Add a wrapper around PushSincResampler and the old Resampler.
The old resampler is used whenever it supports the requested rates. Otherwise
the sinc resampler is enabled.
Integrated with output_mixer in order to test the change through
output_mixer_unittest. The sinc resampler will not yet be used, since we don't
feed VoE with any rates that trigger it.
BUG=webrtc:1395
R=bjornv@webrtc.org
Review URL: https://webrtc-codereview.appspot.com/1355004
git-svn-id: http://webrtc.googlecode.com/svn/trunk@3915 4adac7df-926f-26a2-2b94-8c16560cd09d
diff --git a/webrtc/voice_engine/output_mixer.cc b/webrtc/voice_engine/output_mixer.cc
index a124564..a8e4177 100644
--- a/webrtc/voice_engine/output_mixer.cc
+++ b/webrtc/voice_engine/output_mixer.cc
@@ -8,16 +8,16 @@
* be found in the AUTHORS file in the root of the source tree.
*/
-#include "output_mixer.h"
+#include "webrtc/voice_engine/output_mixer.h"
-#include "audio_processing.h"
-#include "audio_frame_operations.h"
-#include "critical_section_wrapper.h"
-#include "file_wrapper.h"
-#include "output_mixer_internal.h"
-#include "statistics.h"
-#include "trace.h"
-#include "voe_external_media.h"
+#include "webrtc/modules/audio_processing/include/audio_processing.h"
+#include "webrtc/modules/utility/interface/audio_frame_operations.h"
+#include "webrtc/system_wrappers/interface/critical_section_wrapper.h"
+#include "webrtc/system_wrappers/interface/file_wrapper.h"
+#include "webrtc/system_wrappers/interface/trace.h"
+#include "webrtc/voice_engine/include/voe_external_media.h"
+#include "webrtc/voice_engine/output_mixer_internal.h"
+#include "webrtc/voice_engine/statistics.h"
namespace webrtc {
@@ -528,7 +528,7 @@
frame->sample_rate_hz_ = sample_rate_hz;
// TODO(andrew): Ideally the downmixing would occur much earlier, in
// AudioCodingModule.
- return RemixAndResample(_audioFrame, &_resampler, frame);
+ return RemixAndResample(_audioFrame, &resampler_, frame);
}
int32_t
@@ -602,7 +602,7 @@
AudioFrame frame;
frame.num_channels_ = 1;
frame.sample_rate_hz_ = _audioProcessingModulePtr->sample_rate_hz();
- if (RemixAndResample(_audioFrame, &_apmResampler, &frame) == -1)
+ if (RemixAndResample(_audioFrame, &audioproc_resampler_, &frame) == -1)
return;
if (_audioProcessingModulePtr->AnalyzeReverseStream(&frame) == -1) {
diff --git a/webrtc/voice_engine/output_mixer.h b/webrtc/voice_engine/output_mixer.h
index e2ca366..b98f88e 100644
--- a/webrtc/voice_engine/output_mixer.h
+++ b/webrtc/voice_engine/output_mixer.h
@@ -11,14 +11,14 @@
#ifndef WEBRTC_VOICE_ENGINE_OUTPUT_MIXER_H_
#define WEBRTC_VOICE_ENGINE_OUTPUT_MIXER_H_
-#include "audio_conference_mixer.h"
-#include "audio_conference_mixer_defines.h"
-#include "common_types.h"
-#include "dtmf_inband.h"
-#include "file_recorder.h"
-#include "level_indicator.h"
-#include "resampler.h"
-#include "voice_engine_defines.h"
+#include "webrtc/common_audio/resampler/include/push_resampler.h"
+#include "webrtc/common_types.h"
+#include "webrtc/modules/audio_conference_mixer/interface/audio_conference_mixer.h"
+#include "webrtc/modules/audio_conference_mixer/interface/audio_conference_mixer_defines.h"
+#include "webrtc/modules/utility/interface/file_recorder.h"
+#include "webrtc/voice_engine/dtmf_inband.h"
+#include "webrtc/voice_engine/level_indicator.h"
+#include "webrtc/voice_engine/voice_engine_defines.h"
namespace webrtc {
@@ -133,8 +133,8 @@
CriticalSectionWrapper& _fileCritSect;
AudioConferenceMixer& _mixerModule;
AudioFrame _audioFrame;
- Resampler _resampler; // converts mixed audio to fit ADM format
- Resampler _apmResampler; // converts mixed audio to fit APM rate
+ PushResampler resampler_; // converts mixed audio to fit ADM format
+ PushResampler audioproc_resampler_; // converts mixed audio to fit APM rate
AudioLevel _audioLevel; // measures audio level for the combined signal
DtmfInband _dtmfGenerator;
int _instanceId;
diff --git a/webrtc/voice_engine/output_mixer_internal.cc b/webrtc/voice_engine/output_mixer_internal.cc
index dfa7d95..55eedb3 100644
--- a/webrtc/voice_engine/output_mixer_internal.cc
+++ b/webrtc/voice_engine/output_mixer_internal.cc
@@ -8,18 +8,19 @@
* be found in the AUTHORS file in the root of the source tree.
*/
-#include "output_mixer_internal.h"
+#include "webrtc/voice_engine/output_mixer_internal.h"
-#include "audio_frame_operations.h"
-#include "common_audio/resampler/include/resampler.h"
-#include "module_common_types.h"
-#include "trace.h"
+#include "webrtc/common_audio/resampler/include/push_resampler.h"
+#include "webrtc/modules/interface/module_common_types.h"
+#include "webrtc/modules/utility/interface/audio_frame_operations.h"
+#include "webrtc/system_wrappers/interface/logging.h"
+#include "webrtc/system_wrappers/interface/trace.h"
namespace webrtc {
namespace voe {
int RemixAndResample(const AudioFrame& src_frame,
- Resampler* resampler,
+ PushResampler* resampler,
AudioFrame* dst_frame) {
const int16_t* audio_ptr = src_frame.data_;
int audio_ptr_num_channels = src_frame.num_channels_;
@@ -34,30 +35,26 @@
audio_ptr_num_channels = 1;
}
- const ResamplerType resampler_type = audio_ptr_num_channels == 1 ?
- kResamplerSynchronous : kResamplerSynchronousStereo;
- if (resampler->ResetIfNeeded(src_frame.sample_rate_hz_,
- dst_frame->sample_rate_hz_,
- resampler_type) == -1) {
+ if (resampler->InitializeIfNeeded(src_frame.sample_rate_hz_,
+ dst_frame->sample_rate_hz_,
+ audio_ptr_num_channels) == -1) {
dst_frame->CopyFrom(src_frame);
- WEBRTC_TRACE(kTraceError, kTraceVoice, -1,
- "%s ResetIfNeeded failed", __FUNCTION__);
+ LOG_FERR3(LS_ERROR, InitializeIfNeeded, src_frame.sample_rate_hz_,
+ dst_frame->sample_rate_hz_, audio_ptr_num_channels);
return -1;
}
- int out_length = 0;
- if (resampler->Push(audio_ptr,
- src_frame.samples_per_channel_* audio_ptr_num_channels,
- dst_frame->data_,
- AudioFrame::kMaxDataSizeSamples,
- out_length) == 0) {
- dst_frame->samples_per_channel_ = out_length / audio_ptr_num_channels;
- } else {
+ const int src_length = src_frame.samples_per_channel_ *
+ audio_ptr_num_channels;
+ int out_length = resampler->Resample(audio_ptr, src_length, dst_frame->data_,
+ AudioFrame::kMaxDataSizeSamples);
+ if (out_length == -1) {
dst_frame->CopyFrom(src_frame);
- WEBRTC_TRACE(kTraceError, kTraceVoice, -1,
- "%s resampling failed", __FUNCTION__);
+ LOG_FERR3(LS_ERROR, Resample, src_length, dst_frame->data_,
+ AudioFrame::kMaxDataSizeSamples);
return -1;
}
+ dst_frame->samples_per_channel_ = out_length / audio_ptr_num_channels;
// Upmix after resampling.
if (src_frame.num_channels_ == 1 && dst_frame->num_channels_ == 2) {
diff --git a/webrtc/voice_engine/output_mixer_internal.h b/webrtc/voice_engine/output_mixer_internal.h
index 8d23a14..88a3a5b 100644
--- a/webrtc/voice_engine/output_mixer_internal.h
+++ b/webrtc/voice_engine/output_mixer_internal.h
@@ -14,7 +14,7 @@
namespace webrtc {
class AudioFrame;
-class Resampler;
+class PushResampler;
namespace voe {
@@ -24,7 +24,7 @@
//
// On failure, returns -1 and copies |src_frame| to |dst_frame|.
int RemixAndResample(const AudioFrame& src_frame,
- Resampler* resampler,
+ PushResampler* resampler,
AudioFrame* dst_frame);
} // namespace voe
diff --git a/webrtc/voice_engine/output_mixer_unittest.cc b/webrtc/voice_engine/output_mixer_unittest.cc
index dbcb251..24d3917 100644
--- a/webrtc/voice_engine/output_mixer_unittest.cc
+++ b/webrtc/voice_engine/output_mixer_unittest.cc
@@ -10,10 +10,9 @@
#include <math.h>
-#include "gtest/gtest.h"
-
-#include "output_mixer.h"
-#include "output_mixer_internal.h"
+#include "testing/gtest/include/gtest/gtest.h"
+#include "webrtc/voice_engine/output_mixer.h"
+#include "webrtc/voice_engine/output_mixer_internal.h"
namespace webrtc {
namespace voe {
@@ -32,7 +31,7 @@
void RunResampleTest(int src_channels, int src_sample_rate_hz,
int dst_channels, int dst_sample_rate_hz);
- Resampler resampler_;
+ PushResampler resampler_;
AudioFrame src_frame_;
AudioFrame dst_frame_;
AudioFrame golden_frame_;
@@ -42,6 +41,7 @@
// used so non-integer values result in rounding error, but not an accumulating
// error.
void SetMonoFrame(AudioFrame* frame, float data, int sample_rate_hz) {
+ memset(frame->data_, 0, sizeof(frame->data_));
frame->num_channels_ = 1;
frame->sample_rate_hz_ = sample_rate_hz;
frame->samples_per_channel_ = sample_rate_hz / 100;
@@ -59,6 +59,7 @@
// each channel respectively.
void SetStereoFrame(AudioFrame* frame, float left, float right,
int sample_rate_hz) {
+ memset(frame->data_, 0, sizeof(frame->data_));
frame->num_channels_ = 2;
frame->sample_rate_hz_ = sample_rate_hz;
frame->samples_per_channel_ = sample_rate_hz / 100;
@@ -80,13 +81,14 @@
}
// Computes the best SNR based on the error between |ref_frame| and
-// |test_frame|. It allows for up to a 30 sample delay between the signals to
-// compensate for the resampling delay.
-float ComputeSNR(const AudioFrame& ref_frame, const AudioFrame& test_frame) {
+// |test_frame|. It allows for up to a |max_delay| in samples between the
+// signals to compensate for the resampling delay.
+float ComputeSNR(const AudioFrame& ref_frame, const AudioFrame& test_frame,
+ int max_delay) {
VerifyParams(ref_frame, test_frame);
float best_snr = 0;
int best_delay = 0;
- for (int delay = 0; delay < 30; delay++) {
+ for (int delay = 0; delay <= max_delay; delay++) {
float mse = 0;
float variance = 0;
for (int i = 0; i < ref_frame.samples_per_channel_ *
@@ -120,14 +122,14 @@
int src_sample_rate_hz,
int dst_channels,
int dst_sample_rate_hz) {
- Resampler resampler; // Create a new one with every test.
- const int16_t kSrcLeft = 60; // Shouldn't overflow for any used sample rate.
- const int16_t kSrcRight = 30;
- const float kResamplingFactor = (1.0 * src_sample_rate_hz) /
+ PushResampler resampler; // Create a new one with every test.
+ const int16_t kSrcLeft = 30; // Shouldn't overflow for any used sample rate.
+ const int16_t kSrcRight = 15;
+ const float resampling_factor = (1.0 * src_sample_rate_hz) /
dst_sample_rate_hz;
- const float kDstLeft = kResamplingFactor * kSrcLeft;
- const float kDstRight = kResamplingFactor * kSrcRight;
- const float kDstMono = (kDstLeft + kDstRight) / 2;
+ const float dst_left = resampling_factor * kSrcLeft;
+ const float dst_right = resampling_factor * kSrcRight;
+ const float dst_mono = (dst_left + dst_right) / 2;
if (src_channels == 1)
SetMonoFrame(&src_frame_, kSrcLeft, src_sample_rate_hz);
else
@@ -136,27 +138,27 @@
if (dst_channels == 1) {
SetMonoFrame(&dst_frame_, 0, dst_sample_rate_hz);
if (src_channels == 1)
- SetMonoFrame(&golden_frame_, kDstLeft, dst_sample_rate_hz);
+ SetMonoFrame(&golden_frame_, dst_left, dst_sample_rate_hz);
else
- SetMonoFrame(&golden_frame_, kDstMono, dst_sample_rate_hz);
+ SetMonoFrame(&golden_frame_, dst_mono, dst_sample_rate_hz);
} else {
SetStereoFrame(&dst_frame_, 0, 0, dst_sample_rate_hz);
if (src_channels == 1)
- SetStereoFrame(&golden_frame_, kDstLeft, kDstLeft, dst_sample_rate_hz);
+ SetStereoFrame(&golden_frame_, dst_left, dst_left, dst_sample_rate_hz);
else
- SetStereoFrame(&golden_frame_, kDstLeft, kDstRight, dst_sample_rate_hz);
+ SetStereoFrame(&golden_frame_, dst_left, dst_right, dst_sample_rate_hz);
}
+ // The sinc resampler has a known delay, which we compute here. Multiplying by
+ // two gives us a crude maximum for any resampling, as the old resampler
+ // typically (but not always) has lower delay.
+ static const int kInputKernelDelaySamples = 16;
+ const int max_delay = static_cast<double>(dst_sample_rate_hz)
+ / src_sample_rate_hz * kInputKernelDelaySamples * dst_channels * 2;
printf("(%d, %d Hz) -> (%d, %d Hz) ", // SNR reported on the same line later.
src_channels, src_sample_rate_hz, dst_channels, dst_sample_rate_hz);
EXPECT_EQ(0, RemixAndResample(src_frame_, &resampler, &dst_frame_));
- EXPECT_GT(ComputeSNR(golden_frame_, dst_frame_), 40.0f);
-}
-
-TEST_F(OutputMixerTest, RemixAndResampleFailsWithBadSampleRate) {
- SetMonoFrame(&dst_frame_, 10, 44100);
- EXPECT_EQ(-1, RemixAndResample(src_frame_, &resampler_, &dst_frame_));
- VerifyFramesAreEqual(src_frame_, dst_frame_);
+ EXPECT_GT(ComputeSNR(golden_frame_, dst_frame_, max_delay), 39.0f);
}
TEST_F(OutputMixerTest, RemixAndResampleCopyFrameSucceeds) {
@@ -190,10 +192,9 @@
}
TEST_F(OutputMixerTest, RemixAndResampleSucceeds) {
- // We don't attempt to be exhaustive here, but just get good coverage. Some
- // combinations of rates will not be resampled, and some give an odd
- // resampling factor which makes it more difficult to evaluate.
- const int kSampleRates[] = {16000, 32000, 48000};
+ // TODO(ajm): convert this to the parameterized TEST_P style used in
+ // sinc_resampler_unittest.cc. We can then easily add tighter SNR thresholds.
+ const int kSampleRates[] = {8000, 16000, 32000, 44100, 48000, 96000};
const int kSampleRatesSize = sizeof(kSampleRates) / sizeof(*kSampleRates);
const int kChannels[] = {1, 2};
const int kChannelsSize = sizeof(kChannels) / sizeof(*kChannels);