mirror of
https://github.com/oxen-io/session-android.git
synced 2024-12-19 14:37:32 +00:00
388 lines
12 KiB
C++
388 lines
12 KiB
C++
|
/*
|
||
|
* Copyright (c) 2013 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/main/test/opus_test.h"
|
||
|
|
||
|
#include <assert.h>
|
||
|
|
||
|
#include <string>
|
||
|
|
||
|
#include "testing/gtest/include/gtest/gtest.h"
|
||
|
#include "webrtc/common_types.h"
|
||
|
#include "webrtc/engine_configurations.h"
|
||
|
#include "webrtc/modules/audio_coding/codecs/opus/interface/opus_interface.h"
|
||
|
#include "webrtc/modules/audio_coding/main/interface/audio_coding_module_typedefs.h"
|
||
|
#include "webrtc/modules/audio_coding/main/acm2/acm_codec_database.h"
|
||
|
#include "webrtc/modules/audio_coding/main/acm2/acm_opus.h"
|
||
|
#include "webrtc/modules/audio_coding/main/test/TestStereo.h"
|
||
|
#include "webrtc/modules/audio_coding/main/test/utility.h"
|
||
|
#include "webrtc/system_wrappers/interface/trace.h"
|
||
|
#include "webrtc/test/testsupport/fileutils.h"
|
||
|
|
||
|
namespace webrtc {
|
||
|
|
||
|
OpusTest::OpusTest()
|
||
|
: acm_receiver_(AudioCodingModule::Create(0)),
|
||
|
channel_a2b_(NULL),
|
||
|
counter_(0),
|
||
|
payload_type_(255),
|
||
|
rtp_timestamp_(0) {}
|
||
|
|
||
|
OpusTest::~OpusTest() {
|
||
|
if (channel_a2b_ != NULL) {
|
||
|
delete channel_a2b_;
|
||
|
channel_a2b_ = NULL;
|
||
|
}
|
||
|
if (opus_mono_encoder_ != NULL) {
|
||
|
WebRtcOpus_EncoderFree(opus_mono_encoder_);
|
||
|
opus_mono_encoder_ = NULL;
|
||
|
}
|
||
|
if (opus_stereo_encoder_ != NULL) {
|
||
|
WebRtcOpus_EncoderFree(opus_stereo_encoder_);
|
||
|
opus_stereo_encoder_ = NULL;
|
||
|
}
|
||
|
if (opus_mono_decoder_ != NULL) {
|
||
|
WebRtcOpus_DecoderFree(opus_mono_decoder_);
|
||
|
opus_mono_decoder_ = NULL;
|
||
|
}
|
||
|
if (opus_stereo_decoder_ != NULL) {
|
||
|
WebRtcOpus_DecoderFree(opus_stereo_decoder_);
|
||
|
opus_stereo_decoder_ = NULL;
|
||
|
}
|
||
|
}
|
||
|
|
||
|
void OpusTest::Perform() {
|
||
|
#ifndef WEBRTC_CODEC_OPUS
|
||
|
// Opus isn't defined, exit.
|
||
|
return;
|
||
|
#else
|
||
|
uint16_t frequency_hz;
|
||
|
int audio_channels;
|
||
|
int16_t test_cntr = 0;
|
||
|
|
||
|
// Open both mono and stereo test files in 32 kHz.
|
||
|
const std::string file_name_stereo =
|
||
|
webrtc::test::ResourcePath("audio_coding/teststereo32kHz", "pcm");
|
||
|
const std::string file_name_mono =
|
||
|
webrtc::test::ResourcePath("audio_coding/testfile32kHz", "pcm");
|
||
|
frequency_hz = 32000;
|
||
|
in_file_stereo_.Open(file_name_stereo, frequency_hz, "rb");
|
||
|
in_file_stereo_.ReadStereo(true);
|
||
|
in_file_mono_.Open(file_name_mono, frequency_hz, "rb");
|
||
|
in_file_mono_.ReadStereo(false);
|
||
|
|
||
|
// Create Opus encoders for mono and stereo.
|
||
|
ASSERT_GT(WebRtcOpus_EncoderCreate(&opus_mono_encoder_, 1), -1);
|
||
|
ASSERT_GT(WebRtcOpus_EncoderCreate(&opus_stereo_encoder_, 2), -1);
|
||
|
|
||
|
// Create Opus decoders for mono and stereo for stand-alone testing of Opus.
|
||
|
ASSERT_GT(WebRtcOpus_DecoderCreate(&opus_mono_decoder_, 1), -1);
|
||
|
ASSERT_GT(WebRtcOpus_DecoderCreate(&opus_stereo_decoder_, 2), -1);
|
||
|
ASSERT_GT(WebRtcOpus_DecoderInitNew(opus_mono_decoder_), -1);
|
||
|
ASSERT_GT(WebRtcOpus_DecoderInitNew(opus_stereo_decoder_), -1);
|
||
|
|
||
|
ASSERT_TRUE(acm_receiver_.get() != NULL);
|
||
|
EXPECT_EQ(0, acm_receiver_->InitializeReceiver());
|
||
|
|
||
|
// Register Opus stereo as receiving codec.
|
||
|
CodecInst opus_codec_param;
|
||
|
int codec_id = acm_receiver_->Codec("opus", 48000, 2);
|
||
|
EXPECT_EQ(0, acm_receiver_->Codec(codec_id, &opus_codec_param));
|
||
|
payload_type_ = opus_codec_param.pltype;
|
||
|
EXPECT_EQ(0, acm_receiver_->RegisterReceiveCodec(opus_codec_param));
|
||
|
|
||
|
// Create and connect the channel.
|
||
|
channel_a2b_ = new TestPackStereo;
|
||
|
channel_a2b_->RegisterReceiverACM(acm_receiver_.get());
|
||
|
|
||
|
//
|
||
|
// Test Stereo.
|
||
|
//
|
||
|
|
||
|
channel_a2b_->set_codec_mode(kStereo);
|
||
|
audio_channels = 2;
|
||
|
test_cntr++;
|
||
|
OpenOutFile(test_cntr);
|
||
|
|
||
|
// Run Opus with 2.5 ms frame size.
|
||
|
Run(channel_a2b_, audio_channels, 64000, 120);
|
||
|
|
||
|
// Run Opus with 5 ms frame size.
|
||
|
Run(channel_a2b_, audio_channels, 64000, 240);
|
||
|
|
||
|
// Run Opus with 10 ms frame size.
|
||
|
Run(channel_a2b_, audio_channels, 64000, 480);
|
||
|
|
||
|
// Run Opus with 20 ms frame size.
|
||
|
Run(channel_a2b_, audio_channels, 64000, 960);
|
||
|
|
||
|
// Run Opus with 40 ms frame size.
|
||
|
Run(channel_a2b_, audio_channels, 64000, 1920);
|
||
|
|
||
|
// Run Opus with 60 ms frame size.
|
||
|
Run(channel_a2b_, audio_channels, 64000, 2880);
|
||
|
|
||
|
out_file_.Close();
|
||
|
out_file_standalone_.Close();
|
||
|
|
||
|
//
|
||
|
// Test Opus stereo with packet-losses.
|
||
|
//
|
||
|
|
||
|
test_cntr++;
|
||
|
OpenOutFile(test_cntr);
|
||
|
|
||
|
// Run Opus with 20 ms frame size, 1% packet loss.
|
||
|
Run(channel_a2b_, audio_channels, 64000, 960, 1);
|
||
|
|
||
|
// Run Opus with 20 ms frame size, 5% packet loss.
|
||
|
Run(channel_a2b_, audio_channels, 64000, 960, 5);
|
||
|
|
||
|
// Run Opus with 20 ms frame size, 10% packet loss.
|
||
|
Run(channel_a2b_, audio_channels, 64000, 960, 10);
|
||
|
|
||
|
out_file_.Close();
|
||
|
out_file_standalone_.Close();
|
||
|
|
||
|
//
|
||
|
// Test Mono.
|
||
|
//
|
||
|
channel_a2b_->set_codec_mode(kMono);
|
||
|
audio_channels = 1;
|
||
|
test_cntr++;
|
||
|
OpenOutFile(test_cntr);
|
||
|
|
||
|
// Register Opus mono as receiving codec.
|
||
|
opus_codec_param.channels = 1;
|
||
|
EXPECT_EQ(0, acm_receiver_->RegisterReceiveCodec(opus_codec_param));
|
||
|
|
||
|
// Run Opus with 2.5 ms frame size.
|
||
|
Run(channel_a2b_, audio_channels, 32000, 120);
|
||
|
|
||
|
// Run Opus with 5 ms frame size.
|
||
|
Run(channel_a2b_, audio_channels, 32000, 240);
|
||
|
|
||
|
// Run Opus with 10 ms frame size.
|
||
|
Run(channel_a2b_, audio_channels, 32000, 480);
|
||
|
|
||
|
// Run Opus with 20 ms frame size.
|
||
|
Run(channel_a2b_, audio_channels, 32000, 960);
|
||
|
|
||
|
// Run Opus with 40 ms frame size.
|
||
|
Run(channel_a2b_, audio_channels, 32000, 1920);
|
||
|
|
||
|
// Run Opus with 60 ms frame size.
|
||
|
Run(channel_a2b_, audio_channels, 32000, 2880);
|
||
|
|
||
|
out_file_.Close();
|
||
|
out_file_standalone_.Close();
|
||
|
|
||
|
//
|
||
|
// Test Opus mono with packet-losses.
|
||
|
//
|
||
|
test_cntr++;
|
||
|
OpenOutFile(test_cntr);
|
||
|
|
||
|
// Run Opus with 20 ms frame size, 1% packet loss.
|
||
|
Run(channel_a2b_, audio_channels, 64000, 960, 1);
|
||
|
|
||
|
// Run Opus with 20 ms frame size, 5% packet loss.
|
||
|
Run(channel_a2b_, audio_channels, 64000, 960, 5);
|
||
|
|
||
|
// Run Opus with 20 ms frame size, 10% packet loss.
|
||
|
Run(channel_a2b_, audio_channels, 64000, 960, 10);
|
||
|
|
||
|
// Close the files.
|
||
|
in_file_stereo_.Close();
|
||
|
in_file_mono_.Close();
|
||
|
out_file_.Close();
|
||
|
out_file_standalone_.Close();
|
||
|
#endif
|
||
|
}
|
||
|
|
||
|
void OpusTest::Run(TestPackStereo* channel, int channels, int bitrate,
|
||
|
int frame_length, int percent_loss) {
|
||
|
AudioFrame audio_frame;
|
||
|
int32_t out_freq_hz_b = out_file_.SamplingFrequency();
|
||
|
const int kBufferSizeSamples = 480 * 12 * 2; // Can hold 120 ms stereo audio.
|
||
|
int16_t audio[kBufferSizeSamples];
|
||
|
int16_t out_audio[kBufferSizeSamples];
|
||
|
int16_t audio_type;
|
||
|
int written_samples = 0;
|
||
|
int read_samples = 0;
|
||
|
int decoded_samples = 0;
|
||
|
bool first_packet = true;
|
||
|
uint32_t start_time_stamp = 0;
|
||
|
|
||
|
channel->reset_payload_size();
|
||
|
counter_ = 0;
|
||
|
|
||
|
// Set encoder rate.
|
||
|
EXPECT_EQ(0, WebRtcOpus_SetBitRate(opus_mono_encoder_, bitrate));
|
||
|
EXPECT_EQ(0, WebRtcOpus_SetBitRate(opus_stereo_encoder_, bitrate));
|
||
|
|
||
|
#if defined(WEBRTC_ANDROID) || defined(WEBRTC_IOS) || defined(WEBRTC_ARCH_ARM)
|
||
|
// If we are on Android, iOS and/or ARM, use a lower complexity setting as
|
||
|
// default.
|
||
|
const int kOpusComplexity5 = 5;
|
||
|
EXPECT_EQ(0, WebRtcOpus_SetComplexity(opus_mono_encoder_, kOpusComplexity5));
|
||
|
EXPECT_EQ(0, WebRtcOpus_SetComplexity(opus_stereo_encoder_,
|
||
|
kOpusComplexity5));
|
||
|
#endif
|
||
|
|
||
|
// Make sure the runtime is less than 60 seconds to pass Android test.
|
||
|
for (size_t audio_length = 0; audio_length < 10000; audio_length += 10) {
|
||
|
bool lost_packet = false;
|
||
|
|
||
|
// Get 10 msec of audio.
|
||
|
if (channels == 1) {
|
||
|
if (in_file_mono_.EndOfFile()) {
|
||
|
break;
|
||
|
}
|
||
|
in_file_mono_.Read10MsData(audio_frame);
|
||
|
} else {
|
||
|
if (in_file_stereo_.EndOfFile()) {
|
||
|
break;
|
||
|
}
|
||
|
in_file_stereo_.Read10MsData(audio_frame);
|
||
|
}
|
||
|
|
||
|
// If input audio is sampled at 32 kHz, resampling to 48 kHz is required.
|
||
|
EXPECT_EQ(480,
|
||
|
resampler_.Resample10Msec(audio_frame.data_,
|
||
|
audio_frame.sample_rate_hz_,
|
||
|
48000,
|
||
|
channels,
|
||
|
kBufferSizeSamples - written_samples,
|
||
|
&audio[written_samples]));
|
||
|
written_samples += 480 * channels;
|
||
|
|
||
|
// Sometimes we need to loop over the audio vector to produce the right
|
||
|
// number of packets.
|
||
|
int loop_encode = (written_samples - read_samples) /
|
||
|
(channels * frame_length);
|
||
|
|
||
|
if (loop_encode > 0) {
|
||
|
const int kMaxBytes = 1000; // Maximum number of bytes for one packet.
|
||
|
int16_t bitstream_len_byte;
|
||
|
uint8_t bitstream[kMaxBytes];
|
||
|
for (int i = 0; i < loop_encode; i++) {
|
||
|
if (channels == 1) {
|
||
|
bitstream_len_byte = WebRtcOpus_Encode(
|
||
|
opus_mono_encoder_, &audio[read_samples],
|
||
|
frame_length, kMaxBytes, bitstream);
|
||
|
ASSERT_GT(bitstream_len_byte, -1);
|
||
|
} else {
|
||
|
bitstream_len_byte = WebRtcOpus_Encode(
|
||
|
opus_stereo_encoder_, &audio[read_samples],
|
||
|
frame_length, kMaxBytes, bitstream);
|
||
|
ASSERT_GT(bitstream_len_byte, -1);
|
||
|
}
|
||
|
|
||
|
// Simulate packet loss by setting |packet_loss_| to "true" in
|
||
|
// |percent_loss| percent of the loops.
|
||
|
// TODO(tlegrand): Move handling of loss simulation to TestPackStereo.
|
||
|
if (percent_loss > 0) {
|
||
|
if (counter_ == floor((100 / percent_loss) + 0.5)) {
|
||
|
counter_ = 0;
|
||
|
lost_packet = true;
|
||
|
channel->set_lost_packet(true);
|
||
|
} else {
|
||
|
lost_packet = false;
|
||
|
channel->set_lost_packet(false);
|
||
|
}
|
||
|
counter_++;
|
||
|
}
|
||
|
|
||
|
// Run stand-alone Opus decoder, or decode PLC.
|
||
|
if (channels == 1) {
|
||
|
if (!lost_packet) {
|
||
|
decoded_samples += WebRtcOpus_DecodeNew(
|
||
|
opus_mono_decoder_, bitstream, bitstream_len_byte,
|
||
|
&out_audio[decoded_samples * channels], &audio_type);
|
||
|
} else {
|
||
|
decoded_samples += WebRtcOpus_DecodePlc(
|
||
|
opus_mono_decoder_, &out_audio[decoded_samples * channels], 1);
|
||
|
}
|
||
|
} else {
|
||
|
if (!lost_packet) {
|
||
|
decoded_samples += WebRtcOpus_DecodeNew(
|
||
|
opus_stereo_decoder_, bitstream, bitstream_len_byte,
|
||
|
&out_audio[decoded_samples * channels], &audio_type);
|
||
|
} else {
|
||
|
decoded_samples += WebRtcOpus_DecodePlc(
|
||
|
opus_stereo_decoder_, &out_audio[decoded_samples * channels],
|
||
|
1);
|
||
|
}
|
||
|
}
|
||
|
|
||
|
// Send data to the channel. "channel" will handle the loss simulation.
|
||
|
channel->SendData(kAudioFrameSpeech, payload_type_, rtp_timestamp_,
|
||
|
bitstream, bitstream_len_byte, NULL);
|
||
|
if (first_packet) {
|
||
|
first_packet = false;
|
||
|
start_time_stamp = rtp_timestamp_;
|
||
|
}
|
||
|
rtp_timestamp_ += frame_length;
|
||
|
read_samples += frame_length * channels;
|
||
|
}
|
||
|
if (read_samples == written_samples) {
|
||
|
read_samples = 0;
|
||
|
written_samples = 0;
|
||
|
}
|
||
|
}
|
||
|
|
||
|
// Run received side of ACM.
|
||
|
ASSERT_EQ(0, acm_receiver_->PlayoutData10Ms(out_freq_hz_b, &audio_frame));
|
||
|
|
||
|
// Write output speech to file.
|
||
|
out_file_.Write10MsData(
|
||
|
audio_frame.data_,
|
||
|
audio_frame.samples_per_channel_ * audio_frame.num_channels_);
|
||
|
|
||
|
// Write stand-alone speech to file.
|
||
|
out_file_standalone_.Write10MsData(out_audio, decoded_samples * channels);
|
||
|
|
||
|
if (audio_frame.timestamp_ > start_time_stamp) {
|
||
|
// Number of channels should be the same for both stand-alone and
|
||
|
// ACM-decoding.
|
||
|
EXPECT_EQ(audio_frame.num_channels_, channels);
|
||
|
}
|
||
|
|
||
|
decoded_samples = 0;
|
||
|
}
|
||
|
|
||
|
if (in_file_mono_.EndOfFile()) {
|
||
|
in_file_mono_.Rewind();
|
||
|
}
|
||
|
if (in_file_stereo_.EndOfFile()) {
|
||
|
in_file_stereo_.Rewind();
|
||
|
}
|
||
|
// Reset in case we ended with a lost packet.
|
||
|
channel->set_lost_packet(false);
|
||
|
}
|
||
|
|
||
|
void OpusTest::OpenOutFile(int test_number) {
|
||
|
std::string file_name;
|
||
|
std::stringstream file_stream;
|
||
|
file_stream << webrtc::test::OutputPath() << "opustest_out_"
|
||
|
<< test_number << ".pcm";
|
||
|
file_name = file_stream.str();
|
||
|
out_file_.Open(file_name, 48000, "wb");
|
||
|
file_stream.str("");
|
||
|
file_name = file_stream.str();
|
||
|
file_stream << webrtc::test::OutputPath() << "opusstandalone_out_"
|
||
|
<< test_number << ".pcm";
|
||
|
file_name = file_stream.str();
|
||
|
out_file_standalone_.Open(file_name, 48000, "wb");
|
||
|
}
|
||
|
|
||
|
} // namespace webrtc
|