);
-}
-
-bool WebRtcVoiceMediaChannel::ResumeSend() {
- return ChangeSend(desired_send_);
-}
-
-bool WebRtcVoiceMediaChannel::ChangeSend(SendFlags send) {
- if (send_ == send) {
- return true;
- }
-
- // Apply channel specific options when channel is enabled for sending.
- if (send == SEND_MICROPHONE) {
- engine()->ApplyOptions(options_);
- }
-
- // Change the settings on each send channel.
- for (const auto& ch : send_streams_) {
- if (!ChangeSend(ch.second->channel(), send)) {
- return false;
- }
- }
-
- send_ = send;
- return true;
-}
-
-bool WebRtcVoiceMediaChannel::ChangeSend(int channel, SendFlags send) {
- if (send == SEND_MICROPHONE) {
- if (engine()->voe()->base()->StartSend(channel) == -1) {
- LOG_RTCERR1(StartSend, channel);
- return false;
- }
- } else { // SEND_NOTHING
- RTC_DCHECK(send == SEND_NOTHING);
- if (engine()->voe()->base()->StopSend(channel) == -1) {
- LOG_RTCERR1(StopSend, channel);
- return false;
- }
- }
-
- return true;
-}
-
-bool WebRtcVoiceMediaChannel::SetAudioSend(uint32_t ssrc,
- bool enable,
- const AudioOptions* options,
- AudioRenderer* renderer) {
- RTC_DCHECK(worker_thread_checker_.CalledOnValidThread());
- // TODO(solenberg): The state change should be fully rolled back if any one of
- // these calls fail.
- if (!SetLocalRenderer(ssrc, renderer)) {
- return false;
- }
- if (!MuteStream(ssrc, !enable)) {
- return false;
- }
- if (enable && options) {
- return SetOptions(*options);
- }
- return true;
-}
-
-int WebRtcVoiceMediaChannel::CreateVoEChannel() {
- int id = engine()->CreateVoEChannel();
- if (id == -1) {
- LOG_RTCERR0(CreateVoEChannel);
- return -1;
- }
- if (engine()->voe()->network()->RegisterExternalTransport(id, *this) == -1) {
- LOG_RTCERR2(RegisterExternalTransport, id, this);
- engine()->voe()->base()->DeleteChannel(id);
- return -1;
- }
- return id;
-}
-
-bool WebRtcVoiceMediaChannel::DeleteVoEChannel(int channel) {
- if (engine()->voe()->network()->DeRegisterExternalTransport(channel) == -1) {
- LOG_RTCERR1(DeRegisterExternalTransport, channel);
- }
- if (engine()->voe()->base()->DeleteChannel(channel) == -1) {
- LOG_RTCERR1(DeleteChannel, channel);
- return false;
- }
- return true;
-}
-
-bool WebRtcVoiceMediaChannel::AddSendStream(const StreamParams& sp) {
- RTC_DCHECK(worker_thread_checker_.CalledOnValidThread());
- LOG(LS_INFO) << "AddSendStream: " << sp.ToString();
-
- uint32_t ssrc = sp.first_ssrc();
- RTC_DCHECK(0 != ssrc);
-
- if (GetSendChannelId(ssrc) != -1) {
- LOG(LS_ERROR) << "Stream already exists with ssrc " << ssrc;
- return false;
- }
-
- // Create a new channel for sending audio data.
- int channel = CreateVoEChannel();
- if (channel == -1) {
- return false;
- }
-
- // Save the channel to send_streams_, so that RemoveSendStream() can still
- // delete the channel in case failure happens below.
- webrtc::AudioTransport* audio_transport =
- engine()->voe()->base()->audio_transport();
- send_streams_.insert(std::make_pair(ssrc, new WebRtcAudioSendStream(
- channel, audio_transport, ssrc, sp.cname, send_rtp_extensions_, call_)));
-
- // Set the current codecs to be used for the new channel. We need to do this
- // after adding the channel to send_channels_, because of how max bitrate is
- // currently being configured by SetSendCodec().
- if (!send_codecs_.empty() && !SetSendCodecs(channel, send_codecs_)) {
- RemoveSendStream(ssrc);
- return false;
- }
-
- // At this point the channel's local SSRC has been updated. If the channel is
- // the first send channel make sure that all the receive channels are updated
- // with the same SSRC in order to send receiver reports.
- if (send_streams_.size() == 1) {
- receiver_reports_ssrc_ = ssrc;
- for (const auto& stream : recv_streams_) {
- int recv_channel = stream.second->channel();
- if (engine()->voe()->rtp()->SetLocalSSRC(recv_channel, ssrc) != 0) {
- LOG_RTCERR2(SetLocalSSRC, recv_channel, ssrc);
- return false;
- }
- engine()->voe()->base()->AssociateSendChannel(recv_channel, channel);
- LOG(LS_INFO) << "VoiceEngine channel #" << recv_channel
- << " is associated with channel #" << channel << ".";
- }
- }
-
- return ChangeSend(channel, desired_send_);
-}
-
-bool WebRtcVoiceMediaChannel::RemoveSendStream(uint32_t ssrc) {
- RTC_DCHECK(worker_thread_checker_.CalledOnValidThread());
- LOG(LS_INFO) << "RemoveSendStream: " << ssrc;
-
- auto it = send_streams_.find(ssrc);
- if (it == send_streams_.end()) {
- LOG(LS_WARNING) << "Try to remove stream with ssrc " << ssrc
- << " which doesn't exist.";
- return false;
- }
-
- int channel = it->second->channel();
- ChangeSend(channel, SEND_NOTHING);
-
- // Clean up and delete the send stream+channel.
- LOG(LS_INFO) << "Removing audio send stream " << ssrc
- << " with VoiceEngine channel #" << channel << ".";
- delete it->second;
- send_streams_.erase(it);
- if (!DeleteVoEChannel(channel)) {
- return false;
- }
- if (send_streams_.empty()) {
- ChangeSend(SEND_NOTHING);
- }
- return true;
-}
-
-bool WebRtcVoiceMediaChannel::AddRecvStream(const StreamParams& sp) {
- RTC_DCHECK(worker_thread_checker_.CalledOnValidThread());
- LOG(LS_INFO) << "AddRecvStream: " << sp.ToString();
-
- if (!ValidateStreamParams(sp)) {
- return false;
- }
-
- const uint32_t ssrc = sp.first_ssrc();
- if (ssrc == 0) {
- LOG(LS_WARNING) << "AddRecvStream with ssrc==0 is not supported.";
- return false;
- }
-
- // Remove the default receive stream if one had been created with this ssrc;
- // we'll recreate it then.
- if (IsDefaultRecvStream(ssrc)) {
- RemoveRecvStream(ssrc);
- }
-
- if (GetReceiveChannelId(ssrc) != -1) {
- LOG(LS_ERROR) << "Stream already exists with ssrc " << ssrc;
- return false;
- }
-
- // Create a new channel for receiving audio data.
- const int channel = CreateVoEChannel();
- if (channel == -1) {
- return false;
- }
-
- // Turn off all supported codecs.
- // TODO(solenberg): Remove once "no codecs" is the default state of a stream.
- for (webrtc::CodecInst voe_codec : webrtc::acm2::RentACodec::Database()) {
- voe_codec.pltype = -1;
- if (engine()->voe()->codec()->SetRecPayloadType(channel, voe_codec) == -1) {
- LOG_RTCERR2(SetRecPayloadType, channel, ToString(voe_codec));
- DeleteVoEChannel(channel);
- return false;
- }
- }
-
- // Only enable those configured for this channel.
- for (const auto& codec : recv_codecs_) {
- webrtc::CodecInst voe_codec;
- if (WebRtcVoiceEngine::ToCodecInst(codec, &voe_codec)) {
- voe_codec.pltype = codec.id;
- if (engine()->voe()->codec()->SetRecPayloadType(
- channel, voe_codec) == -1) {
- LOG_RTCERR2(SetRecPayloadType, channel, ToString(voe_codec));
- DeleteVoEChannel(channel);
- return false;
- }
- }
- }
-
- const int send_channel = GetSendChannelId(receiver_reports_ssrc_);
- if (send_channel != -1) {
- // Associate receive channel with first send channel (so the receive channel
- // can obtain RTT from the send channel)
- engine()->voe()->base()->AssociateSendChannel(channel, send_channel);
- LOG(LS_INFO) << "VoiceEngine channel #" << channel
- << " is associated with channel #" << send_channel << ".";
- }
-
- transport_cc_enabled_ =
- !send_codecs_.empty() ? HasTransportCc(send_codecs_[0]) : false;
-
- recv_streams_.insert(std::make_pair(
- ssrc, new WebRtcAudioReceiveStream(channel, ssrc, receiver_reports_ssrc_,
- transport_cc_enabled_, sp.sync_label,
- recv_rtp_extensions_, call_)));
-
- SetNack(channel, nack_enabled_);
- SetPlayout(channel, playout_);
-
- return true;
-}
-
-bool WebRtcVoiceMediaChannel::RemoveRecvStream(uint32_t ssrc) {
- RTC_DCHECK(worker_thread_checker_.CalledOnValidThread());
- LOG(LS_INFO) << "RemoveRecvStream: " << ssrc;
-
- const auto it = recv_streams_.find(ssrc);
- if (it == recv_streams_.end()) {
- LOG(LS_WARNING) << "Try to remove stream with ssrc " << ssrc
- << " which doesn't exist.";
- return false;
- }
-
- // Deregister default channel, if that's the one being destroyed.
- if (IsDefaultRecvStream(ssrc)) {
- default_recv_ssrc_ = -1;
- }
-
- const int channel = it->second->channel();
-
- // Clean up and delete the receive stream+channel.
- LOG(LS_INFO) << "Removing audio receive stream " << ssrc
- << " with VoiceEngine channel #" << channel << ".";
- it->second->SetRawAudioSink(nullptr);
- delete it->second;
- recv_streams_.erase(it);
- return DeleteVoEChannel(channel);
-}
-
-bool WebRtcVoiceMediaChannel::SetLocalRenderer(uint32_t ssrc,
- AudioRenderer* renderer) {
- auto it = send_streams_.find(ssrc);
- if (it == send_streams_.end()) {
- if (renderer) {
- // Return an error if trying to set a valid renderer with an invalid ssrc.
- LOG(LS_ERROR) << "SetLocalRenderer failed with ssrc "<< ssrc;
- return false;
- }
-
- // The channel likely has gone away, do nothing.
- return true;
- }
-
- if (renderer) {
- it->second->Start(renderer);
- } else {
- it->second->Stop();
- }
-
- return true;
-}
-
-bool WebRtcVoiceMediaChannel::GetActiveStreams(
- AudioInfo::StreamList* actives) {
- RTC_DCHECK(worker_thread_checker_.CalledOnValidThread());
- actives->clear();
- for (const auto& ch : recv_streams_) {
- int level = GetOutputLevel(ch.second->channel());
- if (level > 0) {
- actives->push_back(std::make_pair(ch.first, level));
- }
- }
- return true;
-}
-
-int WebRtcVoiceMediaChannel::GetOutputLevel() {
- RTC_DCHECK(worker_thread_checker_.CalledOnValidThread());
- int highest = 0;
- for (const auto& ch : recv_streams_) {
- highest = std::max(GetOutputLevel(ch.second->channel()), highest);
- }
- return highest;
-}
-
-int WebRtcVoiceMediaChannel::GetTimeSinceLastTyping() {
- int ret;
- if (engine()->voe()->processing()->TimeSinceLastTyping(ret) == -1) {
- // In case of error, log the info and continue
- LOG_RTCERR0(TimeSinceLastTyping);
- ret = -1;
- } else {
- ret *= 1000; // We return ms, webrtc returns seconds.
- }
- return ret;
-}
-
-void WebRtcVoiceMediaChannel::SetTypingDetectionParameters(int time_window,
- int cost_per_typing, int reporting_threshold, int penalty_decay,
- int type_event_delay) {
- if (engine()->voe()->processing()->SetTypingDetectionParameters(
- time_window, cost_per_typing,
- reporting_threshold, penalty_decay, type_event_delay) == -1) {
- // In case of error, log the info and continue
- LOG_RTCERR5(SetTypingDetectionParameters, time_window,
- cost_per_typing, reporting_threshold, penalty_decay,
- type_event_delay);
- }
-}
-
-bool WebRtcVoiceMediaChannel::SetOutputVolume(uint32_t ssrc, double volume) {
- RTC_DCHECK(worker_thread_checker_.CalledOnValidThread());
- if (ssrc == 0) {
- default_recv_volume_ = volume;
- if (default_recv_ssrc_ == -1) {
- return true;
- }
- ssrc = static_cast<uint32_t>(default_recv_ssrc_);
- }
- int ch_id = GetReceiveChannelId(ssrc);
- if (ch_id < 0) {
- LOG(LS_WARNING) << "Cannot find channel for ssrc:" << ssrc;
- return false;
- }
-
- if (-1 == engine()->voe()->volume()->SetChannelOutputVolumeScaling(ch_id,
- volume)) {
- LOG_RTCERR2(SetChannelOutputVolumeScaling, ch_id, volume);
- return false;
- }
- LOG(LS_INFO) << "SetOutputVolume to " << volume
- << " for channel " << ch_id << " and ssrc " << ssrc;
- return true;
-}
-
-bool WebRtcVoiceMediaChannel::CanInsertDtmf() {
- return dtmf_payload_type_ ? true : false;
-}
-
-bool WebRtcVoiceMediaChannel::InsertDtmf(uint32_t ssrc, int event,
- int duration) {
- RTC_DCHECK(worker_thread_checker_.CalledOnValidThread());
- LOG(LS_INFO) << "WebRtcVoiceMediaChannel::InsertDtmf";
- if (!dtmf_payload_type_) {
- return false;
- }
-
- // Figure out which WebRtcAudioSendStream to send the event on.
- auto it = ssrc != 0 ? send_streams_.find(ssrc) : send_streams_.begin();
- if (it == send_streams_.end()) {
- LOG(LS_WARNING) << "The specified ssrc " << ssrc << " is not in use.";
- return false;
- }
- if (event < kMinTelephoneEventCode ||
- event > kMaxTelephoneEventCode) {
- LOG(LS_WARNING) << "DTMF event code " << event << " out of range.";
- return false;
- }
- if (duration < kMinTelephoneEventDuration ||
- duration > kMaxTelephoneEventDuration) {
- LOG(LS_WARNING) << "DTMF event duration " << duration << " out of range.";
- return false;
- }
- return it->second->SendTelephoneEvent(*dtmf_payload_type_, event, duration);
-}
-
-void WebRtcVoiceMediaChannel::OnPacketReceived(
- rtc::Buffer* packet, const rtc::PacketTime& packet_time) {
- RTC_DCHECK(worker_thread_checker_.CalledOnValidThread());
-
- uint32_t ssrc = 0;
- if (!GetRtpSsrc(packet->data(), packet->size(), &ssrc)) {
- return;
- }
-
- // If we don't have a default channel, and the SSRC is unknown, create a
- // default channel.
- if (default_recv_ssrc_ == -1 && GetReceiveChannelId(ssrc) == -1) {
- StreamParams sp;
- sp.ssrcs.push_back(ssrc);
- LOG(LS_INFO) << "Creating default receive stream for SSRC=" << ssrc << ".";
- if (!AddRecvStream(sp)) {
- LOG(LS_WARNING) << "Could not create default receive stream.";
- return;
- }
- default_recv_ssrc_ = ssrc;
- SetOutputVolume(default_recv_ssrc_, default_recv_volume_);
- if (default_sink_) {
- rtc::scoped_ptr<webrtc::AudioSinkInterface> proxy_sink(
- new ProxySink(default_sink_.get()));
- SetRawAudioSink(default_recv_ssrc_, std::move(proxy_sink));
- }
- }
-
- // Forward packet to Call. If the SSRC is unknown we'll return after this.
- const webrtc::PacketTime webrtc_packet_time(packet_time.timestamp,
- packet_time.not_before);
- webrtc::PacketReceiver::DeliveryStatus delivery_result =
- call_->Receiver()->DeliverPacket(webrtc::MediaType::AUDIO,
- reinterpret_cast<const uint8_t*>(packet->data()), packet->size(),
- webrtc_packet_time);
- if (webrtc::PacketReceiver::DELIVERY_OK != delivery_result) {
- // If the SSRC is unknown here, route it to the default channel, if we have
- // one. See: https://bugs.chromium.org/p/webrtc/issues/detail?id=5208
- if (default_recv_ssrc_ == -1) {
- return;
- } else {
- ssrc = default_recv_ssrc_;
- }
- }
-
- // Find the channel to send this packet to. It must exist since webrtc::Call
- // was able to demux the packet.
- int channel = GetReceiveChannelId(ssrc);
- RTC_DCHECK(channel != -1);
-
- // Pass it off to the decoder.
- engine()->voe()->network()->ReceivedRTPPacket(
- channel, packet->data(), packet->size(), webrtc_packet_time);
-}
-
-void WebRtcVoiceMediaChannel::OnRtcpReceived(
- rtc::Buffer* packet, const rtc::PacketTime& packet_time) {
- RTC_DCHECK(worker_thread_checker_.CalledOnValidThread());
-
- // Forward packet to Call as well.
- const webrtc::PacketTime webrtc_packet_time(packet_time.timestamp,
- packet_time.not_before);
- call_->Receiver()->DeliverPacket(webrtc::MediaType::AUDIO,
- reinterpret_cast<const uint8_t*>(packet->data()), packet->size(),
- webrtc_packet_time);
-
- // Sending channels need all RTCP packets with feedback information.
- // Even sender reports can contain attached report blocks.
- // Receiving channels need sender reports in order to create
- // correct receiver reports.
- int type = 0;
- if (!GetRtcpType(packet->data(), packet->size(), &type)) {
- LOG(LS_WARNING) << "Failed to parse type from received RTCP packet";
- return;
- }
-
- // If it is a sender report, find the receive channel that is listening.
- if (type == kRtcpTypeSR) {
- uint32_t ssrc = 0;
- if (!GetRtcpSsrc(packet->data(), packet->size(), &ssrc)) {
- return;
- }
- int recv_channel_id = GetReceiveChannelId(ssrc);
- if (recv_channel_id != -1) {
- engine()->voe()->network()->ReceivedRTCPPacket(
- recv_channel_id, packet->data(), packet->size());
- }
- }
-
- // SR may continue RR and any RR entry may correspond to any one of the send
- // channels. So all RTCP packets must be forwarded all send channels. VoE
- // will filter out RR internally.
- for (const auto& ch : send_streams_) {
- engine()->voe()->network()->ReceivedRTCPPacket(
- ch.second->channel(), packet->data(), packet->size());
- }
-}
-
-bool WebRtcVoiceMediaChannel::MuteStream(uint32_t ssrc, bool muted) {
- RTC_DCHECK(worker_thread_checker_.CalledOnValidThread());
- int channel = GetSendChannelId(ssrc);
- if (channel == -1) {
- LOG(LS_WARNING) << "The specified ssrc " << ssrc << " is not in use.";
- return false;
- }
- if (engine()->voe()->volume()->SetInputMute(channel, muted) == -1) {
- LOG_RTCERR2(SetInputMute, channel, muted);
- return false;
- }
- // We set the AGC to mute state only when all the channels are muted.
- // This implementation is not ideal, instead we should signal the AGC when
- // the mic channel is muted/unmuted. We can't do it today because there
- // is no good way to know which stream is mapping to the mic channel.
- bool all_muted = muted;
- for (const auto& ch : send_streams_) {
- if (!all_muted) {
- break;
- }
- if (engine()->voe()->volume()->GetInputMute(ch.second->channel(),
- all_muted)) {
- LOG_RTCERR1(GetInputMute, ch.second->channel());
- return false;
- }
- }
-
- webrtc::AudioProcessing* ap = engine()->voe()->base()->audio_processing();
- if (ap) {
- ap->set_output_will_be_muted(all_muted);
- }
- return true;
-}
-
-// TODO(minyue): SetMaxSendBandwidth() is subject to be renamed to
-// SetMaxSendBitrate() in future.
-bool WebRtcVoiceMediaChannel::SetMaxSendBandwidth(int bps) {
- LOG(LS_INFO) << "WebRtcVoiceMediaChannel::SetMaxSendBandwidth.";
- return SetSendBitrateInternal(bps);
-}
-
-bool WebRtcVoiceMediaChannel::SetSendBitrateInternal(int bps) {
- LOG(LS_INFO) << "WebRtcVoiceMediaChannel::SetSendBitrateInternal.";
-
- send_bitrate_setting_ = true;
- send_bitrate_bps_ = bps;
-
- if (!send_codec_) {
- LOG(LS_INFO) << "The send codec has not been set up yet. "
- << "The send bitrate setting will be applied later.";
- return true;
- }
-
- // Bitrate is auto by default.
- // TODO(bemasc): Fix this so that if SetMaxSendBandwidth(50) is followed by
- // SetMaxSendBandwith(0), the second call removes the previous limit.
- if (bps <= 0)
- return true;
-
- webrtc::CodecInst codec = *send_codec_;
- bool is_multi_rate = WebRtcVoiceCodecs::IsCodecMultiRate(codec);
-
- if (is_multi_rate) {
- // If codec is multi-rate then just set the bitrate.
- codec.rate = bps;
- for (const auto& ch : send_streams_) {
- if (!SetSendCodec(ch.second->channel(), codec)) {
- LOG(LS_INFO) << "Failed to set codec " << codec.plname
- << " to bitrate " << bps << " bps.";
- return false;
- }
- }
- return true;
- } else {
- // If codec is not multi-rate and |bps| is less than the fixed bitrate
- // then fail. If codec is not multi-rate and |bps| exceeds or equal the
- // fixed bitrate then ignore.
- if (bps < codec.rate) {
- LOG(LS_INFO) << "Failed to set codec " << codec.plname
- << " to bitrate " << bps << " bps"
- << ", requires at least " << codec.rate << " bps.";
- return false;
- }
- return true;
- }
-}
-
-bool WebRtcVoiceMediaChannel::GetStats(VoiceMediaInfo* info) {
- RTC_DCHECK(worker_thread_checker_.CalledOnValidThread());
- RTC_DCHECK(info);
-
- // Get SSRC and stats for each sender.
- RTC_DCHECK(info->senders.size() == 0);
- for (const auto& stream : send_streams_) {
- webrtc::AudioSendStream::Stats stats = stream.second->GetStats();
- VoiceSenderInfo sinfo;
- sinfo.add_ssrc(stats.local_ssrc);
- sinfo.bytes_sent = stats.bytes_sent;
- sinfo.packets_sent = stats.packets_sent;
- sinfo.packets_lost = stats.packets_lost;
- sinfo.fraction_lost = stats.fraction_lost;
- sinfo.codec_name = stats.codec_name;
- sinfo.ext_seqnum = stats.ext_seqnum;
- sinfo.jitter_ms = stats.jitter_ms;
- sinfo.rtt_ms = stats.rtt_ms;
- sinfo.audio_level = stats.audio_level;
- sinfo.aec_quality_min = stats.aec_quality_min;
- sinfo.echo_delay_median_ms = stats.echo_delay_median_ms;
- sinfo.echo_delay_std_ms = stats.echo_delay_std_ms;
- sinfo.echo_return_loss = stats.echo_return_loss;
- sinfo.echo_return_loss_enhancement = stats.echo_return_loss_enhancement;
- sinfo.typing_noise_detected =
- (send_ == SEND_NOTHING ? false : stats.typing_noise_detected);
- info->senders.push_back(sinfo);
- }
-
- // Get SSRC and stats for each receiver.
- RTC_DCHECK(info->receivers.size() == 0);
- for (const auto& stream : recv_streams_) {
- webrtc::AudioReceiveStream::Stats stats = stream.second->GetStats();
- VoiceReceiverInfo rinfo;
- rinfo.add_ssrc(stats.remote_ssrc);
- rinfo.bytes_rcvd = stats.bytes_rcvd;
- rinfo.packets_rcvd = stats.packets_rcvd;
- rinfo.packets_lost = stats.packets_lost;
- rinfo.fraction_lost = stats.fraction_lost;
- rinfo.codec_name = stats.codec_name;
- rinfo.ext_seqnum = stats.ext_seqnum;
- rinfo.jitter_ms = stats.jitter_ms;
- rinfo.jitter_buffer_ms = stats.jitter_buffer_ms;
- rinfo.jitter_buffer_preferred_ms = stats.jitter_buffer_preferred_ms;
- rinfo.delay_estimate_ms = stats.delay_estimate_ms;
- rinfo.audio_level = stats.audio_level;
- rinfo.expand_rate = stats.expand_rate;
- rinfo.speech_expand_rate = stats.speech_expand_rate;
- rinfo.secondary_decoded_rate = stats.secondary_decoded_rate;
- rinfo.accelerate_rate = stats.accelerate_rate;
- rinfo.preemptive_expand_rate = stats.preemptive_expand_rate;
- rinfo.decoding_calls_to_silence_generator =
- stats.decoding_calls_to_silence_generator;
- rinfo.decoding_calls_to_neteq = stats.decoding_calls_to_neteq;
- rinfo.decoding_normal = stats.decoding_normal;
- rinfo.decoding_plc = stats.decoding_plc;
- rinfo.decoding_cng = stats.decoding_cng;
- rinfo.decoding_plc_cng = stats.decoding_plc_cng;
- rinfo.capture_start_ntp_time_ms = stats.capture_start_ntp_time_ms;
- info->receivers.push_back(rinfo);
- }
-
- return true;
-}
-
-void WebRtcVoiceMediaChannel::SetRawAudioSink(
- uint32_t ssrc,
- rtc::scoped_ptr<webrtc::AudioSinkInterface> sink) {
- RTC_DCHECK(worker_thread_checker_.CalledOnValidThread());
- LOG(LS_VERBOSE) << "WebRtcVoiceMediaChannel::SetRawAudioSink: ssrc:" << ssrc
- << " " << (sink ? "(ptr)" : "NULL");
- if (ssrc == 0) {
- if (default_recv_ssrc_ != -1) {
- rtc::scoped_ptr<webrtc::AudioSinkInterface> proxy_sink(
- sink ? new ProxySink(sink.get()) : nullptr);
- SetRawAudioSink(default_recv_ssrc_, std::move(proxy_sink));
- }
- default_sink_ = std::move(sink);
- return;
- }
- const auto it = recv_streams_.find(ssrc);
- if (it == recv_streams_.end()) {
- LOG(LS_WARNING) << "SetRawAudioSink: no recv stream" << ssrc;
- return;
- }
- it->second->SetRawAudioSink(std::move(sink));
-}
-
-int WebRtcVoiceMediaChannel::GetOutputLevel(int channel) {
- unsigned int ulevel = 0;
- int ret = engine()->voe()->volume()->GetSpeechOutputLevel(channel, ulevel);
- return (ret == 0) ? static_cast<int>(ulevel) : -1;
-}
-
-int WebRtcVoiceMediaChannel::GetReceiveChannelId(uint32_t ssrc) const {
- RTC_DCHECK(worker_thread_checker_.CalledOnValidThread());
- const auto it = recv_streams_.find(ssrc);
- if (it != recv_streams_.end()) {
- return it->second->channel();
- }
- return -1;
-}
-
-int WebRtcVoiceMediaChannel::GetSendChannelId(uint32_t ssrc) const {
- RTC_DCHECK(worker_thread_checker_.CalledOnValidThread());
- const auto it = send_streams_.find(ssrc);
- if (it != send_streams_.end()) {
- return it->second->channel();
- }
- return -1;
-}
-
-bool WebRtcVoiceMediaChannel::SetPlayout(int channel, bool playout) {
- if (playout) {
- LOG(LS_INFO) << "Starting playout for channel #" << channel;
- if (engine()->voe()->base()->StartPlayout(channel) == -1) {
- LOG_RTCERR1(StartPlayout, channel);
- return false;
- }
- } else {
- LOG(LS_INFO) << "Stopping playout for channel #" << channel;
- engine()->voe()->base()->StopPlayout(channel);
- }
- return true;
-}
-} // namespace cricket
-
-#endif // HAVE_WEBRTC_VOICE
« no previous file with comments | « webrtc/media/webrtc/webrtcvoiceengine.h ('k') | webrtc/media/webrtc/webrtcvoiceengine_unittest.cc » ('j') | no next file with comments »

Powered by Google App Engine
This is Rietveld 408576698