amass/FaceAccess
1
0
Fork 0
Code Issues Pull Requests Actions Packages Projects Releases Wiki Activity

add vad code.

Browse Source
This commit is contained in:
luocai
2024-09-06 18:26:45 +08:00
parent 35bf68338f
commit 2bed1dacf2
93 changed files with 12362 additions and 2 deletions
Download Patch File Download Diff File Expand all files Collapse all files

235
VocieProcess/api/audio/audio_frame.cc Normal file
View File

@ -0,0 +1,235 @@
/*
* Copyright (c) 2018 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 "api/audio/audio_frame.h"
#include <string.h>
#include <cstdint>
#include "absl/types/optional.h"
#include "api/array_view.h"
#include "api/audio/audio_view.h"
#include "api/audio/channel_layout.h"
#include "api/rtp_packet_infos.h"
#include "rtc_base/checks.h"
#include "rtc_base/time_utils.h"
namespace webrtc {
AudioFrame::AudioFrame() {
// Visual Studio doesn't like this in the class definition.
static_assert(sizeof(data_) == kMaxDataSizeBytes, "kMaxDataSizeBytes");
}
AudioFrame::AudioFrame(int sample_rate_hz,
size_t num_channels,
ChannelLayout layout /*= CHANNEL_LAYOUT_UNSUPPORTED*/)
: samples_per_channel_(SampleRateToDefaultChannelSize(sample_rate_hz)),
sample_rate_hz_(sample_rate_hz),
num_channels_(num_channels),
channel_layout_(layout == CHANNEL_LAYOUT_UNSUPPORTED
? GuessChannelLayout(num_channels)
: layout) {
RTC_DCHECK_LE(num_channels_, kMaxConcurrentChannels);
RTC_DCHECK_GT(sample_rate_hz_, 0);
RTC_DCHECK_GT(samples_per_channel_, 0u);
}
void AudioFrame::Reset() {
ResetWithoutMuting();
muted_ = true;
}
void AudioFrame::ResetWithoutMuting() {
// TODO(wu): Zero is a valid value for `timestamp_`. We should initialize
// to an invalid value, or add a new member to indicate invalidity.
timestamp_ = 0;
elapsed_time_ms_ = -1;
ntp_time_ms_ = -1;
samples_per_channel_ = 0;
sample_rate_hz_ = 0;
num_channels_ = 0;
channel_layout_ = CHANNEL_LAYOUT_NONE;
speech_type_ = kUndefined;
vad_activity_ = kVadUnknown;
profile_timestamp_ms_ = 0;
packet_infos_ = RtpPacketInfos();
absolute_capture_timestamp_ms_ = absl::nullopt;
}
void AudioFrame::UpdateFrame(uint32_t timestamp,
const int16_t* data,
size_t samples_per_channel,
int sample_rate_hz,
SpeechType speech_type,
VADActivity vad_activity,
size_t num_channels) {
RTC_CHECK_LE(num_channels, kMaxConcurrentChannels);
timestamp_ = timestamp;
samples_per_channel_ = samples_per_channel;
sample_rate_hz_ = sample_rate_hz;
speech_type_ = speech_type;
vad_activity_ = vad_activity;
num_channels_ = num_channels;
channel_layout_ = GuessChannelLayout(num_channels);
if (channel_layout_ != CHANNEL_LAYOUT_UNSUPPORTED) {
RTC_DCHECK_EQ(num_channels, ChannelLayoutToChannelCount(channel_layout_));
}
const size_t length = samples_per_channel * num_channels;
RTC_CHECK_LE(length, data_.size());
if (data != nullptr) {
memcpy(data_.data(), data, sizeof(int16_t) * length);
muted_ = false;
} else {
muted_ = true;
}
}
void AudioFrame::CopyFrom(const AudioFrame& src) {
if (this == &src)
return;
if (muted_ && !src.muted()) {
// TODO: bugs.webrtc.org/5647 - Since the default value for `muted_` is
// false and `data_` may still be uninitialized (because we don't initialize
// data_ as part of construction), we clear the full buffer here before
// copying over new values. If we don't, msan might complain in some tests.
// Consider locking down construction, avoiding the default constructor and
// prefering construction that initializes all state.
ClearSamples(data_);
}
timestamp_ = src.timestamp_;
elapsed_time_ms_ = src.elapsed_time_ms_;
ntp_time_ms_ = src.ntp_time_ms_;
packet_infos_ = src.packet_infos_;
muted_ = src.muted();
samples_per_channel_ = src.samples_per_channel_;
sample_rate_hz_ = src.sample_rate_hz_;
speech_type_ = src.speech_type_;
vad_activity_ = src.vad_activity_;
num_channels_ = src.num_channels_;
channel_layout_ = src.channel_layout_;
absolute_capture_timestamp_ms_ = src.absolute_capture_timestamp_ms();
auto data = src.data_view();
RTC_CHECK_LE(data.size(), data_.size());
if (!muted_ && !data.empty()) {
memcpy(&data_[0], &data[0], sizeof(int16_t) * data.size());
}
}
void AudioFrame::UpdateProfileTimeStamp() {
profile_timestamp_ms_ = rtc::TimeMillis();
}
int64_t AudioFrame::ElapsedProfileTimeMs() const {
if (profile_timestamp_ms_ == 0) {
// Profiling has not been activated.
return -1;
}
return rtc::TimeSince(profile_timestamp_ms_);
}
const int16_t* AudioFrame::data() const {
return muted_ ? zeroed_data().begin() : data_.data();
}
InterleavedView<const int16_t> AudioFrame::data_view() const {
// If you get a nullptr from `data_view()`, it's likely because the
// samples_per_channel_ and/or num_channels_ members haven't been properly
// set. Since `data_view()` returns an InterleavedView<> (which internally
// uses rtc::ArrayView<>), we inherit the behavior in InterleavedView when the
// view size is 0 that ArrayView<>::data() returns nullptr. So, even when an
// AudioFrame is muted and we want to return `zeroed_data()`, if
// samples_per_channel_ or num_channels_ is 0, the view will point to
// nullptr.
return InterleavedView<const int16_t>(muted_ ? &zeroed_data()[0] : &data_[0],
samples_per_channel_, num_channels_);
}
int16_t* AudioFrame::mutable_data() {
// TODO: bugs.webrtc.org/5647 - Can we skip zeroing the buffer?
// Consider instead if we should rather zero the buffer when `muted_` is set
// to `true`.
if (muted_) {
ClearSamples(data_);
muted_ = false;
}
return &data_[0];
}
InterleavedView<int16_t> AudioFrame::mutable_data(size_t samples_per_channel,
size_t num_channels) {
const size_t total_samples = samples_per_channel * num_channels;
RTC_CHECK_LE(total_samples, data_.size());
RTC_CHECK_LE(num_channels, kMaxConcurrentChannels);
// Sanity check for valid argument values during development.
// If `samples_per_channel` is < `num_channels` but larger than 0,
// then chances are the order of arguments is incorrect.
RTC_DCHECK((samples_per_channel == 0 && num_channels == 0) ||
num_channels <= samples_per_channel)
<< "samples_per_channel=" << samples_per_channel
<< "num_channels=" << num_channels;
// TODO: bugs.webrtc.org/5647 - Can we skip zeroing the buffer?
// Consider instead if we should rather zero the whole buffer when `muted_` is
// set to `true`.
if (muted_) {
ClearSamples(data_, total_samples);
muted_ = false;
}
samples_per_channel_ = samples_per_channel;
num_channels_ = num_channels;
return InterleavedView<int16_t>(&data_[0], samples_per_channel, num_channels);
}
void AudioFrame::Mute() {
muted_ = true;
}
bool AudioFrame::muted() const {
return muted_;
}
void AudioFrame::SetLayoutAndNumChannels(ChannelLayout layout,
size_t num_channels) {
channel_layout_ = layout;
num_channels_ = num_channels;
#if RTC_DCHECK_IS_ON
// Do a sanity check that the layout and num_channels match.
// If this lookup yield 0u, then the layout is likely CHANNEL_LAYOUT_DISCRETE.
auto expected_num_channels = ChannelLayoutToChannelCount(layout);
if (expected_num_channels) { // If expected_num_channels is 0
RTC_DCHECK_EQ(expected_num_channels, num_channels_);
}
#endif
RTC_CHECK_LE(samples_per_channel_ * num_channels_, data_.size());
}
void AudioFrame::SetSampleRateAndChannelSize(int sample_rate) {
sample_rate_hz_ = sample_rate;
// We could call `AudioProcessing::GetFrameSize()` here, but that requires
// adding a dependency on the ":audio_processing" build target, which can
// complicate the dependency tree. Some refactoring is probably in order to
// get some consistency around this since there are many places across the
// code that assume this default buffer size.
samples_per_channel_ = SampleRateToDefaultChannelSize(sample_rate_hz_);
}
// static
rtc::ArrayView<const int16_t> AudioFrame::zeroed_data() {
static int16_t* null_data = new int16_t[kMaxDataSizeSamples]();
return rtc::ArrayView<const int16_t>(null_data, kMaxDataSizeSamples);
}
} // namespace webrtc

230
VocieProcess/api/audio/audio_frame.h Normal file
View File

@ -0,0 +1,230 @@
/*
* Copyright (c) 2018 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 API_AUDIO_AUDIO_FRAME_H_
#define API_AUDIO_AUDIO_FRAME_H_
#include <stddef.h>
#include <stdint.h>
#include <array>
#include "absl/types/optional.h"
#include "api/array_view.h"
#include "api/audio/audio_view.h"
#include "api/audio/channel_layout.h"
#include "api/rtp_packet_infos.h"
#include "rtc_base/checks.h"
namespace webrtc {
// Default webrtc buffer size in milliseconds.
constexpr size_t kDefaultAudioBufferLengthMs = 10u;
// Default total number of audio buffers per second based on the default length.
constexpr size_t kDefaultAudioBuffersPerSec =
1000u / kDefaultAudioBufferLengthMs;
// Returns the number of samples a buffer needs to hold for ~10ms of a single
// audio channel at a given sample rate.
// See also `AudioProcessing::GetFrameSize()`.
inline size_t SampleRateToDefaultChannelSize(size_t sample_rate) {
// Basic sanity check. 192kHz is the highest supported input sample rate.
RTC_DCHECK_LE(sample_rate, 192000);
return sample_rate / kDefaultAudioBuffersPerSec;
}
/////////////////////////////////////////////////////////////////////
/* This class holds up to 120 ms of super-wideband (32 kHz) stereo audio. It
* allows for adding and subtracting frames while keeping track of the resulting
* states.
*
* Notes
* - This is a de-facto api, not designed for external use. The AudioFrame class
* is in need of overhaul or even replacement, and anyone depending on it
* should be prepared for that.
* - The total number of samples is samples_per_channel_ * num_channels_.
* - Stereo data is interleaved starting with the left channel.
*/
class AudioFrame {
public:
// Using constexpr here causes linker errors unless the variable also has an
// out-of-class definition, which is impractical in this header-only class.
// (This makes no sense because it compiles as an enum value, which we most
// certainly cannot take the address of, just fine.) C++17 introduces inline
// variables which should allow us to switch to constexpr and keep this a
// header-only class.
enum : size_t {
// Stereo, 32 kHz, 120 ms (2 * 32 * 120)
// Stereo, 192 kHz, 20 ms (2 * 192 * 20)
kMaxDataSizeSamples = 7680,
kMaxDataSizeBytes = kMaxDataSizeSamples * sizeof(int16_t),
};
enum VADActivity { kVadActive = 0, kVadPassive = 1, kVadUnknown = 2 };
enum SpeechType {
kNormalSpeech = 0,
kPLC = 1,
kCNG = 2,
kPLCCNG = 3,
kCodecPLC = 5,
kUndefined = 4
};
AudioFrame();
// Construct an audio frame with frame length properties and channel
// information. `samples_per_channel()` will be initialized to a 10ms buffer
// size and if `layout` is not specified (default value of
// CHANNEL_LAYOUT_UNSUPPORTED is set), then the channel layout is derived
// (guessed) from `num_channels`.
AudioFrame(int sample_rate_hz,
size_t num_channels,
ChannelLayout layout = CHANNEL_LAYOUT_UNSUPPORTED);
AudioFrame(const AudioFrame&) = delete;
AudioFrame& operator=(const AudioFrame&) = delete;
// Resets all members to their default state.
void Reset();
// Same as Reset(), but leaves mute state unchanged. Muting a frame requires
// the buffer to be zeroed on the next call to mutable_data(). Callers
// intending to write to the buffer immediately after Reset() can instead use
// ResetWithoutMuting() to skip this wasteful zeroing.
void ResetWithoutMuting();
// TODO: b/335805780 - Accept InterleavedView.
void UpdateFrame(uint32_t timestamp,
const int16_t* data,
size_t samples_per_channel,
int sample_rate_hz,
SpeechType speech_type,
VADActivity vad_activity,
size_t num_channels = 1);
void CopyFrom(const AudioFrame& src);
// Sets a wall-time clock timestamp in milliseconds to be used for profiling
// of time between two points in the audio chain.
// Example:
// t0: UpdateProfileTimeStamp()
// t1: ElapsedProfileTimeMs() => t1 - t0 [msec]
void UpdateProfileTimeStamp();
// Returns the time difference between now and when UpdateProfileTimeStamp()
// was last called. Returns -1 if UpdateProfileTimeStamp() has not yet been
// called.
int64_t ElapsedProfileTimeMs() const;
// data() returns a zeroed static buffer if the frame is muted.
// TODO: b/335805780 - Return InterleavedView.
const int16_t* data() const;
// Returns a read-only view of all the valid samples held by the AudioFrame.
// For a muted AudioFrame, the samples will all be 0.
InterleavedView<const int16_t> data_view() const;
// mutable_frame() always returns a non-static buffer; the first call to
// mutable_frame() zeros the buffer and marks the frame as unmuted.
// TODO: b/335805780 - Return an InterleavedView.
int16_t* mutable_data();
// Grants write access to the audio buffer. The size of the returned writable
// view is determined by the `samples_per_channel` and `num_channels`
// dimensions which the function checks for correctness and stores in the
// internal member variables; `samples_per_channel()` and `num_channels()`
// respectively.
// If the state is currently muted, the returned view will be zeroed out.
InterleavedView<int16_t> mutable_data(size_t samples_per_channel,
size_t num_channels);
// Prefer to mute frames using AudioFrameOperations::Mute.
void Mute();
// Frame is muted by default.
bool muted() const;
size_t max_16bit_samples() const { return data_.size(); }
size_t samples_per_channel() const { return samples_per_channel_; }
size_t num_channels() const { return num_channels_; }
ChannelLayout channel_layout() const { return channel_layout_; }
// Sets the `channel_layout` property as well as `num_channels`.
void SetLayoutAndNumChannels(ChannelLayout layout, size_t num_channels);
int sample_rate_hz() const { return sample_rate_hz_; }
void set_absolute_capture_timestamp_ms(
int64_t absolute_capture_time_stamp_ms) {
absolute_capture_timestamp_ms_ = absolute_capture_time_stamp_ms;
}
absl::optional<int64_t> absolute_capture_timestamp_ms() const {
return absolute_capture_timestamp_ms_;
}
// Sets the sample_rate_hz and samples_per_channel properties based on a
// given sample rate and calculates a default 10ms samples_per_channel value.
void SetSampleRateAndChannelSize(int sample_rate);
// RTP timestamp of the first sample in the AudioFrame.
uint32_t timestamp_ = 0;
// Time since the first frame in milliseconds.
// -1 represents an uninitialized value.
int64_t elapsed_time_ms_ = -1;
// NTP time of the estimated capture time in local timebase in milliseconds.
// -1 represents an uninitialized value.
int64_t ntp_time_ms_ = -1;
size_t samples_per_channel_ = 0;
int sample_rate_hz_ = 0;
size_t num_channels_ = 0;
SpeechType speech_type_ = kUndefined;
VADActivity vad_activity_ = kVadUnknown;
// Monotonically increasing timestamp intended for profiling of audio frames.
// Typically used for measuring elapsed time between two different points in
// the audio path. No lock is used to save resources and we are thread safe
// by design.
// TODO(nisse@webrtc.org): consider using absl::optional.
int64_t profile_timestamp_ms_ = 0;
// Information about packets used to assemble this audio frame. This is needed
// by `SourceTracker` when the frame is delivered to the RTCRtpReceiver's
// MediaStreamTrack, in order to implement getContributingSources(). See:
// https://w3c.github.io/webrtc-pc/#dom-rtcrtpreceiver-getcontributingsources
//
// TODO(bugs.webrtc.org/10757):
// Note that this information might not be fully accurate since we currently
// don't have a proper way to track it across the audio sync buffer. The
// sync buffer is the small sample-holding buffer located after the audio
// decoder and before where samples are assembled into output frames.
//
// `RtpPacketInfos` may also be empty if the audio samples did not come from
// RTP packets. E.g. if the audio were locally generated by packet loss
// concealment, comfort noise generation, etc.
RtpPacketInfos packet_infos_;
private:
// A permanently zeroed out buffer to represent muted frames. This is a
// header-only class, so the only way to avoid creating a separate zeroed
// buffer per translation unit is to wrap a static in an inline function.
static rtc::ArrayView<const int16_t> zeroed_data();
std::array<int16_t, kMaxDataSizeSamples> data_;
bool muted_ = true;
ChannelLayout channel_layout_ = CHANNEL_LAYOUT_NONE;
// Absolute capture timestamp when this audio frame was originally captured.
// This is only valid for audio frames captured on this machine. The absolute
// capture timestamp of a received frame is found in `packet_infos_`.
// This timestamp MUST be based on the same clock as rtc::TimeMillis().
absl::optional<int64_t> absolute_capture_timestamp_ms_;
};
} // namespace webrtc
#endif // API_AUDIO_AUDIO_FRAME_H_