Multi frequency DTMF support - receiver side

webrtc/modules/audio_coding/neteq/neteq_impl_unittest.cc

Index: webrtc/modules/audio_coding/neteq/neteq_impl_unittest.cc
diff --git a/webrtc/modules/audio_coding/neteq/neteq_impl_unittest.cc b/webrtc/modules/audio_coding/neteq/neteq_impl_unittest.cc
index 1859e566e7d2cb4d728d69effea09932619d8c34..777a05c1c40421031d9c8d47f8d65a3fb556dc05 100644
--- a/webrtc/modules/audio_coding/neteq/neteq_impl_unittest.cc
+++ b/webrtc/modules/audio_coding/neteq/neteq_impl_unittest.cc
@@ -173,6 +173,40 @@ class NetEqImplTest : public ::testing::Test {
     }
   }
 
+  void TestDtmfPacket(NetEqDecoder decoder_type) {
+    UseNoMocks();
+    CreateInstance();
+    const size_t kPayloadLength = 4;
+    const uint8_t kPayloadType = 110;
+    const uint32_t kReceiveTime = 17;
+    const int kSampleRateHz = 8000;

[ossu, 2016/10/11 17:56:17]

So DTMF sample rate doesn't depend on clock rate, but the clock rate NetEq runs
at (which is determined by the most recently used speech codec)? Am I
understanding this correctly?

[the sun, 2016/10/11 19:17:17]

The sample rate is set when NetEq is constructed and defaults to 8000. This is
the output rate and has nothing to do with the rate of encoded audio (either
bit-, rtp clock-, or actual sample rate). Since DTMF is only encoded events, the
only rate that matters here is the output sample rate, so that the tone
generator can give us something sensible. Henrik already answered about the rtp
rate - IIUC it is duly ignored for DTMF, regardless of rate, and the rate of the
audio codec is used to align playout times.

+    // Event: 2, E bit, Volume: 0x3F, Length: 0x10F0.
+    uint8_t payload[kPayloadLength] = { 0x02, 0x80 + 0x3F, 0x10, 0xF0 };
+    WebRtcRTPHeader rtp_header;
+    rtp_header.header.payloadType = kPayloadType;
+    rtp_header.header.sequenceNumber = 0x1234;
+    rtp_header.header.timestamp = 0x12345678;
+    rtp_header.header.ssrc = 0x87654321;
+
+    EXPECT_EQ(NetEq::kOK, neteq_->RegisterPayloadType(
+        decoder_type, "telephone-event", kPayloadType));
+
+    // Insert first packet.
+    EXPECT_EQ(NetEq::kOK,
+              neteq_->InsertPacket(rtp_header, payload, kReceiveTime));
+
+    // Pull audio once.
+    const size_t kMaxOutputSize =
+        static_cast<size_t>(10 * kSampleRateHz / 1000);
+    AudioFrame output;
+    bool muted;
+    EXPECT_EQ(NetEq::kOK, neteq_->GetAudio(&output, &muted));
+    ASSERT_FALSE(muted);
+    ASSERT_EQ(kMaxOutputSize, output.samples_per_channel_);
+    EXPECT_EQ(1u, output.num_channels_);
+    EXPECT_EQ(AudioFrame::kNormalSpeech, output.speech_type_);

[ossu, 2016/10/11 17:56:18]

Do you want to check that there is actually audio generated as well? Like that
the frame isn't all zeroes or something? (I guess _actually_ validating the
waveform is a bit much?)

[the sun, 2016/10/11 19:17:17]

Done.

+  }
+
   std::unique_ptr<NetEqImpl> neteq_;
   NetEq::Config config_;
   TickTimer* tick_timer_ = nullptr;
@@ -381,6 +415,22 @@ TEST_F(NetEqImplTest, InsertPacketsUntilBufferIsFull) {
   EXPECT_EQ(rtp_header.header.sequenceNumber, test_header->sequenceNumber);
 }
 
+TEST_F(NetEqImplTest, TestDtmfPacketAVT) {
+  TestDtmfPacket(NetEqDecoder::kDecoderAVT);
+}
+
+TEST_F(NetEqImplTest, TestDtmfPacketAVT16kHz) {
+  TestDtmfPacket(NetEqDecoder::kDecoderAVT16kHz);
+}
+
+TEST_F(NetEqImplTest, TestDtmfPacketAVT32kHz) {
+  TestDtmfPacket(NetEqDecoder::kDecoderAVT32kHz);
+}
+
+TEST_F(NetEqImplTest, TestDtmfPacketAVT48kHz) {
+  TestDtmfPacket(NetEqDecoder::kDecoderAVT48kHz);
+}
+
 // This test verifies that timestamps propagate from the incoming packets
 // through to the sync buffer and to the playout timestamp.
 TEST_F(NetEqImplTest, VerifyTimestampPropagation) {