NetEq: Don't interpolate longer than the output size

webrtc/modules/audio_coding/neteq/normal.cc

 /*
  *  Copyright (c) 2012 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.
  */
 
 #include "webrtc/modules/audio_coding/neteq/normal.h"
 
 #include <string.h>  // memset, memcpy
 
 #include <algorithm>  // min
 
+#include "webrtc/base/checks.h"
 #include "webrtc/common_audio/signal_processing/include/signal_processing_library.h"
 #include "webrtc/modules/audio_coding/codecs/audio_decoder.h"
 #include "webrtc/modules/audio_coding/neteq/audio_multi_vector.h"
 #include "webrtc/modules/audio_coding/neteq/background_noise.h"
 #include "webrtc/modules/audio_coding/neteq/decoder_database.h"
 #include "webrtc/modules/audio_coding/neteq/expand.h"
 
 namespace webrtc {
 
 int Normal::Process(const int16_t* input,
                     size_t length,
                     Modes last_mode,
                     int16_t* external_mute_factor_array,
                     AudioMultiVector* output) {
   if (length == 0) {
     // Nothing to process.
     output->Clear();
     return static_cast<int>(length);
   }
 
-  assert(output->Empty());
+  RTC_DCHECK(output->Empty());
   // Output should be empty at this point.
   if (length % output->Channels() != 0) {
     // The length does not match the number of channels.
     output->Clear();
     return 0;
   }
   output->PushBackInterleaved(input, length);
 
   const int fs_mult = fs_hz_ / 8000;
-  assert(fs_mult > 0);
+  RTC_DCHECK_GT(fs_mult, 0);
   // fs_shift = log2(fs_mult), rounded down.
   // Note that |fs_shift| is not "exact" for 48 kHz.
   // TODO(hlundin): Investigate this further.
   const int fs_shift = 30 - WebRtcSpl_NormW32(fs_mult);
 
   // Check if last RecOut call resulted in an Expand. If so, we have to take
   // care of some cross-fading and unmuting.
   if (last_mode == kModeExpand) {
     // Generate interpolation data using Expand.
     // First, set Expand parameters to appropriate values.
@@ skipping 50 matching lines @@
       }
       if (mute_factor > external_mute_factor_array[channel_ix]) {
         external_mute_factor_array[channel_ix] =
             static_cast<int16_t>(std::min(mute_factor, 16384));
       }
 
       // If muted increase by 0.64 for every 20 ms (NB/WB 0.0040/0.0020 in Q14).
       int increment = 64 / fs_mult;
       for (size_t i = 0; i < length_per_channel; i++) {
         // Scale with mute factor.
-        assert(channel_ix < output->Channels());
-        assert(i < output->Size());
+        RTC_DCHECK_LT(channel_ix, output->Channels());
+        RTC_DCHECK_LT(i, output->Size());
         int32_t scaled_signal = (*output)[channel_ix][i] *
             external_mute_factor_array[channel_ix];
         // Shift 14 with proper rounding.
         (*output)[channel_ix][i] =
             static_cast<int16_t>((scaled_signal + 8192) >> 14);
         // Increase mute_factor towards 16384.
         external_mute_factor_array[channel_ix] = static_cast<int16_t>(std::min(
             external_mute_factor_array[channel_ix] + increment, 16384));
       }
 
       // Interpolate the expanded data into the new vector.
       // (NB/WB/SWB32/SWB48 8/16/32/48 samples.)
-      assert(fs_shift < 3);  // Will always be 0, 1, or, 2.
+      RTC_DCHECK_LT(fs_shift, 3);  // Will always be 0, 1, or, 2.
       increment = 4 >> fs_shift;
       int fraction = increment;
-      for (size_t i = 0; i < static_cast<size_t>(8 * fs_mult); i++) {
+      // Don't interpolate over more samples than what is in output. When this
+      // cap strikes, the interpolation will likely sound worse, but this is an
+      // emergency operation in response to unexpected input.
+      const size_t interp_len_samples =
+          std::min(static_cast<size_t>(8 * fs_mult), output->Size());

[kwiberg-webrtc, 2016/11/14 14:36:33]

Oooh, I just realized that it's possible to define min and max so that we can
feed them any pair of integer types, and still always be able to represent the
answer in a statically determined type that's one of the two input types.

+      for (size_t i = 0; i < interp_len_samples; ++i) {
         // TODO(hlundin): Add 16 instead of 8 for correct rounding. Keeping 8
         // now for legacy bit-exactness.
-        assert(channel_ix < output->Channels());
-        assert(i < output->Size());
+        RTC_DCHECK_LT(channel_ix, output->Channels());
+        RTC_DCHECK_LT(i, output->Size());
         (*output)[channel_ix][i] =
             static_cast<int16_t>((fraction * (*output)[channel_ix][i] +
                 (32 - fraction) * expanded[channel_ix][i] + 8) >> 5);
         fraction += increment;
       }
     }
   } else if (last_mode == kModeRfc3389Cng) {
-    assert(output->Channels() == 1);  // Not adapted for multi-channel yet.
+    RTC_DCHECK_EQ(output->Channels(), 1);  // Not adapted for multi-channel yet.
     static const size_t kCngLength = 48;
     RTC_DCHECK_LE(static_cast<size_t>(8 * fs_mult), kCngLength);
     int16_t cng_output[kCngLength];
     // Reset mute factor and start up fresh.
     external_mute_factor_array[0] = 16384;
     ComfortNoiseDecoder* cng_decoder = decoder_database_->GetActiveCngDecoder();
 
     if (cng_decoder) {
       // Generate long enough for 48kHz.
       if (!cng_decoder->Generate(cng_output, 0)) {
         // Error returned; set return vector to all zeros.
         memset(cng_output, 0, sizeof(cng_output));
       }
     } else {
       // If no CNG instance is defined, just copy from the decoded data.
       // (This will result in interpolating the decoded with itself.)
       (*output)[0].CopyTo(fs_mult * 8, 0, cng_output);
     }
     // Interpolate the CNG into the new vector.
     // (NB/WB/SWB32/SWB48 8/16/32/48 samples.)
-    assert(fs_shift < 3);  // Will always be 0, 1, or, 2.
+    RTC_DCHECK_LT(fs_shift, 3);  // Will always be 0, 1, or, 2.
     int16_t increment = 4 >> fs_shift;
     int16_t fraction = increment;
     for (size_t i = 0; i < static_cast<size_t>(8 * fs_mult); i++) {
       // TODO(hlundin): Add 16 instead of 8 for correct rounding. Keeping 8 now
       // for legacy bit-exactness.
       (*output)[0][i] = (fraction * (*output)[0][i] +
           (32 - fraction) * cng_output[i] + 8) >> 5;
       fraction += increment;
     }
   } else if (external_mute_factor_array[0] < 16384) {
     // Previous was neither of Expand, FadeToBGN or RFC3389_CNG, but we are
     // still ramping up from previous muting.
     // If muted increase by 0.64 for every 20 ms (NB/WB 0.0040/0.0020 in Q14).
     int increment = 64 / fs_mult;
     size_t length_per_channel = length / output->Channels();
     for (size_t i = 0; i < length_per_channel; i++) {
       for (size_t channel_ix = 0; channel_ix < output->Channels();
           ++channel_ix) {
         // Scale with mute factor.
-        assert(channel_ix < output->Channels());
-        assert(i < output->Size());
+        RTC_DCHECK_LT(channel_ix, output->Channels());
+        RTC_DCHECK_LT(i, output->Size());
         int32_t scaled_signal = (*output)[channel_ix][i] *
             external_mute_factor_array[channel_ix];
         // Shift 14 with proper rounding.
         (*output)[channel_ix][i] =
             static_cast<int16_t>((scaled_signal + 8192) >> 14);
         // Increase mute_factor towards 16384.
         external_mute_factor_array[channel_ix] = static_cast<int16_t>(std::min(
             16384, external_mute_factor_array[channel_ix] + increment));
       }
     }
   }
 
   return static_cast<int>(length);
 }
 
 }  // namespace webrtc