zqz/platform-external-webrtc
1
0
Fork 0
Code Issues Pull Requests Actions Packages Projects Releases Wiki Activity

Convert CNG into C++ and remove it from AudioDecoder

Browse Source 
Broke out CNG from AudioDecoder as they didn't really share an interface.

Converted the CNG code to C++, to make initialization and resource handling easier. This includes several changes to the behavior, favoring RTC_CHECKs over returning error codes.

Review URL: https://codereview.webrtc.org/1868143002

Cr-Commit-Position: refs/heads/master@{#12491}
This commit is contained in:
ossu
2016-04-25 07:55:58 -07:00
committed by Commit bot
parent f55f58d45d
commit 97ba30eedf
24 changed files with 744 additions and 1251 deletions
Download Patch File Download Diff File Expand all files Collapse all files

5
webrtc/modules/audio_coding/codecs/audio_decoder.cc
View File

@ -82,11 +82,6 @@ bool AudioDecoder::PacketHasFec(const uint8_t* encoded,
return false;
}
CNG_dec_inst* AudioDecoder::CngDecoderInstance() {
FATAL() << "Not a CNG decoder";
return NULL;
}
AudioDecoder::SpeechType AudioDecoder::ConvertSpeechType(int16_t type) {
switch (type) {
case 0: // TODO(hlundin): Both iSAC and Opus return 0 for speech.

5
webrtc/modules/audio_coding/codecs/audio_decoder.h
View File

@ -14,7 +14,6 @@
#include <stdlib.h> // NULL
#include "webrtc/base/constructormagic.h"
#include "webrtc/modules/audio_coding/codecs/cng/webrtc_cng.h"
#include "webrtc/typedefs.h"
namespace webrtc {
@ -94,10 +93,6 @@ class AudioDecoder {
// Returns true if the packet has FEC and false otherwise.
virtual bool PacketHasFec(const uint8_t* encoded, size_t encoded_len) const;
// If this is a CNG decoder, return the underlying CNG_dec_inst*. If this
// isn't a CNG decoder, don't call this method.
virtual CNG_dec_inst* CngDecoderInstance();
virtual size_t Channels() const = 0;
protected:

66
webrtc/modules/audio_coding/codecs/cng/audio_encoder_cng.cc
View File

@ -21,19 +21,6 @@ namespace {
const int kMaxFrameSizeMs = 60;
std::unique_ptr<CNG_enc_inst, CngInstDeleter> CreateCngInst(
int sample_rate_hz,
int sid_frame_interval_ms,
int num_cng_coefficients) {
CNG_enc_inst* ci;
RTC_CHECK_EQ(0, WebRtcCng_CreateEnc(&ci));
std::unique_ptr<CNG_enc_inst, CngInstDeleter> cng_inst(ci);
RTC_CHECK_EQ(0,
WebRtcCng_InitEnc(cng_inst.get(), sample_rate_hz,
sid_frame_interval_ms, num_cng_coefficients));
return cng_inst;
}
} // namespace
AudioEncoderCng::Config::Config() = default;
@ -65,9 +52,10 @@ AudioEncoderCng::AudioEncoderCng(Config&& config)
sid_frame_interval_ms_(config.sid_frame_interval_ms),
last_frame_active_(true),
vad_(config.vad ? std::unique_ptr<Vad>(config.vad)
: CreateVad(config.vad_mode)) {
cng_inst_ = CreateCngInst(SampleRateHz(), sid_frame_interval_ms_,
num_cng_coefficients_);
: CreateVad(config.vad_mode)),
cng_encoder_(new ComfortNoiseEncoder(SampleRateHz(),
sid_frame_interval_ms_,
num_cng_coefficients_)) {
}
AudioEncoderCng::~AudioEncoderCng() = default;
@ -170,8 +158,9 @@ void AudioEncoderCng::Reset() {
rtp_timestamps_.clear();
last_frame_active_ = true;
vad_->Reset();
cng_inst_ = CreateCngInst(SampleRateHz(), sid_frame_interval_ms_,
num_cng_coefficients_);
cng_encoder_.reset(
new ComfortNoiseEncoder(SampleRateHz(), sid_frame_interval_ms_,
num_cng_coefficients_));
}
bool AudioEncoderCng::SetFec(bool enable) {
@ -204,32 +193,27 @@ AudioEncoder::EncodedInfo AudioEncoderCng::EncodePassive(
bool force_sid = last_frame_active_;
bool output_produced = false;
const size_t samples_per_10ms_frame = SamplesPer10msFrame();
const size_t bytes_to_encode = frames_to_encode * samples_per_10ms_frame;
AudioEncoder::EncodedInfo info;
encoded->AppendData(bytes_to_encode, [&] (rtc::ArrayView<uint8_t> encoded) {
for (size_t i = 0; i < frames_to_encode; ++i) {
// It's important not to pass &info.encoded_bytes directly to
// WebRtcCng_Encode(), since later loop iterations may return zero in
// that value, in which case we don't want to overwrite any value from
// an earlier iteration.
size_t encoded_bytes_tmp = 0;
RTC_CHECK_GE(
WebRtcCng_Encode(cng_inst_.get(),
&speech_buffer_[i * samples_per_10ms_frame],
samples_per_10ms_frame, encoded.data(),
&encoded_bytes_tmp, force_sid),
0);
if (encoded_bytes_tmp > 0) {
RTC_CHECK(!output_produced);
info.encoded_bytes = encoded_bytes_tmp;
output_produced = true;
force_sid = false;
}
}
for (size_t i = 0; i < frames_to_encode; ++i) {
// It's important not to pass &info.encoded_bytes directly to
// WebRtcCng_Encode(), since later loop iterations may return zero in
// that value, in which case we don't want to overwrite any value from
// an earlier iteration.
size_t encoded_bytes_tmp =
cng_encoder_->Encode(
rtc::ArrayView<const int16_t>(
&speech_buffer_[i * samples_per_10ms_frame],
samples_per_10ms_frame),
force_sid, encoded);
return info.encoded_bytes;
});
if (encoded_bytes_tmp > 0) {
RTC_CHECK(!output_produced);
info.encoded_bytes = encoded_bytes_tmp;
output_produced = true;
force_sid = false;
}
}
info.encoded_timestamp = rtp_timestamps_.front();
info.payload_type = cng_payload_type_;

7
webrtc/modules/audio_coding/codecs/cng/audio_encoder_cng.h
View File

@ -21,11 +21,6 @@
namespace webrtc {
// Deleter for use with unique_ptr.
struct CngInstDeleter {
void operator()(CNG_enc_inst* ptr) const { WebRtcCng_FreeEnc(ptr); }
};
class Vad;
class AudioEncoderCng final : public AudioEncoder {
@ -84,7 +79,7 @@ class AudioEncoderCng final : public AudioEncoder {
std::vector<uint32_t> rtp_timestamps_;
bool last_frame_active_;
std::unique_ptr<Vad> vad_;
std::unique_ptr<CNG_enc_inst, CngInstDeleter> cng_inst_;
std::unique_ptr<ComfortNoiseEncoder> cng_encoder_;
RTC_DISALLOW_COPY_AND_ASSIGN(AudioEncoderCng);
};

4
webrtc/modules/audio_coding/codecs/cng/cng.gypi
View File

@ -18,9 +18,7 @@
'sources': [
'audio_encoder_cng.cc',
'audio_encoder_cng.h',
'cng_helpfuns.c',
'cng_helpfuns.h',
'webrtc_cng.c',
'webrtc_cng.cc',
'webrtc_cng.h',
],
},

48
webrtc/modules/audio_coding/codecs/cng/cng_helpfuns.c
View File

@ -1,48 +0,0 @@
/*
* Copyright (c) 2011 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 "cng_helpfuns.h"
#include "signal_processing_library.h"
#include "webrtc/typedefs.h"
#include "webrtc_cng.h"
/* Values in |k| are Q15, and |a| Q12. */
void WebRtcCng_K2a16(int16_t* k, int useOrder, int16_t* a) {
int16_t any[WEBRTC_SPL_MAX_LPC_ORDER + 1];
int16_t *aptr, *aptr2, *anyptr;
const int16_t *kptr;
int m, i;
kptr = k;
*a = 4096; /* i.e., (Word16_MAX >> 3) + 1 */
*any = *a;
a[1] = (*k + 4) >> 3;
for (m = 1; m < useOrder; m++) {
kptr++;
aptr = a;
aptr++;
aptr2 = &a[m];
anyptr = any;
anyptr++;
any[m + 1] = (*kptr + 4) >> 3;
for (i = 0; i < m; i++) {
*anyptr++ = (*aptr++) +
(int16_t)((((int32_t)(*aptr2--) * (int32_t) * kptr) + 16384) >> 15);
}
aptr = a;
anyptr = any;
for (i = 0; i < (m + 2); i++) {
*aptr++ = *anyptr++;
}
}
}

25
webrtc/modules/audio_coding/codecs/cng/cng_helpfuns.h
View File

@ -1,25 +0,0 @@
/*
* Copyright (c) 2011 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.
*/
#ifndef WEBRTC_MODULES_AUDIO_CODING_CODECS_CNG_CNG_HELPFUNS_H_
#define WEBRTC_MODULES_AUDIO_CODING_CODECS_CNG_CNG_HELPFUNS_H_
#include "webrtc/typedefs.h"
#ifdef __cplusplus
extern "C" {
#endif
void WebRtcCng_K2a16(int16_t* k, int useOrder, int16_t* a);
#ifdef __cplusplus
}
#endif
#endif // WEBRTC_MODULES_AUDIO_CODING_CODECS_CNG_CNG_HELPFUNS_H_

327
webrtc/modules/audio_coding/codecs/cng/cng_unittest.cc
View File

@ -7,11 +7,12 @@
* in the file PATENTS. All contributing project authors may
* be found in the AUTHORS file in the root of the source tree.
*/
#include <memory>
#include <string>
#include "testing/gtest/include/gtest/gtest.h"
#include "webrtc/test/testsupport/fileutils.h"
#include "webrtc_cng.h"
#include "webrtc/modules/audio_coding/codecs/cng/webrtc_cng.h"
namespace webrtc {
@ -21,7 +22,7 @@ enum {
kSidLongIntervalUpdate = 10000
};
enum {
enum : size_t {
kCNGNumParamsLow = 0,
kCNGNumParamsNormal = 8,
kCNGNumParamsHigh = WEBRTC_CNG_MAX_LPC_ORDER,
@ -35,19 +36,13 @@ enum {
class CngTest : public ::testing::Test {
protected:
CngTest();
virtual void SetUp();
CNG_enc_inst* cng_enc_inst_;
CNG_dec_inst* cng_dec_inst_;
void TestCngEncode(int sample_rate_hz, int quality);
int16_t speech_data_[640]; // Max size of CNG internal buffers.
};
CngTest::CngTest()
: cng_enc_inst_(NULL),
cng_dec_inst_(NULL) {
}
void CngTest::SetUp() {
FILE* input_file;
const std::string file_name =
@ -60,289 +55,187 @@ void CngTest::SetUp() {
input_file = NULL;
}
// Test failing Create.
TEST_F(CngTest, CngCreateFail) {
// Test to see that an invalid pointer is caught.
EXPECT_EQ(-1, WebRtcCng_CreateEnc(NULL));
EXPECT_EQ(-1, WebRtcCng_CreateDec(NULL));
}
// Test normal Create.
TEST_F(CngTest, CngCreate) {
EXPECT_EQ(0, WebRtcCng_CreateEnc(&cng_enc_inst_));
EXPECT_EQ(0, WebRtcCng_CreateDec(&cng_dec_inst_));
EXPECT_TRUE(cng_enc_inst_ != NULL);
EXPECT_TRUE(cng_dec_inst_ != NULL);
// Free encoder and decoder memory.
EXPECT_EQ(0, WebRtcCng_FreeEnc(cng_enc_inst_));
EXPECT_EQ(0, WebRtcCng_FreeDec(cng_dec_inst_));
void CngTest::TestCngEncode(int sample_rate_hz, int quality) {
const size_t num_samples_10ms = rtc::CheckedDivExact(sample_rate_hz, 100);
rtc::Buffer sid_data;
ComfortNoiseEncoder cng_encoder(sample_rate_hz, kSidNormalIntervalUpdate,
quality);
EXPECT_EQ(0U, cng_encoder.Encode(rtc::ArrayView<const int16_t>(
speech_data_, num_samples_10ms),
kNoSid, &sid_data));
EXPECT_EQ(static_cast<size_t>(quality + 1),
cng_encoder.Encode(
rtc::ArrayView<const int16_t>(speech_data_, num_samples_10ms),
kForceSid, &sid_data));
}
#if GTEST_HAS_DEATH_TEST && !defined(WEBRTC_ANDROID)
// Create CNG encoder, init with faulty values, free CNG encoder.
TEST_F(CngTest, CngInitFail) {
// Create encoder memory.
EXPECT_EQ(0, WebRtcCng_CreateEnc(&cng_enc_inst_));
// Call with too few parameters.
EXPECT_EQ(-1, WebRtcCng_InitEnc(cng_enc_inst_, 8000, kSidNormalIntervalUpdate,
kCNGNumParamsLow));
EXPECT_EQ(6130, WebRtcCng_GetErrorCodeEnc(cng_enc_inst_));
EXPECT_DEATH({ ComfortNoiseEncoder(8000, kSidNormalIntervalUpdate,
kCNGNumParamsLow); }, "");
// Call with too many parameters.
EXPECT_EQ(-1, WebRtcCng_InitEnc(cng_enc_inst_, 8000, kSidNormalIntervalUpdate,
kCNGNumParamsTooHigh));
EXPECT_EQ(6130, WebRtcCng_GetErrorCodeEnc(cng_enc_inst_));
// Free encoder memory.
EXPECT_EQ(0, WebRtcCng_FreeEnc(cng_enc_inst_));
}
TEST_F(CngTest, CngEncode) {
uint8_t sid_data[WEBRTC_CNG_MAX_LPC_ORDER + 1];
size_t number_bytes;
// Create encoder memory.
EXPECT_EQ(0, WebRtcCng_CreateEnc(&cng_enc_inst_));
// 8 kHz, Normal number of parameters
EXPECT_EQ(0, WebRtcCng_InitEnc(cng_enc_inst_, 8000, kSidNormalIntervalUpdate,
kCNGNumParamsNormal));
EXPECT_EQ(0, WebRtcCng_Encode(cng_enc_inst_, speech_data_, 80, sid_data,
&number_bytes, kNoSid));
EXPECT_EQ(kCNGNumParamsNormal + 1, WebRtcCng_Encode(
cng_enc_inst_, speech_data_, 80, sid_data, &number_bytes, kForceSid));
// 16 kHz, Normal number of parameters
EXPECT_EQ(0, WebRtcCng_InitEnc(cng_enc_inst_, 16000, kSidNormalIntervalUpdate,
kCNGNumParamsNormal));
EXPECT_EQ(0, WebRtcCng_Encode(cng_enc_inst_, speech_data_, 160, sid_data,
&number_bytes, kNoSid));
EXPECT_EQ(kCNGNumParamsNormal + 1, WebRtcCng_Encode(
cng_enc_inst_, speech_data_, 160, sid_data, &number_bytes, kForceSid));
// 32 kHz, Max number of parameters
EXPECT_EQ(0, WebRtcCng_InitEnc(cng_enc_inst_, 32000, kSidNormalIntervalUpdate,
kCNGNumParamsHigh));
EXPECT_EQ(0, WebRtcCng_Encode(cng_enc_inst_, speech_data_, 320, sid_data,
&number_bytes, kNoSid));
EXPECT_EQ(kCNGNumParamsHigh + 1, WebRtcCng_Encode(
cng_enc_inst_, speech_data_, 320, sid_data, &number_bytes, kForceSid));
// 48 kHz, Normal number of parameters
EXPECT_EQ(0, WebRtcCng_InitEnc(cng_enc_inst_, 48000, kSidNormalIntervalUpdate,
kCNGNumParamsNormal));
EXPECT_EQ(0, WebRtcCng_Encode(cng_enc_inst_, speech_data_, 480, sid_data,
&number_bytes, kNoSid));
EXPECT_EQ(kCNGNumParamsNormal + 1, WebRtcCng_Encode(
cng_enc_inst_, speech_data_, 480, sid_data, &number_bytes, kForceSid));
// 64 kHz, Normal number of parameters
EXPECT_EQ(0, WebRtcCng_InitEnc(cng_enc_inst_, 64000, kSidNormalIntervalUpdate,
kCNGNumParamsNormal));
EXPECT_EQ(0, WebRtcCng_Encode(cng_enc_inst_, speech_data_, 640, sid_data,
&number_bytes, kNoSid));
EXPECT_EQ(kCNGNumParamsNormal + 1, WebRtcCng_Encode(
cng_enc_inst_, speech_data_, 640, sid_data, &number_bytes, kForceSid));
// Free encoder memory.
EXPECT_EQ(0, WebRtcCng_FreeEnc(cng_enc_inst_));
EXPECT_DEATH({ ComfortNoiseEncoder(8000, kSidNormalIntervalUpdate,
kCNGNumParamsTooHigh); }, "");
}
// Encode Cng with too long input vector.
TEST_F(CngTest, CngEncodeTooLong) {
uint8_t sid_data[WEBRTC_CNG_MAX_LPC_ORDER + 1];
size_t number_bytes;
// Create and init encoder memory.
EXPECT_EQ(0, WebRtcCng_CreateEnc(&cng_enc_inst_));
EXPECT_EQ(0, WebRtcCng_InitEnc(cng_enc_inst_, 8000, kSidNormalIntervalUpdate,
kCNGNumParamsNormal));
rtc::Buffer sid_data;
// Create encoder.
ComfortNoiseEncoder cng_encoder(8000, kSidNormalIntervalUpdate,
kCNGNumParamsNormal);
// Run encoder with too much data.
EXPECT_EQ(-1, WebRtcCng_Encode(cng_enc_inst_, speech_data_, 641, sid_data,
&number_bytes, kNoSid));
EXPECT_EQ(6140, WebRtcCng_GetErrorCodeEnc(cng_enc_inst_));
EXPECT_DEATH(
cng_encoder.Encode(rtc::ArrayView<const int16_t>(speech_data_, 641),
kNoSid, &sid_data),
"");
}
#endif // GTEST_HAS_DEATH_TEST && !defined(WEBRTC_ANDROID)
// Free encoder memory.
EXPECT_EQ(0, WebRtcCng_FreeEnc(cng_enc_inst_));
TEST_F(CngTest, CngEncode8000) {
TestCngEncode(8000, kCNGNumParamsNormal);
}
// Call encode without calling init.
TEST_F(CngTest, CngEncodeNoInit) {
uint8_t sid_data[WEBRTC_CNG_MAX_LPC_ORDER + 1];
size_t number_bytes;
TEST_F(CngTest, CngEncode16000) {
TestCngEncode(16000, kCNGNumParamsNormal);
}
// Create encoder memory.
EXPECT_EQ(0, WebRtcCng_CreateEnc(&cng_enc_inst_));
TEST_F(CngTest, CngEncode32000) {
TestCngEncode(32000, kCNGNumParamsHigh);
}
// Run encoder without calling init.
EXPECT_EQ(-1, WebRtcCng_Encode(cng_enc_inst_, speech_data_, 640, sid_data,
&number_bytes, kNoSid));
EXPECT_EQ(6120, WebRtcCng_GetErrorCodeEnc(cng_enc_inst_));
TEST_F(CngTest, CngEncode48000) {
TestCngEncode(48000, kCNGNumParamsNormal);
}
// Free encoder memory.
EXPECT_EQ(0, WebRtcCng_FreeEnc(cng_enc_inst_));
TEST_F(CngTest, CngEncode64000) {
TestCngEncode(64000, kCNGNumParamsNormal);
}
// Update SID parameters, for both 9 and 16 parameters.
TEST_F(CngTest, CngUpdateSid) {
uint8_t sid_data[WEBRTC_CNG_MAX_LPC_ORDER + 1];
size_t number_bytes;
rtc::Buffer sid_data;
// Create and initialize encoder and decoder memory.
EXPECT_EQ(0, WebRtcCng_CreateEnc(&cng_enc_inst_));
EXPECT_EQ(0, WebRtcCng_CreateDec(&cng_dec_inst_));
EXPECT_EQ(0, WebRtcCng_InitEnc(cng_enc_inst_, 16000, kSidNormalIntervalUpdate,
kCNGNumParamsNormal));
WebRtcCng_InitDec(cng_dec_inst_);
// Create and initialize encoder and decoder.
ComfortNoiseEncoder cng_encoder(16000, kSidNormalIntervalUpdate,
kCNGNumParamsNormal);
ComfortNoiseDecoder cng_decoder;
// Run normal Encode and UpdateSid.
EXPECT_EQ(kCNGNumParamsNormal + 1, WebRtcCng_Encode(
cng_enc_inst_, speech_data_, 160, sid_data, &number_bytes, kForceSid));
EXPECT_EQ(0, WebRtcCng_UpdateSid(cng_dec_inst_, sid_data,
kCNGNumParamsNormal + 1));
EXPECT_EQ(kCNGNumParamsNormal + 1,
cng_encoder.Encode(rtc::ArrayView<const int16_t>(speech_data_, 160),
kForceSid, &sid_data));
cng_decoder.UpdateSid(sid_data);
// Reinit with new length.
EXPECT_EQ(0, WebRtcCng_InitEnc(cng_enc_inst_, 16000, kSidNormalIntervalUpdate,
kCNGNumParamsHigh));
WebRtcCng_InitDec(cng_dec_inst_);
cng_encoder.Reset(16000, kSidNormalIntervalUpdate, kCNGNumParamsHigh);
cng_decoder.Reset();
// Expect 0 because of unstable parameters after switching length.
EXPECT_EQ(0, WebRtcCng_Encode(cng_enc_inst_, speech_data_, 160, sid_data,
&number_bytes, kForceSid));
EXPECT_EQ(kCNGNumParamsHigh + 1, WebRtcCng_Encode(
cng_enc_inst_, speech_data_ + 160, 160, sid_data, &number_bytes,
kForceSid));
EXPECT_EQ(0, WebRtcCng_UpdateSid(cng_dec_inst_, sid_data,
kCNGNumParamsNormal + 1));
// Free encoder and decoder memory.
EXPECT_EQ(0, WebRtcCng_FreeEnc(cng_enc_inst_));
EXPECT_EQ(0, WebRtcCng_FreeDec(cng_dec_inst_));
EXPECT_EQ(0U,
cng_encoder.Encode(rtc::ArrayView<const int16_t>(speech_data_, 160),
kForceSid, &sid_data));
EXPECT_EQ(
kCNGNumParamsHigh + 1,
cng_encoder.Encode(rtc::ArrayView<const int16_t>(speech_data_ + 160, 160),
kForceSid, &sid_data));
cng_decoder.UpdateSid(
rtc::ArrayView<const uint8_t>(sid_data.data(), kCNGNumParamsNormal + 1));
}
// Update SID parameters, with wrong parameters or without calling decode.
TEST_F(CngTest, CngUpdateSidErroneous) {
uint8_t sid_data[WEBRTC_CNG_MAX_LPC_ORDER + 1];
size_t number_bytes;
// Create encoder and decoder memory.
EXPECT_EQ(0, WebRtcCng_CreateEnc(&cng_enc_inst_));
EXPECT_EQ(0, WebRtcCng_CreateDec(&cng_dec_inst_));
rtc::Buffer sid_data;
// Encode.
EXPECT_EQ(0, WebRtcCng_InitEnc(cng_enc_inst_, 16000, kSidNormalIntervalUpdate,
kCNGNumParamsNormal));
EXPECT_EQ(kCNGNumParamsNormal + 1, WebRtcCng_Encode(
cng_enc_inst_, speech_data_, 160, sid_data, &number_bytes, kForceSid));
// Update Sid before initializing decoder.
EXPECT_EQ(-1, WebRtcCng_UpdateSid(cng_dec_inst_, sid_data,
kCNGNumParamsNormal + 1));
EXPECT_EQ(6220, WebRtcCng_GetErrorCodeDec(cng_dec_inst_));
// Initialize decoder.
WebRtcCng_InitDec(cng_dec_inst_);
ComfortNoiseEncoder cng_encoder(16000, kSidNormalIntervalUpdate,
kCNGNumParamsNormal);
ComfortNoiseDecoder cng_decoder;
EXPECT_EQ(kCNGNumParamsNormal + 1,
cng_encoder.Encode(rtc::ArrayView<const int16_t>(speech_data_, 160),
kForceSid, &sid_data));
// First run with valid parameters, then with too many CNG parameters.
// The function will operate correctly by only reading the maximum number of
// parameters, skipping the extra.
EXPECT_EQ(0, WebRtcCng_UpdateSid(cng_dec_inst_, sid_data,
kCNGNumParamsNormal + 1));
EXPECT_EQ(0, WebRtcCng_UpdateSid(cng_dec_inst_, sid_data,
kCNGNumParamsTooHigh + 1));
EXPECT_EQ(kCNGNumParamsNormal + 1, sid_data.size());
cng_decoder.UpdateSid(sid_data);
// Free encoder and decoder memory.
EXPECT_EQ(0, WebRtcCng_FreeEnc(cng_enc_inst_));
EXPECT_EQ(0, WebRtcCng_FreeDec(cng_dec_inst_));
// Make sure the input buffer is large enough. Since Encode() appends data, we
// need to set the size manually only afterwards, or the buffer will be bigger
// than anticipated.
sid_data.SetSize(kCNGNumParamsTooHigh + 1);
cng_decoder.UpdateSid(sid_data);
}
// Test to generate cng data, by forcing SID. Both normal and faulty condition.
TEST_F(CngTest, CngGenerate) {
uint8_t sid_data[WEBRTC_CNG_MAX_LPC_ORDER + 1];
rtc::Buffer sid_data;
int16_t out_data[640];
size_t number_bytes;
// Create and initialize encoder and decoder memory.
EXPECT_EQ(0, WebRtcCng_CreateEnc(&cng_enc_inst_));
EXPECT_EQ(0, WebRtcCng_CreateDec(&cng_dec_inst_));
EXPECT_EQ(0, WebRtcCng_InitEnc(cng_enc_inst_, 16000, kSidNormalIntervalUpdate,
kCNGNumParamsNormal));
WebRtcCng_InitDec(cng_dec_inst_);
// Create and initialize encoder and decoder.
ComfortNoiseEncoder cng_encoder(16000, kSidNormalIntervalUpdate,
kCNGNumParamsNormal);
ComfortNoiseDecoder cng_decoder;
// Normal Encode.
EXPECT_EQ(kCNGNumParamsNormal + 1, WebRtcCng_Encode(
cng_enc_inst_, speech_data_, 160, sid_data, &number_bytes, kForceSid));
EXPECT_EQ(kCNGNumParamsNormal + 1,
cng_encoder.Encode(rtc::ArrayView<const int16_t>(speech_data_, 160),
kForceSid, &sid_data));
// Normal UpdateSid.
EXPECT_EQ(0, WebRtcCng_UpdateSid(cng_dec_inst_, sid_data,
kCNGNumParamsNormal + 1));
cng_decoder.UpdateSid(sid_data);
// Two normal Generate, one with new_period.
EXPECT_EQ(0, WebRtcCng_Generate(cng_dec_inst_, out_data, 640, 1));
EXPECT_EQ(0, WebRtcCng_Generate(cng_dec_inst_, out_data, 640, 0));
EXPECT_TRUE(cng_decoder.Generate(rtc::ArrayView<int16_t>(out_data, 640), 1));
EXPECT_TRUE(cng_decoder.Generate(rtc::ArrayView<int16_t>(out_data, 640), 0));
// Call Genereate with too much data.
EXPECT_EQ(-1, WebRtcCng_Generate(cng_dec_inst_, out_data, 641, 0));
EXPECT_EQ(6140, WebRtcCng_GetErrorCodeDec(cng_dec_inst_));
// Free encoder and decoder memory.
EXPECT_EQ(0, WebRtcCng_FreeEnc(cng_enc_inst_));
EXPECT_EQ(0, WebRtcCng_FreeDec(cng_dec_inst_));
EXPECT_FALSE(cng_decoder.Generate(rtc::ArrayView<int16_t>(out_data, 641), 0));
}
// Test automatic SID.
TEST_F(CngTest, CngAutoSid) {
uint8_t sid_data[WEBRTC_CNG_MAX_LPC_ORDER + 1];
size_t number_bytes;
rtc::Buffer sid_data;
// Create and initialize encoder and decoder memory.
EXPECT_EQ(0, WebRtcCng_CreateEnc(&cng_enc_inst_));
EXPECT_EQ(0, WebRtcCng_CreateDec(&cng_dec_inst_));
EXPECT_EQ(0, WebRtcCng_InitEnc(cng_enc_inst_, 16000, kSidNormalIntervalUpdate,
kCNGNumParamsNormal));
WebRtcCng_InitDec(cng_dec_inst_);
// Create and initialize encoder and decoder.
ComfortNoiseEncoder cng_encoder(16000, kSidNormalIntervalUpdate,
kCNGNumParamsNormal);
ComfortNoiseDecoder cng_decoder;
// Normal Encode, 100 msec, where no SID data should be generated.
for (int i = 0; i < 10; i++) {
EXPECT_EQ(0, WebRtcCng_Encode(cng_enc_inst_, speech_data_, 160, sid_data,
&number_bytes, kNoSid));
EXPECT_EQ(0U, cng_encoder.Encode(
rtc::ArrayView<const int16_t>(speech_data_, 160), kNoSid, &sid_data));
}
// We have reached 100 msec, and SID data should be generated.
EXPECT_EQ(kCNGNumParamsNormal + 1, WebRtcCng_Encode(
cng_enc_inst_, speech_data_, 160, sid_data, &number_bytes, kNoSid));
// Free encoder and decoder memory.
EXPECT_EQ(0, WebRtcCng_FreeEnc(cng_enc_inst_));
EXPECT_EQ(0, WebRtcCng_FreeDec(cng_dec_inst_));
EXPECT_EQ(kCNGNumParamsNormal + 1, cng_encoder.Encode(
rtc::ArrayView<const int16_t>(speech_data_, 160), kNoSid, &sid_data));
}
// Test automatic SID, with very short interval.
TEST_F(CngTest, CngAutoSidShort) {
uint8_t sid_data[WEBRTC_CNG_MAX_LPC_ORDER + 1];
size_t number_bytes;
rtc::Buffer sid_data;
// Create and initialize encoder and decoder memory.
EXPECT_EQ(0, WebRtcCng_CreateEnc(&cng_enc_inst_));
EXPECT_EQ(0, WebRtcCng_CreateDec(&cng_dec_inst_));
EXPECT_EQ(0, WebRtcCng_InitEnc(cng_enc_inst_, 16000, kSidShortIntervalUpdate,
kCNGNumParamsNormal));
WebRtcCng_InitDec(cng_dec_inst_);
// Create and initialize encoder and decoder.
ComfortNoiseEncoder cng_encoder(16000, kSidShortIntervalUpdate,
kCNGNumParamsNormal);
ComfortNoiseDecoder cng_decoder;
// First call will never generate SID, unless forced to.
EXPECT_EQ(0, WebRtcCng_Encode(cng_enc_inst_, speech_data_, 160, sid_data,
&number_bytes, kNoSid));
EXPECT_EQ(0U, cng_encoder.Encode(
rtc::ArrayView<const int16_t>(speech_data_, 160), kNoSid, &sid_data));
// Normal Encode, 100 msec, SID data should be generated all the time.
for (int i = 0; i < 10; i++) {
EXPECT_EQ(kCNGNumParamsNormal + 1, WebRtcCng_Encode(
cng_enc_inst_, speech_data_, 160, sid_data, &number_bytes, kNoSid));
EXPECT_EQ(kCNGNumParamsNormal + 1, cng_encoder.Encode(
rtc::ArrayView<const int16_t>(speech_data_, 160), kNoSid, &sid_data));
}
// Free encoder and decoder memory.
EXPECT_EQ(0, WebRtcCng_FreeEnc(cng_enc_inst_));
EXPECT_EQ(0, WebRtcCng_FreeDec(cng_dec_inst_));
}
} // namespace webrtc

603
webrtc/modules/audio_coding/codecs/cng/webrtc_cng.c
View File

@ -1,603 +0,0 @@
/*
* 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_cng.h"
#include <string.h>
#include <stdlib.h>
#include "cng_helpfuns.h"
#include "signal_processing_library.h"
typedef struct WebRtcCngDecoder_ {
uint32_t dec_seed;
int32_t dec_target_energy;
int32_t dec_used_energy;
int16_t dec_target_reflCoefs[WEBRTC_CNG_MAX_LPC_ORDER + 1];
int16_t dec_used_reflCoefs[WEBRTC_CNG_MAX_LPC_ORDER + 1];
int16_t dec_filtstate[WEBRTC_CNG_MAX_LPC_ORDER + 1];
int16_t dec_filtstateLow[WEBRTC_CNG_MAX_LPC_ORDER + 1];
int16_t dec_Efiltstate[WEBRTC_CNG_MAX_LPC_ORDER + 1];
int16_t dec_EfiltstateLow[WEBRTC_CNG_MAX_LPC_ORDER + 1];
int16_t dec_order;
int16_t dec_target_scale_factor; /* Q29 */
int16_t dec_used_scale_factor; /* Q29 */
int16_t target_scale_factor; /* Q13 */
int16_t errorcode;
int16_t initflag;
} WebRtcCngDecoder;
typedef struct WebRtcCngEncoder_ {
size_t enc_nrOfCoefs;
int enc_sampfreq;
int16_t enc_interval;
int16_t enc_msSinceSID;
int32_t enc_Energy;
int16_t enc_reflCoefs[WEBRTC_CNG_MAX_LPC_ORDER + 1];
int32_t enc_corrVector[WEBRTC_CNG_MAX_LPC_ORDER + 1];
uint32_t enc_seed;
int16_t errorcode;
int16_t initflag;
} WebRtcCngEncoder;
const int32_t WebRtcCng_kDbov[94] = {
1081109975, 858756178, 682134279, 541838517, 430397633, 341876992,
271562548, 215709799, 171344384, 136103682, 108110997, 85875618,
68213428, 54183852, 43039763, 34187699, 27156255, 21570980,
17134438, 13610368, 10811100, 8587562, 6821343, 5418385,
4303976, 3418770, 2715625, 2157098, 1713444, 1361037,
1081110, 858756, 682134, 541839, 430398, 341877,
271563, 215710, 171344, 136104, 108111, 85876,
68213, 54184, 43040, 34188, 27156, 21571,
17134, 13610, 10811, 8588, 6821, 5418,
4304, 3419, 2716, 2157, 1713, 1361,
1081, 859, 682, 542, 430, 342,
272, 216, 171, 136, 108, 86,
68, 54, 43, 34, 27, 22,
17, 14, 11, 9, 7, 5,
4, 3, 3, 2, 2, 1,
1, 1, 1, 1
};
const int16_t WebRtcCng_kCorrWindow[WEBRTC_CNG_MAX_LPC_ORDER] = {
32702, 32636, 32570, 32505, 32439, 32374,
32309, 32244, 32179, 32114, 32049, 31985
};
/****************************************************************************
* WebRtcCng_CreateEnc/Dec(...)
*
* These functions create an instance to the specified structure
*
* Input:
* - XXX_inst : Pointer to created instance that should be created
*
* Return value : 0 - Ok
* -1 - Error
*/
int16_t WebRtcCng_CreateEnc(CNG_enc_inst** cng_inst) {
if (cng_inst != NULL) {
*cng_inst = (CNG_enc_inst*) malloc(sizeof(WebRtcCngEncoder));
if (*cng_inst != NULL) {
(*(WebRtcCngEncoder**) cng_inst)->errorcode = 0;
(*(WebRtcCngEncoder**) cng_inst)->initflag = 0;
/* Needed to get the right function pointers in SPLIB. */
WebRtcSpl_Init();
return 0;
} else {
/* The memory could not be allocated. */
return -1;
}
} else {
/* The input pointer is invalid (NULL). */
return -1;
}
}
int16_t WebRtcCng_CreateDec(CNG_dec_inst** cng_inst) {
if (cng_inst != NULL ) {
*cng_inst = (CNG_dec_inst*) malloc(sizeof(WebRtcCngDecoder));
if (*cng_inst != NULL ) {
(*(WebRtcCngDecoder**) cng_inst)->errorcode = 0;
(*(WebRtcCngDecoder**) cng_inst)->initflag = 0;
/* Needed to get the right function pointers in SPLIB. */
WebRtcSpl_Init();
return 0;
} else {
/* The memory could not be allocated */
return -1;
}
} else {
/* The input pointer is invalid (NULL). */
return -1;
}
}
/****************************************************************************
* WebRtcCng_InitEnc/Dec(...)
*
* This function initializes a instance
*
* Input:
* - cng_inst : Instance that should be initialized
*
* - fs : 8000 for narrowband and 16000 for wideband
* - interval : generate SID data every interval ms
* - quality : TBD
*
* Output:
* - cng_inst : Initialized instance
*
* Return value : 0 - Ok
* -1 - Error
*/
int WebRtcCng_InitEnc(CNG_enc_inst* cng_inst, int fs, int16_t interval,
int16_t quality) {
int i;
WebRtcCngEncoder* inst = (WebRtcCngEncoder*) cng_inst;
memset(inst, 0, sizeof(WebRtcCngEncoder));
/* Check LPC order */
if (quality > WEBRTC_CNG_MAX_LPC_ORDER || quality <= 0) {
inst->errorcode = CNG_DISALLOWED_LPC_ORDER;
return -1;
}
inst->enc_sampfreq = fs;
inst->enc_interval = interval;
inst->enc_nrOfCoefs = quality;
inst->enc_msSinceSID = 0;
inst->enc_seed = 7777; /* For debugging only. */
inst->enc_Energy = 0;
for (i = 0; i < (WEBRTC_CNG_MAX_LPC_ORDER + 1); i++) {
inst->enc_reflCoefs[i] = 0;
inst->enc_corrVector[i] = 0;
}
inst->initflag = 1;
return 0;
}
void WebRtcCng_InitDec(CNG_dec_inst* cng_inst) {
int i;
WebRtcCngDecoder* inst = (WebRtcCngDecoder*) cng_inst;
memset(inst, 0, sizeof(WebRtcCngDecoder));
inst->dec_seed = 7777; /* For debugging only. */
inst->dec_order = 5;
inst->dec_target_scale_factor = 0;
inst->dec_used_scale_factor = 0;
for (i = 0; i < (WEBRTC_CNG_MAX_LPC_ORDER + 1); i++) {
inst->dec_filtstate[i] = 0;
inst->dec_target_reflCoefs[i] = 0;
inst->dec_used_reflCoefs[i] = 0;
}
inst->dec_target_reflCoefs[0] = 0;
inst->dec_used_reflCoefs[0] = 0;
inst->dec_used_energy = 0;
inst->initflag = 1;
}
/****************************************************************************
* WebRtcCng_FreeEnc/Dec(...)
*
* These functions frees the dynamic memory of a specified instance
*
* Input:
* - cng_inst : Pointer to created instance that should be freed
*
* Return value : 0 - Ok
* -1 - Error
*/
int16_t WebRtcCng_FreeEnc(CNG_enc_inst* cng_inst) {
free(cng_inst);
return 0;
}
int16_t WebRtcCng_FreeDec(CNG_dec_inst* cng_inst) {
free(cng_inst);
return 0;
}
/****************************************************************************
* WebRtcCng_Encode(...)
*
* These functions analyzes background noise
*
* Input:
* - cng_inst : Pointer to created instance
* - speech : Signal (noise) to be analyzed
* - nrOfSamples : Size of speech vector
* - bytesOut : Nr of bytes to transmit, might be 0
*
* Return value : 0 - Ok
* -1 - Error
*/
int WebRtcCng_Encode(CNG_enc_inst* cng_inst, int16_t* speech,
size_t nrOfSamples, uint8_t* SIDdata,
size_t* bytesOut, int16_t forceSID) {
WebRtcCngEncoder* inst = (WebRtcCngEncoder*) cng_inst;
int16_t arCoefs[WEBRTC_CNG_MAX_LPC_ORDER + 1];
int32_t corrVector[WEBRTC_CNG_MAX_LPC_ORDER + 1];
int16_t refCs[WEBRTC_CNG_MAX_LPC_ORDER + 1];
int16_t hanningW[WEBRTC_CNG_MAX_OUTSIZE_ORDER];
int16_t ReflBeta = 19661; /* 0.6 in q15. */
int16_t ReflBetaComp = 13107; /* 0.4 in q15. */
int32_t outEnergy;
int outShifts;
size_t i;
int stab;
int acorrScale;
size_t index;
size_t ind, factor;
int32_t* bptr;
int32_t blo, bhi;
int16_t negate;
const int16_t* aptr;
int16_t speechBuf[WEBRTC_CNG_MAX_OUTSIZE_ORDER];
/* Check if encoder initiated. */
if (inst->initflag != 1) {
inst->errorcode = CNG_ENCODER_NOT_INITIATED;
return -1;
}
/* Check framesize. */
if (nrOfSamples > WEBRTC_CNG_MAX_OUTSIZE_ORDER) {
inst->errorcode = CNG_DISALLOWED_FRAME_SIZE;
return -1;
}
for (i = 0; i < nrOfSamples; i++) {
speechBuf[i] = speech[i];
}
factor = nrOfSamples;
/* Calculate energy and a coefficients. */
outEnergy = WebRtcSpl_Energy(speechBuf, nrOfSamples, &outShifts);
while (outShifts > 0) {
/* We can only do 5 shifts without destroying accuracy in
* division factor. */
if (outShifts > 5) {
outEnergy <<= (outShifts - 5);
outShifts = 5;
} else {
factor /= 2;
outShifts--;
}
}
outEnergy = WebRtcSpl_DivW32W16(outEnergy, (int16_t)factor);
if (outEnergy > 1) {
/* Create Hanning Window. */
WebRtcSpl_GetHanningWindow(hanningW, nrOfSamples / 2);
for (i = 0; i < (nrOfSamples / 2); i++)
hanningW[nrOfSamples - i - 1] = hanningW[i];
WebRtcSpl_ElementwiseVectorMult(speechBuf, hanningW, speechBuf, nrOfSamples,
14);
WebRtcSpl_AutoCorrelation(speechBuf, nrOfSamples, inst->enc_nrOfCoefs,
corrVector, &acorrScale);
if (*corrVector == 0)
*corrVector = WEBRTC_SPL_WORD16_MAX;
/* Adds the bandwidth expansion. */
aptr = WebRtcCng_kCorrWindow;
bptr = corrVector;
/* (zzz) lpc16_1 = 17+1+820+2+2 = 842 (ordo2=700). */
for (ind = 0; ind < inst->enc_nrOfCoefs; ind++) {
/* The below code multiplies the 16 b corrWindow values (Q15) with
* the 32 b corrvector (Q0) and shifts the result down 15 steps. */
negate = *bptr < 0;
if (negate)
*bptr = -*bptr;
blo = (int32_t) * aptr * (*bptr & 0xffff);
bhi = ((blo >> 16) & 0xffff)
+ ((int32_t)(*aptr++) * ((*bptr >> 16) & 0xffff));
blo = (blo & 0xffff) | ((bhi & 0xffff) << 16);
*bptr = (((bhi >> 16) & 0x7fff) << 17) | ((uint32_t) blo >> 15);
if (negate)
*bptr = -*bptr;
bptr++;
}
/* End of bandwidth expansion. */
stab = WebRtcSpl_LevinsonDurbin(corrVector, arCoefs, refCs,
inst->enc_nrOfCoefs);
if (!stab) {
/* Disregard from this frame */
*bytesOut = 0;
return 0;
}
} else {
for (i = 0; i < inst->enc_nrOfCoefs; i++)
refCs[i] = 0;
}
if (forceSID) {
/* Read instantaneous values instead of averaged. */
for (i = 0; i < inst->enc_nrOfCoefs; i++)
inst->enc_reflCoefs[i] = refCs[i];
inst->enc_Energy = outEnergy;
} else {
/* Average history with new values. */
for (i = 0; i < (inst->enc_nrOfCoefs); i++) {
inst->enc_reflCoefs[i] = (int16_t) WEBRTC_SPL_MUL_16_16_RSFT(
inst->enc_reflCoefs[i], ReflBeta, 15);
inst->enc_reflCoefs[i] += (int16_t) WEBRTC_SPL_MUL_16_16_RSFT(
refCs[i], ReflBetaComp, 15);
}
inst->enc_Energy = (outEnergy >> 2) + (inst->enc_Energy >> 1)
+ (inst->enc_Energy >> 2);
}
if (inst->enc_Energy < 1) {
inst->enc_Energy = 1;
}
if ((inst->enc_msSinceSID > (inst->enc_interval - 1)) || forceSID) {
/* Search for best dbov value. */
index = 0;
for (i = 1; i < 93; i++) {
/* Always round downwards. */
if ((inst->enc_Energy - WebRtcCng_kDbov[i]) > 0) {
index = i;
break;
}
}
if ((i == 93) && (index == 0))
index = 94;
SIDdata[0] = (uint8_t)index;
/* Quantize coefficients with tweak for WebRtc implementation of RFC3389. */
if (inst->enc_nrOfCoefs == WEBRTC_CNG_MAX_LPC_ORDER) {
for (i = 0; i < inst->enc_nrOfCoefs; i++) {
/* Q15 to Q7 with rounding. */
SIDdata[i + 1] = ((inst->enc_reflCoefs[i] + 128) >> 8);
}
} else {
for (i = 0; i < inst->enc_nrOfCoefs; i++) {
/* Q15 to Q7 with rounding. */
SIDdata[i + 1] = (127 + ((inst->enc_reflCoefs[i] + 128) >> 8));
}
}
inst->enc_msSinceSID = 0;
*bytesOut = inst->enc_nrOfCoefs + 1;
inst->enc_msSinceSID +=
(int16_t)((1000 * nrOfSamples) / inst->enc_sampfreq);
return (int)(inst->enc_nrOfCoefs + 1);
} else {
inst->enc_msSinceSID +=
(int16_t)((1000 * nrOfSamples) / inst->enc_sampfreq);
*bytesOut = 0;
return 0;
}
}
/****************************************************************************
* WebRtcCng_UpdateSid(...)
*
* These functions updates the CN state, when a new SID packet arrives
*
* Input:
* - cng_inst : Pointer to created instance that should be freed
* - SID : SID packet, all headers removed
* - length : Length in bytes of SID packet
*
* Return value : 0 - Ok
* -1 - Error
*/
int16_t WebRtcCng_UpdateSid(CNG_dec_inst* cng_inst, uint8_t* SID,
size_t length) {
WebRtcCngDecoder* inst = (WebRtcCngDecoder*) cng_inst;
int16_t refCs[WEBRTC_CNG_MAX_LPC_ORDER];
int32_t targetEnergy;
int i;
if (inst->initflag != 1) {
inst->errorcode = CNG_DECODER_NOT_INITIATED;
return -1;
}
/* Throw away reflection coefficients of higher order than we can handle. */
if (length > (WEBRTC_CNG_MAX_LPC_ORDER + 1))
length = WEBRTC_CNG_MAX_LPC_ORDER + 1;
inst->dec_order = (int16_t)length - 1;
if (SID[0] > 93)
SID[0] = 93;
targetEnergy = WebRtcCng_kDbov[SID[0]];
/* Take down target energy to 75%. */
targetEnergy = targetEnergy >> 1;
targetEnergy += targetEnergy >> 2;
inst->dec_target_energy = targetEnergy;
/* Reconstruct coeffs with tweak for WebRtc implementation of RFC3389. */
if (inst->dec_order == WEBRTC_CNG_MAX_LPC_ORDER) {
for (i = 0; i < (inst->dec_order); i++) {
refCs[i] = SID[i + 1] << 8; /* Q7 to Q15*/
inst->dec_target_reflCoefs[i] = refCs[i];
}
} else {
for (i = 0; i < (inst->dec_order); i++) {
refCs[i] = (SID[i + 1] - 127) << 8; /* Q7 to Q15. */
inst->dec_target_reflCoefs[i] = refCs[i];
}
}
for (i = (inst->dec_order); i < WEBRTC_CNG_MAX_LPC_ORDER; i++) {
refCs[i] = 0;
inst->dec_target_reflCoefs[i] = refCs[i];
}
return 0;
}
/****************************************************************************
* WebRtcCng_Generate(...)
*
* These functions generates CN data when needed
*
* Input:
* - cng_inst : Pointer to created instance that should be freed
* - outData : pointer to area to write CN data
* - nrOfSamples : How much data to generate
*
* Return value : 0 - Ok
* -1 - Error
*/
int16_t WebRtcCng_Generate(CNG_dec_inst* cng_inst, int16_t* outData,
size_t nrOfSamples, int16_t new_period) {
WebRtcCngDecoder* inst = (WebRtcCngDecoder*) cng_inst;
size_t i;
int16_t excitation[WEBRTC_CNG_MAX_OUTSIZE_ORDER];
int16_t low[WEBRTC_CNG_MAX_OUTSIZE_ORDER];
int16_t lpPoly[WEBRTC_CNG_MAX_LPC_ORDER + 1];
int16_t ReflBetaStd = 26214; /* 0.8 in q15. */
int16_t ReflBetaCompStd = 6553; /* 0.2 in q15. */
int16_t ReflBetaNewP = 19661; /* 0.6 in q15. */
int16_t ReflBetaCompNewP = 13107; /* 0.4 in q15. */
int16_t Beta, BetaC, tmp1, tmp2, tmp3;
int32_t targetEnergy;
int16_t En;
int16_t temp16;
if (nrOfSamples > WEBRTC_CNG_MAX_OUTSIZE_ORDER) {
inst->errorcode = CNG_DISALLOWED_FRAME_SIZE;
return -1;
}
if (new_period) {
inst->dec_used_scale_factor = inst->dec_target_scale_factor;
Beta = ReflBetaNewP;
BetaC = ReflBetaCompNewP;
} else {
Beta = ReflBetaStd;
BetaC = ReflBetaCompStd;
}
/* Here we use a 0.5 weighting, should possibly be modified to 0.6. */
tmp1 = inst->dec_used_scale_factor << 2; /* Q13->Q15 */
tmp2 = inst->dec_target_scale_factor << 2; /* Q13->Q15 */
tmp3 = (int16_t) WEBRTC_SPL_MUL_16_16_RSFT(tmp1, Beta, 15);
tmp3 += (int16_t) WEBRTC_SPL_MUL_16_16_RSFT(tmp2, BetaC, 15);
inst->dec_used_scale_factor = tmp3 >> 2; /* Q15->Q13 */
inst->dec_used_energy = inst->dec_used_energy >> 1;
inst->dec_used_energy += inst->dec_target_energy >> 1;
/* Do the same for the reflection coeffs. */
for (i = 0; i < WEBRTC_CNG_MAX_LPC_ORDER; i++) {
inst->dec_used_reflCoefs[i] = (int16_t) WEBRTC_SPL_MUL_16_16_RSFT(
inst->dec_used_reflCoefs[i], Beta, 15);
inst->dec_used_reflCoefs[i] += (int16_t) WEBRTC_SPL_MUL_16_16_RSFT(
inst->dec_target_reflCoefs[i], BetaC, 15);
}
/* Compute the polynomial coefficients. */
WebRtcCng_K2a16(inst->dec_used_reflCoefs, WEBRTC_CNG_MAX_LPC_ORDER, lpPoly);
targetEnergy = inst->dec_used_energy;
/* Calculate scaling factor based on filter energy. */
En = 8192; /* 1.0 in Q13. */
for (i = 0; i < (WEBRTC_CNG_MAX_LPC_ORDER); i++) {
/* Floating point value for reference.
E *= 1.0 - (inst->dec_used_reflCoefs[i] / 32768.0) *
(inst->dec_used_reflCoefs[i] / 32768.0);
*/
/* Same in fixed point. */
/* K(i).^2 in Q15. */
temp16 = (int16_t) WEBRTC_SPL_MUL_16_16_RSFT(
inst->dec_used_reflCoefs[i], inst->dec_used_reflCoefs[i], 15);
/* 1 - K(i).^2 in Q15. */
temp16 = 0x7fff - temp16;
En = (int16_t) WEBRTC_SPL_MUL_16_16_RSFT(En, temp16, 15);
}
/* float scaling= sqrt(E * inst->dec_target_energy / (1 << 24)); */
/* Calculate sqrt(En * target_energy / excitation energy) */
targetEnergy = WebRtcSpl_Sqrt(inst->dec_used_energy);
En = (int16_t) WebRtcSpl_Sqrt(En) << 6;
En = (En * 3) >> 1; /* 1.5 estimates sqrt(2). */
inst->dec_used_scale_factor = (int16_t)((En * targetEnergy) >> 12);
/* Generate excitation. */
/* Excitation energy per sample is 2.^24 - Q13 N(0,1). */
for (i = 0; i < nrOfSamples; i++) {
excitation[i] = WebRtcSpl_RandN(&inst->dec_seed) >> 1;
}
/* Scale to correct energy. */
WebRtcSpl_ScaleVector(excitation, excitation, inst->dec_used_scale_factor,
nrOfSamples, 13);
/* |lpPoly| - Coefficients in Q12.
* |excitation| - Speech samples.
* |nst->dec_filtstate| - State preservation.
* |outData| - Filtered speech samples. */
WebRtcSpl_FilterAR(lpPoly, WEBRTC_CNG_MAX_LPC_ORDER + 1, excitation,
nrOfSamples, inst->dec_filtstate, WEBRTC_CNG_MAX_LPC_ORDER,
inst->dec_filtstateLow, WEBRTC_CNG_MAX_LPC_ORDER, outData,
low, nrOfSamples);
return 0;
}
/****************************************************************************
* WebRtcCng_GetErrorCodeEnc/Dec(...)
*
* This functions can be used to check the error code of a CNG instance. When
* a function returns -1 a error code will be set for that instance. The
* function below extract the code of the last error that occured in the
* specified instance.
*
* Input:
* - CNG_inst : CNG enc/dec instance
*
* Return value : Error code
*/
int16_t WebRtcCng_GetErrorCodeEnc(CNG_enc_inst* cng_inst) {
/* Typecast pointer to real structure. */
WebRtcCngEncoder* inst = (WebRtcCngEncoder*) cng_inst;
return inst->errorcode;
}
int16_t WebRtcCng_GetErrorCodeDec(CNG_dec_inst* cng_inst) {
/* Typecast pointer to real structure. */
WebRtcCngDecoder* inst = (WebRtcCngDecoder*) cng_inst;
return inst->errorcode;
}

442
webrtc/modules/audio_coding/codecs/cng/webrtc_cng.cc Normal file
View File

@ -0,0 +1,442 @@
/*
* 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/codecs/cng/webrtc_cng.h"
#include <algorithm>
#include "webrtc/common_audio/signal_processing/include/signal_processing_library.h"
namespace webrtc {
namespace {
const size_t kCngMaxOutsizeOrder = 640;
// TODO(ossu): Rename the left-over WebRtcCng according to style guide.
void WebRtcCng_K2a16(int16_t* k, int useOrder, int16_t* a);
const int32_t WebRtcCng_kDbov[94] = {
1081109975, 858756178, 682134279, 541838517, 430397633, 341876992,
271562548, 215709799, 171344384, 136103682, 108110997, 85875618,
68213428, 54183852, 43039763, 34187699, 27156255, 21570980,
17134438, 13610368, 10811100, 8587562, 6821343, 5418385,
4303976, 3418770, 2715625, 2157098, 1713444, 1361037,
1081110, 858756, 682134, 541839, 430398, 341877,
271563, 215710, 171344, 136104, 108111, 85876,
68213, 54184, 43040, 34188, 27156, 21571,
17134, 13610, 10811, 8588, 6821, 5418,
4304, 3419, 2716, 2157, 1713, 1361,
1081, 859, 682, 542, 430, 342,
272, 216, 171, 136, 108, 86,
68, 54, 43, 34, 27, 22,
17, 14, 11, 9, 7, 5,
4, 3, 3, 2, 2, 1,
1, 1, 1, 1
};
const int16_t WebRtcCng_kCorrWindow[WEBRTC_CNG_MAX_LPC_ORDER] = {
32702, 32636, 32570, 32505, 32439, 32374,
32309, 32244, 32179, 32114, 32049, 31985
};
} // namespace
ComfortNoiseDecoder::ComfortNoiseDecoder() {
/* Needed to get the right function pointers in SPLIB. */
WebRtcSpl_Init();
Reset();
}
void ComfortNoiseDecoder::Reset() {
dec_seed_ = 7777; /* For debugging only. */
dec_target_energy_ = 0;
dec_used_energy_ = 0;
for (auto& c : dec_target_reflCoefs_)
c = 0;
for (auto& c : dec_used_reflCoefs_)
c = 0;
for (auto& c : dec_filtstate_)
c = 0;
for (auto& c : dec_filtstateLow_)
c = 0;
dec_order_ = 5;
dec_target_scale_factor_ = 0;
dec_used_scale_factor_ = 0;
}
void ComfortNoiseDecoder::UpdateSid(rtc::ArrayView<const uint8_t> sid) {
int16_t refCs[WEBRTC_CNG_MAX_LPC_ORDER];
int32_t targetEnergy;
size_t length = sid.size();
/* Throw away reflection coefficients of higher order than we can handle. */
if (length > (WEBRTC_CNG_MAX_LPC_ORDER + 1))
length = WEBRTC_CNG_MAX_LPC_ORDER + 1;
dec_order_ = static_cast<uint16_t>(length - 1);
uint8_t sid0 = std::min<uint8_t>(sid[0], 93);
targetEnergy = WebRtcCng_kDbov[sid0];
/* Take down target energy to 75%. */
targetEnergy = targetEnergy >> 1;
targetEnergy += targetEnergy >> 2;
dec_target_energy_ = targetEnergy;
/* Reconstruct coeffs with tweak for WebRtc implementation of RFC3389. */
if (dec_order_ == WEBRTC_CNG_MAX_LPC_ORDER) {
for (size_t i = 0; i < (dec_order_); i++) {
refCs[i] = sid[i + 1] << 8; /* Q7 to Q15*/
dec_target_reflCoefs_[i] = refCs[i];
}
} else {
for (size_t i = 0; i < (dec_order_); i++) {
refCs[i] = (sid[i + 1] - 127) << 8; /* Q7 to Q15. */
dec_target_reflCoefs_[i] = refCs[i];
}
}
for (size_t i = (dec_order_); i < WEBRTC_CNG_MAX_LPC_ORDER; i++) {
refCs[i] = 0;
dec_target_reflCoefs_[i] = refCs[i];
}
}
bool ComfortNoiseDecoder::Generate(rtc::ArrayView<int16_t> out_data,
bool new_period) {
int16_t excitation[kCngMaxOutsizeOrder];
int16_t low[kCngMaxOutsizeOrder];
int16_t lpPoly[WEBRTC_CNG_MAX_LPC_ORDER + 1];
int16_t ReflBetaStd = 26214; /* 0.8 in q15. */
int16_t ReflBetaCompStd = 6553; /* 0.2 in q15. */
int16_t ReflBetaNewP = 19661; /* 0.6 in q15. */
int16_t ReflBetaCompNewP = 13107; /* 0.4 in q15. */
int16_t Beta, BetaC, tmp1, tmp2, tmp3;
int32_t targetEnergy;
int16_t En;
int16_t temp16;
const size_t num_samples = out_data.size();
if (num_samples > kCngMaxOutsizeOrder) {
return false;
}
if (new_period) {
dec_used_scale_factor_ = dec_target_scale_factor_;
Beta = ReflBetaNewP;
BetaC = ReflBetaCompNewP;
} else {
Beta = ReflBetaStd;
BetaC = ReflBetaCompStd;
}
/* Here we use a 0.5 weighting, should possibly be modified to 0.6. */
tmp1 = dec_used_scale_factor_ << 2; /* Q13->Q15 */
tmp2 = dec_target_scale_factor_ << 2; /* Q13->Q15 */
tmp3 = (int16_t) WEBRTC_SPL_MUL_16_16_RSFT(tmp1, Beta, 15);
tmp3 += (int16_t) WEBRTC_SPL_MUL_16_16_RSFT(tmp2, BetaC, 15);
dec_used_scale_factor_ = tmp3 >> 2; /* Q15->Q13 */
dec_used_energy_ = dec_used_energy_ >> 1;
dec_used_energy_ += dec_target_energy_ >> 1;
/* Do the same for the reflection coeffs. */
for (size_t i = 0; i < WEBRTC_CNG_MAX_LPC_ORDER; i++) {
dec_used_reflCoefs_[i] = (int16_t) WEBRTC_SPL_MUL_16_16_RSFT(
dec_used_reflCoefs_[i], Beta, 15);
dec_used_reflCoefs_[i] += (int16_t) WEBRTC_SPL_MUL_16_16_RSFT(
dec_target_reflCoefs_[i], BetaC, 15);
}
/* Compute the polynomial coefficients. */
WebRtcCng_K2a16(dec_used_reflCoefs_, WEBRTC_CNG_MAX_LPC_ORDER, lpPoly);
targetEnergy = dec_used_energy_;
/* Calculate scaling factor based on filter energy. */
En = 8192; /* 1.0 in Q13. */
for (size_t i = 0; i < (WEBRTC_CNG_MAX_LPC_ORDER); i++) {
/* Floating point value for reference.
E *= 1.0 - (dec_used_reflCoefs_[i] / 32768.0) *
(dec_used_reflCoefs_[i] / 32768.0);
*/
/* Same in fixed point. */
/* K(i).^2 in Q15. */
temp16 = (int16_t) WEBRTC_SPL_MUL_16_16_RSFT(
dec_used_reflCoefs_[i], dec_used_reflCoefs_[i], 15);
/* 1 - K(i).^2 in Q15. */
temp16 = 0x7fff - temp16;
En = (int16_t) WEBRTC_SPL_MUL_16_16_RSFT(En, temp16, 15);
}
/* float scaling= sqrt(E * dec_target_energy_ / (1 << 24)); */
/* Calculate sqrt(En * target_energy / excitation energy) */
targetEnergy = WebRtcSpl_Sqrt(dec_used_energy_);
En = (int16_t) WebRtcSpl_Sqrt(En) << 6;
En = (En * 3) >> 1; /* 1.5 estimates sqrt(2). */
dec_used_scale_factor_ = (int16_t)((En * targetEnergy) >> 12);
/* Generate excitation. */
/* Excitation energy per sample is 2.^24 - Q13 N(0,1). */
for (size_t i = 0; i < num_samples; i++) {
excitation[i] = WebRtcSpl_RandN(&dec_seed_) >> 1;
}
/* Scale to correct energy. */
WebRtcSpl_ScaleVector(excitation, excitation, dec_used_scale_factor_,
num_samples, 13);
/* |lpPoly| - Coefficients in Q12.
* |excitation| - Speech samples.
* |nst->dec_filtstate| - State preservation.
* |out_data| - Filtered speech samples. */
WebRtcSpl_FilterAR(lpPoly, WEBRTC_CNG_MAX_LPC_ORDER + 1, excitation,
num_samples, dec_filtstate_, WEBRTC_CNG_MAX_LPC_ORDER,
dec_filtstateLow_, WEBRTC_CNG_MAX_LPC_ORDER,
out_data.data(), low, num_samples);
return true;
}
ComfortNoiseEncoder::ComfortNoiseEncoder(int fs, int interval, int quality)
: enc_nrOfCoefs_(quality),
enc_sampfreq_(fs),
enc_interval_(interval),
enc_msSinceSid_(0),
enc_Energy_(0),
enc_reflCoefs_{0},
enc_corrVector_{0},
enc_seed_(7777) /* For debugging only. */ {
RTC_CHECK(quality <= WEBRTC_CNG_MAX_LPC_ORDER && quality > 0);
/* Needed to get the right function pointers in SPLIB. */
WebRtcSpl_Init();
}
void ComfortNoiseEncoder::Reset(int fs, int interval, int quality) {
RTC_CHECK(quality <= WEBRTC_CNG_MAX_LPC_ORDER && quality > 0);
enc_nrOfCoefs_ = quality;
enc_sampfreq_ = fs;
enc_interval_ = interval;
enc_msSinceSid_ = 0;
enc_Energy_ = 0;
for (auto& c : enc_reflCoefs_)
c = 0;
for (auto& c : enc_corrVector_)
c = 0;
enc_seed_ = 7777; /* For debugging only. */
}
size_t ComfortNoiseEncoder::Encode(rtc::ArrayView<const int16_t> speech,
bool force_sid,
rtc::Buffer* output) {
int16_t arCoefs[WEBRTC_CNG_MAX_LPC_ORDER + 1];
int32_t corrVector[WEBRTC_CNG_MAX_LPC_ORDER + 1];
int16_t refCs[WEBRTC_CNG_MAX_LPC_ORDER + 1];
int16_t hanningW[kCngMaxOutsizeOrder];
int16_t ReflBeta = 19661; /* 0.6 in q15. */
int16_t ReflBetaComp = 13107; /* 0.4 in q15. */
int32_t outEnergy;
int outShifts;
size_t i;
int stab;
int acorrScale;
size_t index;
size_t ind, factor;
int32_t* bptr;
int32_t blo, bhi;
int16_t negate;
const int16_t* aptr;
int16_t speechBuf[kCngMaxOutsizeOrder];
const size_t num_samples = speech.size();
RTC_CHECK_LE(num_samples, static_cast<size_t>(kCngMaxOutsizeOrder));
for (i = 0; i < num_samples; i++) {
speechBuf[i] = speech[i];
}
factor = num_samples;
/* Calculate energy and a coefficients. */
outEnergy = WebRtcSpl_Energy(speechBuf, num_samples, &outShifts);
while (outShifts > 0) {
/* We can only do 5 shifts without destroying accuracy in
* division factor. */
if (outShifts > 5) {
outEnergy <<= (outShifts - 5);
outShifts = 5;
} else {
factor /= 2;
outShifts--;
}
}
outEnergy = WebRtcSpl_DivW32W16(outEnergy, (int16_t)factor);
if (outEnergy > 1) {
/* Create Hanning Window. */
WebRtcSpl_GetHanningWindow(hanningW, num_samples / 2);
for (i = 0; i < (num_samples / 2); i++)
hanningW[num_samples - i - 1] = hanningW[i];
WebRtcSpl_ElementwiseVectorMult(speechBuf, hanningW, speechBuf, num_samples,
14);
WebRtcSpl_AutoCorrelation(speechBuf, num_samples, enc_nrOfCoefs_,
corrVector, &acorrScale);
if (*corrVector == 0)
*corrVector = WEBRTC_SPL_WORD16_MAX;
/* Adds the bandwidth expansion. */
aptr = WebRtcCng_kCorrWindow;
bptr = corrVector;
/* (zzz) lpc16_1 = 17+1+820+2+2 = 842 (ordo2=700). */
for (ind = 0; ind < enc_nrOfCoefs_; ind++) {
/* The below code multiplies the 16 b corrWindow values (Q15) with
* the 32 b corrvector (Q0) and shifts the result down 15 steps. */
negate = *bptr < 0;
if (negate)
*bptr = -*bptr;
blo = (int32_t) * aptr * (*bptr & 0xffff);
bhi = ((blo >> 16) & 0xffff)
+ ((int32_t)(*aptr++) * ((*bptr >> 16) & 0xffff));
blo = (blo & 0xffff) | ((bhi & 0xffff) << 16);
*bptr = (((bhi >> 16) & 0x7fff) << 17) | ((uint32_t) blo >> 15);
if (negate)
*bptr = -*bptr;
bptr++;
}
/* End of bandwidth expansion. */
stab = WebRtcSpl_LevinsonDurbin(corrVector, arCoefs, refCs,
enc_nrOfCoefs_);
if (!stab) {
/* Disregard from this frame */
return 0;
}
} else {
for (i = 0; i < enc_nrOfCoefs_; i++)
refCs[i] = 0;
}
if (force_sid) {
/* Read instantaneous values instead of averaged. */
for (i = 0; i < enc_nrOfCoefs_; i++)
enc_reflCoefs_[i] = refCs[i];
enc_Energy_ = outEnergy;
} else {
/* Average history with new values. */
for (i = 0; i < enc_nrOfCoefs_; i++) {
enc_reflCoefs_[i] = (int16_t) WEBRTC_SPL_MUL_16_16_RSFT(
enc_reflCoefs_[i], ReflBeta, 15);
enc_reflCoefs_[i] +=
(int16_t) WEBRTC_SPL_MUL_16_16_RSFT(refCs[i], ReflBetaComp, 15);
}
enc_Energy_ =
(outEnergy >> 2) + (enc_Energy_ >> 1) + (enc_Energy_ >> 2);
}
if (enc_Energy_ < 1) {
enc_Energy_ = 1;
}
if ((enc_msSinceSid_ > (enc_interval_ - 1)) || force_sid) {
/* Search for best dbov value. */
index = 0;
for (i = 1; i < 93; i++) {
/* Always round downwards. */
if ((enc_Energy_ - WebRtcCng_kDbov[i]) > 0) {
index = i;
break;
}
}
if ((i == 93) && (index == 0))
index = 94;
const size_t output_coefs = enc_nrOfCoefs_ + 1;
output->AppendData(output_coefs, [&] (rtc::ArrayView<uint8_t> output) {
output[0] = (uint8_t)index;
/* Quantize coefficients with tweak for WebRtc implementation of
* RFC3389. */
if (enc_nrOfCoefs_ == WEBRTC_CNG_MAX_LPC_ORDER) {
for (i = 0; i < enc_nrOfCoefs_; i++) {
/* Q15 to Q7 with rounding. */
output[i + 1] = ((enc_reflCoefs_[i] + 128) >> 8);
}
} else {
for (i = 0; i < enc_nrOfCoefs_; i++) {
/* Q15 to Q7 with rounding. */
output[i + 1] = (127 + ((enc_reflCoefs_[i] + 128) >> 8));
}
}
return output_coefs;
});
enc_msSinceSid_ =
static_cast<int16_t>((1000 * num_samples) / enc_sampfreq_);
return output_coefs;
} else {
enc_msSinceSid_ +=
static_cast<int16_t>((1000 * num_samples) / enc_sampfreq_);
return 0;
}
}
namespace {
/* Values in |k| are Q15, and |a| Q12. */
void WebRtcCng_K2a16(int16_t* k, int useOrder, int16_t* a) {
int16_t any[WEBRTC_SPL_MAX_LPC_ORDER + 1];
int16_t* aptr;
int16_t* aptr2;
int16_t* anyptr;
const int16_t* kptr;
int m, i;
kptr = k;
*a = 4096; /* i.e., (Word16_MAX >> 3) + 1 */
*any = *a;
a[1] = (*k + 4) >> 3;
for (m = 1; m < useOrder; m++) {
kptr++;
aptr = a;
aptr++;
aptr2 = &a[m];
anyptr = any;
anyptr++;
any[m + 1] = (*kptr + 4) >> 3;
for (i = 0; i < m; i++) {
*anyptr++ =
(*aptr++) +
(int16_t)((((int32_t)(*aptr2--) * (int32_t)*kptr) + 16384) >> 15);
}
aptr = a;
anyptr = any;
for (i = 0; i < (m + 2); i++) {
*aptr++ = *anyptr++;
}
}
}
} // namespace
} // namespace webrtc

200
webrtc/modules/audio_coding/codecs/cng/webrtc_cng.h
View File

@ -12,152 +12,88 @@
#ifndef WEBRTC_MODULES_AUDIO_CODING_CODECS_CNG_WEBRTC_CNG_H_
#define WEBRTC_MODULES_AUDIO_CODING_CODECS_CNG_WEBRTC_CNG_H_
#include <stddef.h>
#include <cstddef>
#include "webrtc/base/array_view.h"
#include "webrtc/base/buffer.h"
#include "webrtc/typedefs.h"
#ifdef __cplusplus
extern "C" {
#endif
#define WEBRTC_CNG_MAX_LPC_ORDER 12
#define WEBRTC_CNG_MAX_OUTSIZE_ORDER 640
/* Define Error codes. */
namespace webrtc {
/* 6100 Encoder */
#define CNG_ENCODER_NOT_INITIATED 6120
#define CNG_DISALLOWED_LPC_ORDER 6130
#define CNG_DISALLOWED_FRAME_SIZE 6140
#define CNG_DISALLOWED_SAMPLING_FREQUENCY 6150
/* 6200 Decoder */
#define CNG_DECODER_NOT_INITIATED 6220
class ComfortNoiseDecoder {
public:
ComfortNoiseDecoder();
~ComfortNoiseDecoder() = default;
typedef struct WebRtcCngEncInst CNG_enc_inst;
typedef struct WebRtcCngDecInst CNG_dec_inst;
ComfortNoiseDecoder(const ComfortNoiseDecoder&) = delete;
ComfortNoiseDecoder& operator=(const ComfortNoiseDecoder&) = delete;
/****************************************************************************
* WebRtcCng_CreateEnc/Dec(...)
*
* These functions create an instance to the specified structure
*
* Input:
* - XXX_inst : Pointer to created instance that should be created
*
* Return value : 0 - Ok
* -1 - Error
*/
int16_t WebRtcCng_CreateEnc(CNG_enc_inst** cng_inst);
int16_t WebRtcCng_CreateDec(CNG_dec_inst** cng_inst);
void Reset();
/****************************************************************************
* WebRtcCng_InitEnc/Dec(...)
*
* This function initializes a instance
*
* Input:
* - cng_inst : Instance that should be initialized
*
* - fs : 8000 for narrowband and 16000 for wideband
* - interval : generate SID data every interval ms
* - quality : Number of refl. coefs, maximum allowed is 12
*
* Output:
* - cng_inst : Initialized instance
*
* Return value : 0 - Ok
* -1 - Error
*/
// Updates the CN state when a new SID packet arrives.
// |sid| is a view of the SID packet without the headers.
void UpdateSid(rtc::ArrayView<const uint8_t> sid);
int WebRtcCng_InitEnc(CNG_enc_inst* cng_inst, int fs, int16_t interval,
int16_t quality);
void WebRtcCng_InitDec(CNG_dec_inst* cng_inst);
// Generates comfort noise.
// |out_data| will be filled with samples - its size determines the number of
// samples generated. When |new_period| is true, CNG history will be reset
// before any audio is generated. Returns |false| if outData is too large -
// currently 640 bytes (equalling 10ms at 64kHz).
// TODO(ossu): Specify better limits for the size of out_data. Either let it
// be unbounded or limit to 10ms in the current sample rate.
bool Generate(rtc::ArrayView<int16_t> out_data, bool new_period);
/****************************************************************************
* WebRtcCng_FreeEnc/Dec(...)
*
* These functions frees the dynamic memory of a specified instance
*
* Input:
* - cng_inst : Pointer to created instance that should be freed
*
* Return value : 0 - Ok
* -1 - Error
*/
int16_t WebRtcCng_FreeEnc(CNG_enc_inst* cng_inst);
int16_t WebRtcCng_FreeDec(CNG_dec_inst* cng_inst);
private:
uint32_t dec_seed_;
int32_t dec_target_energy_;
int32_t dec_used_energy_;
int16_t dec_target_reflCoefs_[WEBRTC_CNG_MAX_LPC_ORDER + 1];
int16_t dec_used_reflCoefs_[WEBRTC_CNG_MAX_LPC_ORDER + 1];
int16_t dec_filtstate_[WEBRTC_CNG_MAX_LPC_ORDER + 1];
int16_t dec_filtstateLow_[WEBRTC_CNG_MAX_LPC_ORDER + 1];
uint16_t dec_order_;
int16_t dec_target_scale_factor_; /* Q29 */
int16_t dec_used_scale_factor_; /* Q29 */
};
/****************************************************************************
* WebRtcCng_Encode(...)
*
* These functions analyzes background noise
*
* Input:
* - cng_inst : Pointer to created instance
* - speech : Signal to be analyzed
* - nrOfSamples : Size of speech vector
* - forceSID : not zero to force SID frame and reset
*
* Output:
* - bytesOut : Nr of bytes to transmit, might be 0
*
* Return value : 0 - Ok
* -1 - Error
*/
int WebRtcCng_Encode(CNG_enc_inst* cng_inst, int16_t* speech,
size_t nrOfSamples, uint8_t* SIDdata,
size_t* bytesOut, int16_t forceSID);
class ComfortNoiseEncoder {
public:
// Creates a comfort noise encoder.
// |fs| selects sample rate: 8000 for narrowband or 16000 for wideband.
// |interval| sets the interval at which to generate SID data (in ms).
// |quality| selects the number of refl. coeffs. Maximum allowed is 12.
ComfortNoiseEncoder(int fs, int interval, int quality);
~ComfortNoiseEncoder() = default;
/****************************************************************************
* WebRtcCng_UpdateSid(...)
*
* These functions updates the CN state, when a new SID packet arrives
*
* Input:
* - cng_inst : Pointer to created instance that should be freed
* - SID : SID packet, all headers removed
* - length : Length in bytes of SID packet
*
* Return value : 0 - Ok
* -1 - Error
*/
int16_t WebRtcCng_UpdateSid(CNG_dec_inst* cng_inst, uint8_t* SID,
size_t length);
ComfortNoiseEncoder(const ComfortNoiseEncoder&) = delete;
ComfortNoiseEncoder& operator=(const ComfortNoiseEncoder&) = delete;
/****************************************************************************
* WebRtcCng_Generate(...)
*
* These functions generates CN data when needed
*
* Input:
* - cng_inst : Pointer to created instance that should be freed
* - outData : pointer to area to write CN data
* - nrOfSamples : How much data to generate
* - new_period : >0 if a new period of CNG, will reset history
*
* Return value : 0 - Ok
* -1 - Error
*/
int16_t WebRtcCng_Generate(CNG_dec_inst* cng_inst, int16_t* outData,
size_t nrOfSamples, int16_t new_period);
// Resets the comfort noise encoder to its initial state.
// Parameters are set as during construction.
void Reset(int fs, int interval, int quality);
/*****************************************************************************
* WebRtcCng_GetErrorCodeEnc/Dec(...)
*
* This functions can be used to check the error code of a CNG instance. When
* a function returns -1 a error code will be set for that instance. The
* function below extract the code of the last error that occurred in the
* specified instance.
*
* Input:
* - CNG_inst : CNG enc/dec instance
*
* Return value : Error code
*/
int16_t WebRtcCng_GetErrorCodeEnc(CNG_enc_inst* cng_inst);
int16_t WebRtcCng_GetErrorCodeDec(CNG_dec_inst* cng_inst);
// Analyzes background noise from |speech| and appends coefficients to
// |output|. Returns the number of coefficients generated. If |force_sid| is
// true, a SID frame is forced and the internal sid interval counter is reset.
// Will fail if the input size is too large (> 640 samples, see
// ComfortNoiseDecoder::Generate).
size_t Encode(rtc::ArrayView<const int16_t> speech,
bool force_sid,
rtc::Buffer* output);
#ifdef __cplusplus
}
#endif
private:
size_t enc_nrOfCoefs_;
int enc_sampfreq_;
int16_t enc_interval_;
int16_t enc_msSinceSid_;
int32_t enc_Energy_;
int16_t enc_reflCoefs_[WEBRTC_CNG_MAX_LPC_ORDER + 1];
int32_t enc_corrVector_[WEBRTC_CNG_MAX_LPC_ORDER + 1];
uint32_t enc_seed_;
};
} // namespace webrtc
#endif // WEBRTC_MODULES_AUDIO_CODING_CODECS_CNG_WEBRTC_CNG_H_