Mirrors/overte
3
0
Fork 0
mirror of https://github.com/overte-org/overte.git synced 2025-04-21 08:04:01 +02:00
Code Issues Projects Releases Packages Wiki Activity Actions 2

Merge branch 'master' of https://github.com/highfidelity/hifi into metavoxels

Browse source
This commit is contained in:
Andrzej Kapolka 2014-06-06 11:43:12 -07:00
commit f78187c256
12 changed files with 357 additions and 261 deletions
Show stats Download patch file Download diff file Expand all files Collapse all files

242
assignment-client/src/audio/AudioMixer.cpp
View file

@ -173,134 +173,158 @@ void AudioMixer::addBufferToMixForListeningNodeWithBuffer(PositionalAudioRingBuf
weakChannelAmplitudeRatio = 1 - (PHASE_AMPLITUDE_RATIO_AT_90 * sinRatio);
}
}
// if the bearing relative angle to source is > 0 then the delayed channel is the right one
int delayedChannelOffset = (bearingRelativeAngleToSource > 0.0f) ? 1 : 0;
int goodChannelOffset = delayedChannelOffset == 0 ? 1 : 0;
const int16_t* nextOutputStart = bufferToAdd->getNextOutput();
const int16_t* bufferStart = bufferToAdd->getBuffer();
int ringBufferSampleCapacity = bufferToAdd->getSampleCapacity();
int16_t correctBufferSample[2], delayBufferSample[2];
int delayedChannelIndex = 0;
const int SINGLE_STEREO_OFFSET = 2;
for (int s = 0; s < NETWORK_BUFFER_LENGTH_SAMPLES_STEREO; s += 4) {
if (!bufferToAdd->isStereo()) {
// this is a mono buffer, which means it gets full attenuation and spatialization
// setup the int16_t variables for the two sample sets
correctBufferSample[0] = nextOutputStart[s / 2] * attenuationCoefficient;
correctBufferSample[1] = nextOutputStart[(s / 2) + 1] * attenuationCoefficient;
// if the bearing relative angle to source is > 0 then the delayed channel is the right one
int delayedChannelOffset = (bearingRelativeAngleToSource > 0.0f) ? 1 : 0;
int goodChannelOffset = delayedChannelOffset == 0 ? 1 : 0;
delayedChannelIndex = s + (numSamplesDelay * 2) + delayedChannelOffset;
const int16_t* bufferStart = bufferToAdd->getBuffer();
int ringBufferSampleCapacity = bufferToAdd->getSampleCapacity();
delayBufferSample[0] = correctBufferSample[0] * weakChannelAmplitudeRatio;
delayBufferSample[1] = correctBufferSample[1] * weakChannelAmplitudeRatio;
int16_t correctBufferSample[2], delayBufferSample[2];
int delayedChannelIndex = 0;
__m64 bufferSamples = _mm_set_pi16(_clientSamples[s + goodChannelOffset],
_clientSamples[s + goodChannelOffset + SINGLE_STEREO_OFFSET],
_clientSamples[delayedChannelIndex],
_clientSamples[delayedChannelIndex + SINGLE_STEREO_OFFSET]);
__m64 addedSamples = _mm_set_pi16(correctBufferSample[0], correctBufferSample[1],
delayBufferSample[0], delayBufferSample[1]);
const int SINGLE_STEREO_OFFSET = 2;
// perform the MMX add (with saturation) of two correct and delayed samples
__m64 mmxResult = _mm_adds_pi16(bufferSamples, addedSamples);
int16_t* shortResults = reinterpret_cast<int16_t*>(&mmxResult);
// assign the results from the result of the mmx arithmetic
_clientSamples[s + goodChannelOffset] = shortResults[3];
_clientSamples[s + goodChannelOffset + SINGLE_STEREO_OFFSET] = shortResults[2];
_clientSamples[delayedChannelIndex] = shortResults[1];
_clientSamples[delayedChannelIndex + SINGLE_STEREO_OFFSET] = shortResults[0];
}
// The following code is pretty gross and redundant, but AFAIK it's the best way to avoid
// too many conditionals in handling the delay samples at the beginning of _clientSamples.
// Basically we try to take the samples in batches of four, and then handle the remainder
// conditionally to get rid of the rest.
const int DOUBLE_STEREO_OFFSET = 4;
const int TRIPLE_STEREO_OFFSET = 6;
if (numSamplesDelay > 0) {
// if there was a sample delay for this buffer, we need to pull samples prior to the nextOutput
// to stick at the beginning
float attenuationAndWeakChannelRatio = attenuationCoefficient * weakChannelAmplitudeRatio;
const int16_t* delayNextOutputStart = nextOutputStart - numSamplesDelay;
if (delayNextOutputStart < bufferStart) {
delayNextOutputStart = bufferStart + ringBufferSampleCapacity - numSamplesDelay;
for (int s = 0; s < NETWORK_BUFFER_LENGTH_SAMPLES_STEREO; s += 4) {
// setup the int16_t variables for the two sample sets
correctBufferSample[0] = nextOutputStart[s / 2] * attenuationCoefficient;
correctBufferSample[1] = nextOutputStart[(s / 2) + 1] * attenuationCoefficient;
delayedChannelIndex = s + (numSamplesDelay * 2) + delayedChannelOffset;
delayBufferSample[0] = correctBufferSample[0] * weakChannelAmplitudeRatio;
delayBufferSample[1] = correctBufferSample[1] * weakChannelAmplitudeRatio;
__m64 bufferSamples = _mm_set_pi16(_clientSamples[s + goodChannelOffset],
_clientSamples[s + goodChannelOffset + SINGLE_STEREO_OFFSET],
_clientSamples[delayedChannelIndex],
_clientSamples[delayedChannelIndex + SINGLE_STEREO_OFFSET]);
__m64 addedSamples = _mm_set_pi16(correctBufferSample[0], correctBufferSample[1],
delayBufferSample[0], delayBufferSample[1]);
// perform the MMX add (with saturation) of two correct and delayed samples
__m64 mmxResult = _mm_adds_pi16(bufferSamples, addedSamples);
int16_t* shortResults = reinterpret_cast<int16_t*>(&mmxResult);
// assign the results from the result of the mmx arithmetic
_clientSamples[s + goodChannelOffset] = shortResults[3];
_clientSamples[s + goodChannelOffset + SINGLE_STEREO_OFFSET] = shortResults[2];
_clientSamples[delayedChannelIndex] = shortResults[1];
_clientSamples[delayedChannelIndex + SINGLE_STEREO_OFFSET] = shortResults[0];
}
int i = 0;
// The following code is pretty gross and redundant, but AFAIK it's the best way to avoid
// too many conditionals in handling the delay samples at the beginning of _clientSamples.
// Basically we try to take the samples in batches of four, and then handle the remainder
// conditionally to get rid of the rest.
while (i + 3 < numSamplesDelay) {
// handle the first cases where we can MMX add four samples at once
const int DOUBLE_STEREO_OFFSET = 4;
const int TRIPLE_STEREO_OFFSET = 6;
if (numSamplesDelay > 0) {
// if there was a sample delay for this buffer, we need to pull samples prior to the nextOutput
// to stick at the beginning
float attenuationAndWeakChannelRatio = attenuationCoefficient * weakChannelAmplitudeRatio;
const int16_t* delayNextOutputStart = nextOutputStart - numSamplesDelay;
if (delayNextOutputStart < bufferStart) {
delayNextOutputStart = bufferStart + ringBufferSampleCapacity - numSamplesDelay;
}
int i = 0;
while (i + 3 < numSamplesDelay) {
// handle the first cases where we can MMX add four samples at once
int parentIndex = i * 2;
__m64 bufferSamples = _mm_set_pi16(_clientSamples[parentIndex + delayedChannelOffset],
_clientSamples[parentIndex + SINGLE_STEREO_OFFSET + delayedChannelOffset],
_clientSamples[parentIndex + DOUBLE_STEREO_OFFSET + delayedChannelOffset],
_clientSamples[parentIndex + TRIPLE_STEREO_OFFSET + delayedChannelOffset]);
__m64 addSamples = _mm_set_pi16(delayNextOutputStart[i] * attenuationAndWeakChannelRatio,
delayNextOutputStart[i + 1] * attenuationAndWeakChannelRatio,
delayNextOutputStart[i + 2] * attenuationAndWeakChannelRatio,
delayNextOutputStart[i + 3] * attenuationAndWeakChannelRatio);
__m64 mmxResult = _mm_adds_pi16(bufferSamples, addSamples);
int16_t* shortResults = reinterpret_cast<int16_t*>(&mmxResult);
_clientSamples[parentIndex + delayedChannelOffset] = shortResults[3];
_clientSamples[parentIndex + SINGLE_STEREO_OFFSET + delayedChannelOffset] = shortResults[2];
_clientSamples[parentIndex + DOUBLE_STEREO_OFFSET + delayedChannelOffset] = shortResults[1];
_clientSamples[parentIndex + TRIPLE_STEREO_OFFSET + delayedChannelOffset] = shortResults[0];
// push the index
i += 4;
}
int parentIndex = i * 2;
__m64 bufferSamples = _mm_set_pi16(_clientSamples[parentIndex + delayedChannelOffset],
_clientSamples[parentIndex + SINGLE_STEREO_OFFSET + delayedChannelOffset],
_clientSamples[parentIndex + DOUBLE_STEREO_OFFSET + delayedChannelOffset],
_clientSamples[parentIndex + TRIPLE_STEREO_OFFSET + delayedChannelOffset]);
__m64 addSamples = _mm_set_pi16(delayNextOutputStart[i] * attenuationAndWeakChannelRatio,
delayNextOutputStart[i + 1] * attenuationAndWeakChannelRatio,
delayNextOutputStart[i + 2] * attenuationAndWeakChannelRatio,
delayNextOutputStart[i + 3] * attenuationAndWeakChannelRatio);
__m64 mmxResult = _mm_adds_pi16(bufferSamples, addSamples);
int16_t* shortResults = reinterpret_cast<int16_t*>(&mmxResult);
_clientSamples[parentIndex + delayedChannelOffset] = shortResults[3];
_clientSamples[parentIndex + SINGLE_STEREO_OFFSET + delayedChannelOffset] = shortResults[2];
_clientSamples[parentIndex + DOUBLE_STEREO_OFFSET + delayedChannelOffset] = shortResults[1];
_clientSamples[parentIndex + TRIPLE_STEREO_OFFSET + delayedChannelOffset] = shortResults[0];
// push the index
i += 4;
if (i + 2 < numSamplesDelay) {
// MMX add only three delayed samples
__m64 bufferSamples = _mm_set_pi16(_clientSamples[parentIndex + delayedChannelOffset],
_clientSamples[parentIndex + SINGLE_STEREO_OFFSET + delayedChannelOffset],
_clientSamples[parentIndex + DOUBLE_STEREO_OFFSET + delayedChannelOffset],
0);
__m64 addSamples = _mm_set_pi16(delayNextOutputStart[i] * attenuationAndWeakChannelRatio,
delayNextOutputStart[i + 1] * attenuationAndWeakChannelRatio,
delayNextOutputStart[i + 2] * attenuationAndWeakChannelRatio,
0);
__m64 mmxResult = _mm_adds_pi16(bufferSamples, addSamples);
int16_t* shortResults = reinterpret_cast<int16_t*>(&mmxResult);
_clientSamples[parentIndex + delayedChannelOffset] = shortResults[3];
_clientSamples[parentIndex + SINGLE_STEREO_OFFSET + delayedChannelOffset] = shortResults[2];
_clientSamples[parentIndex + DOUBLE_STEREO_OFFSET + delayedChannelOffset] = shortResults[1];
} else if (i + 1 < numSamplesDelay) {
// MMX add two delayed samples
__m64 bufferSamples = _mm_set_pi16(_clientSamples[parentIndex + delayedChannelOffset],
_clientSamples[parentIndex + SINGLE_STEREO_OFFSET + delayedChannelOffset], 0, 0);
__m64 addSamples = _mm_set_pi16(delayNextOutputStart[i] * attenuationAndWeakChannelRatio,
delayNextOutputStart[i + 1] * attenuationAndWeakChannelRatio, 0, 0);
__m64 mmxResult = _mm_adds_pi16(bufferSamples, addSamples);
int16_t* shortResults = reinterpret_cast<int16_t*>(&mmxResult);
_clientSamples[parentIndex + delayedChannelOffset] = shortResults[3];
_clientSamples[parentIndex + SINGLE_STEREO_OFFSET + delayedChannelOffset] = shortResults[2];
} else if (i < numSamplesDelay) {
// MMX add a single delayed sample
__m64 bufferSamples = _mm_set_pi16(_clientSamples[parentIndex + delayedChannelOffset], 0, 0, 0);
__m64 addSamples = _mm_set_pi16(delayNextOutputStart[i] * attenuationAndWeakChannelRatio, 0, 0, 0);
__m64 mmxResult = _mm_adds_pi16(bufferSamples, addSamples);
int16_t* shortResults = reinterpret_cast<int16_t*>(&mmxResult);
_clientSamples[parentIndex + delayedChannelOffset] = shortResults[3];
}
}
int parentIndex = i * 2;
if (i + 2 < numSamplesDelay) {
// MMX add only three delayed samples
} else {
// stereo buffer - do attenuation but no sample delay for spatialization
for (int s = 0; s < NETWORK_BUFFER_LENGTH_SAMPLES_STEREO; s += 4) {
// use MMX to clamp four additions at a time
__m64 bufferSamples = _mm_set_pi16(_clientSamples[parentIndex + delayedChannelOffset],
_clientSamples[parentIndex + SINGLE_STEREO_OFFSET + delayedChannelOffset],
_clientSamples[parentIndex + DOUBLE_STEREO_OFFSET + delayedChannelOffset],
0);
__m64 addSamples = _mm_set_pi16(delayNextOutputStart[i] * attenuationAndWeakChannelRatio,
delayNextOutputStart[i + 1] * attenuationAndWeakChannelRatio,
delayNextOutputStart[i + 2] * attenuationAndWeakChannelRatio,
0);
__m64 mmxResult = _mm_adds_pi16(bufferSamples, addSamples);
int16_t* shortResults = reinterpret_cast<int16_t*>(&mmxResult);
_clientSamples[parentIndex + delayedChannelOffset] = shortResults[3];
_clientSamples[parentIndex + SINGLE_STEREO_OFFSET + delayedChannelOffset] = shortResults[2];
_clientSamples[parentIndex + DOUBLE_STEREO_OFFSET + delayedChannelOffset] = shortResults[1];
} else if (i + 1 < numSamplesDelay) {
// MMX add two delayed samples
__m64 bufferSamples = _mm_set_pi16(_clientSamples[parentIndex + delayedChannelOffset],
_clientSamples[parentIndex + SINGLE_STEREO_OFFSET + delayedChannelOffset], 0, 0);
__m64 addSamples = _mm_set_pi16(delayNextOutputStart[i] * attenuationAndWeakChannelRatio,
delayNextOutputStart[i + 1] * attenuationAndWeakChannelRatio, 0, 0);
__m64 bufferSamples = _mm_set_pi16(_clientSamples[s], _clientSamples[s + 1],
_clientSamples[s + 2], _clientSamples[s + 3]);
__m64 addSamples = _mm_set_pi16(nextOutputStart[s] * attenuationCoefficient,
nextOutputStart[s + 1] * attenuationCoefficient,
nextOutputStart[s + 2] * attenuationCoefficient,
nextOutputStart[s + 3] * attenuationCoefficient);
__m64 mmxResult = _mm_adds_pi16(bufferSamples, addSamples);
int16_t* shortResults = reinterpret_cast<int16_t*>(&mmxResult);
_clientSamples[parentIndex + delayedChannelOffset] = shortResults[3];
_clientSamples[parentIndex + SINGLE_STEREO_OFFSET + delayedChannelOffset] = shortResults[2];
} else if (i < numSamplesDelay) {
// MMX add a single delayed sample
__m64 bufferSamples = _mm_set_pi16(_clientSamples[parentIndex + delayedChannelOffset], 0, 0, 0);
__m64 addSamples = _mm_set_pi16(delayNextOutputStart[i] * attenuationAndWeakChannelRatio, 0, 0, 0);
__m64 mmxResult = _mm_adds_pi16(bufferSamples, addSamples);
int16_t* shortResults = reinterpret_cast<int16_t*>(&mmxResult);
_clientSamples[parentIndex + delayedChannelOffset] = shortResults[3];
_clientSamples[s] = shortResults[3];
_clientSamples[s + 1] = shortResults[2];
_clientSamples[s + 2] = shortResults[1];
_clientSamples[s + 3] = shortResults[0];
}
}
}

17
assignment-client/src/audio/AudioMixerClientData.cpp
View file

@ -50,10 +50,22 @@ int AudioMixerClientData::parseData(const QByteArray& packet) {
// grab the AvatarAudioRingBuffer from the vector (or create it if it doesn't exist)
AvatarAudioRingBuffer* avatarRingBuffer = getAvatarAudioRingBuffer();
// read the first byte after the header to see if this is a stereo or mono buffer
quint8 channelFlag = packet.at(numBytesForPacketHeader(packet));
bool isStereo = channelFlag == 1;
if (avatarRingBuffer && avatarRingBuffer->isStereo() != isStereo) {
// there's a mismatch in the buffer channels for the incoming and current buffer
// so delete our current buffer and create a new one
_ringBuffers.removeOne(avatarRingBuffer);
avatarRingBuffer->deleteLater();
avatarRingBuffer = NULL;
}
if (!avatarRingBuffer) {
// we don't have an AvatarAudioRingBuffer yet, so add it
avatarRingBuffer = new AvatarAudioRingBuffer();
avatarRingBuffer = new AvatarAudioRingBuffer(isStereo);
_ringBuffers.push_back(avatarRingBuffer);
}
@ -106,7 +118,8 @@ void AudioMixerClientData::pushBuffersAfterFrameSend() {
PositionalAudioRingBuffer* audioBuffer = _ringBuffers[i];
if (audioBuffer->willBeAddedToMix()) {
audioBuffer->shiftReadPosition(NETWORK_BUFFER_LENGTH_SAMPLES_PER_CHANNEL);
audioBuffer->shiftReadPosition(audioBuffer->isStereo()
? NETWORK_BUFFER_LENGTH_SAMPLES_STEREO : NETWORK_BUFFER_LENGTH_SAMPLES_PER_CHANNEL);
audioBuffer->setWillBeAddedToMix(false);
} else if (audioBuffer->getType() == PositionalAudioRingBuffer::Injector

4
assignment-client/src/audio/AudioMixerClientData.h
View file

@ -24,14 +24,14 @@ public:
AudioMixerClientData();
~AudioMixerClientData();
const std::vector<PositionalAudioRingBuffer*> getRingBuffers() const { return _ringBuffers; }
const QList<PositionalAudioRingBuffer*> getRingBuffers() const { return _ringBuffers; }
AvatarAudioRingBuffer* getAvatarAudioRingBuffer() const;
int parseData(const QByteArray& packet);
void checkBuffersBeforeFrameSend(int jitterBufferLengthSamples);
void pushBuffersAfterFrameSend();
private:
std::vector<PositionalAudioRingBuffer*> _ringBuffers;
QList<PositionalAudioRingBuffer*> _ringBuffers;
};
#endif // hifi_AudioMixerClientData_h

4
assignment-client/src/audio/AvatarAudioRingBuffer.cpp
View file

@ -13,8 +13,8 @@
#include "AvatarAudioRingBuffer.h"
AvatarAudioRingBuffer::AvatarAudioRingBuffer() :
PositionalAudioRingBuffer(PositionalAudioRingBuffer::Microphone) {
AvatarAudioRingBuffer::AvatarAudioRingBuffer(bool isStereo) :
PositionalAudioRingBuffer(PositionalAudioRingBuffer::Microphone, isStereo) {
}

2
assignment-client/src/audio/AvatarAudioRingBuffer.h
View file

@ -18,7 +18,7 @@
class AvatarAudioRingBuffer : public PositionalAudioRingBuffer {
public:
AvatarAudioRingBuffer();
AvatarAudioRingBuffer(bool isStereo = false);
int parseData(const QByteArray& packet);
private:

326
interface/src/Audio.cpp
View file

@ -68,6 +68,7 @@ Audio::Audio(int16_t initialJitterBufferSamples, QObject* parent) :
_proceduralOutputDevice(NULL),
_inputRingBuffer(0),
_ringBuffer(NETWORK_BUFFER_LENGTH_BYTES_PER_CHANNEL),
_isStereoInput(false),
_averagedLatency(0.0),
_measuredJitter(0),
_jitterBufferSamples(initialJitterBufferSamples),
@ -289,20 +290,27 @@ void linearResampling(int16_t* sourceSamples, int16_t* destinationSamples,
if (sourceToDestinationFactor >= 2) {
// we need to downsample from 48 to 24
// for now this only supports a mono output - this would be the case for audio input
for (unsigned int i = sourceAudioFormat.channelCount(); i < numSourceSamples; i += 2 * sourceAudioFormat.channelCount()) {
if (i + (sourceAudioFormat.channelCount()) >= numSourceSamples) {
destinationSamples[(i - sourceAudioFormat.channelCount()) / (int) sourceToDestinationFactor] =
if (destinationAudioFormat.channelCount() == 1) {
for (unsigned int i = sourceAudioFormat.channelCount(); i < numSourceSamples; i += 2 * sourceAudioFormat.channelCount()) {
if (i + (sourceAudioFormat.channelCount()) >= numSourceSamples) {
destinationSamples[(i - sourceAudioFormat.channelCount()) / (int) sourceToDestinationFactor] =
(sourceSamples[i - sourceAudioFormat.channelCount()] / 2)
+ (sourceSamples[i] / 2);
} else {
destinationSamples[(i - sourceAudioFormat.channelCount()) / (int) sourceToDestinationFactor] =
} else {
destinationSamples[(i - sourceAudioFormat.channelCount()) / (int) sourceToDestinationFactor] =
(sourceSamples[i - sourceAudioFormat.channelCount()] / 4)
+ (sourceSamples[i] / 2)
+ (sourceSamples[i + sourceAudioFormat.channelCount()] / 4);
}
}
} else {
// this is a 48 to 24 resampling but both source and destination are two channels
// squish two samples into one in each channel
for (int i = 0; i < numSourceSamples; i += 4) {
destinationSamples[i / 2] = (sourceSamples[i] / 2) + (sourceSamples[i + 2] / 2);
destinationSamples[(i / 2) + 1] = (sourceSamples[i + 1] / 2) + (sourceSamples[i + 3] / 2);
}
}
} else {
if (sourceAudioFormat.sampleRate() == destinationAudioFormat.sampleRate()) {
// mono to stereo, same sample rate
@ -405,12 +413,12 @@ bool Audio::switchOutputToAudioDevice(const QString& outputDeviceName) {
}
void Audio::handleAudioInput() {
static char monoAudioDataPacket[MAX_PACKET_SIZE];
static char audioDataPacket[MAX_PACKET_SIZE];
static int numBytesPacketHeader = numBytesForPacketHeaderGivenPacketType(PacketTypeMicrophoneAudioNoEcho);
static int leadingBytes = numBytesPacketHeader + sizeof(glm::vec3) + sizeof(glm::quat);
static int leadingBytes = numBytesPacketHeader + sizeof(glm::vec3) + sizeof(glm::quat) + sizeof(quint8);
static int16_t* monoAudioSamples = (int16_t*) (monoAudioDataPacket + leadingBytes);
static int16_t* networkAudioSamples = (int16_t*) (audioDataPacket + leadingBytes);
float inputToNetworkInputRatio = calculateDeviceToNetworkInputRatio(_numInputCallbackBytes);
@ -452,126 +460,139 @@ void Audio::handleAudioInput() {
int16_t* inputAudioSamples = new int16_t[inputSamplesRequired];
_inputRingBuffer.readSamples(inputAudioSamples, inputSamplesRequired);
int numNetworkBytes = _isStereoInput ? NETWORK_BUFFER_LENGTH_BYTES_STEREO : NETWORK_BUFFER_LENGTH_BYTES_PER_CHANNEL;
int numNetworkSamples = _isStereoInput ? NETWORK_BUFFER_LENGTH_SAMPLES_STEREO : NETWORK_BUFFER_LENGTH_SAMPLES_PER_CHANNEL;
// zero out the monoAudioSamples array and the locally injected audio
memset(monoAudioSamples, 0, NETWORK_BUFFER_LENGTH_BYTES_PER_CHANNEL);
memset(networkAudioSamples, 0, numNetworkBytes);
if (!_muted) {
// we aren't muted, downsample the input audio
linearResampling((int16_t*) inputAudioSamples,
monoAudioSamples,
inputSamplesRequired,
NETWORK_BUFFER_LENGTH_SAMPLES_PER_CHANNEL,
linearResampling((int16_t*) inputAudioSamples, networkAudioSamples,
inputSamplesRequired, numNetworkSamples,
_inputFormat, _desiredInputFormat);
//
// Impose Noise Gate
//
// The Noise Gate is used to reject constant background noise by measuring the noise
// floor observed at the microphone and then opening the 'gate' to allow microphone
// signals to be transmitted when the microphone samples average level exceeds a multiple
// of the noise floor.
//
// NOISE_GATE_HEIGHT: How loud you have to speak relative to noise background to open the gate.
// Make this value lower for more sensitivity and less rejection of noise.
// NOISE_GATE_WIDTH: The number of samples in an audio frame for which the height must be exceeded
// to open the gate.
// NOISE_GATE_CLOSE_FRAME_DELAY: Once the noise is below the gate height for the frame, how many frames
// will we wait before closing the gate.
// NOISE_GATE_FRAMES_TO_AVERAGE: How many audio frames should we average together to compute noise floor.
// More means better rejection but also can reject continuous things like singing.
// NUMBER_OF_NOISE_SAMPLE_FRAMES: How often should we re-evaluate the noise floor?
float loudness = 0;
float thisSample = 0;
int samplesOverNoiseGate = 0;
const float NOISE_GATE_HEIGHT = 7.0f;
const int NOISE_GATE_WIDTH = 5;
const int NOISE_GATE_CLOSE_FRAME_DELAY = 5;
const int NOISE_GATE_FRAMES_TO_AVERAGE = 5;
const float DC_OFFSET_AVERAGING = 0.99f;
const float CLIPPING_THRESHOLD = 0.90f;
//
// Check clipping, adjust DC offset, and check if should open noise gate
//
float measuredDcOffset = 0.0f;
// Increment the time since the last clip
if (_timeSinceLastClip >= 0.0f) {
_timeSinceLastClip += (float) NETWORK_BUFFER_LENGTH_SAMPLES_PER_CHANNEL / (float) SAMPLE_RATE;
}
for (int i = 0; i < NETWORK_BUFFER_LENGTH_SAMPLES_PER_CHANNEL; i++) {
measuredDcOffset += monoAudioSamples[i];
monoAudioSamples[i] -= (int16_t) _dcOffset;
thisSample = fabsf(monoAudioSamples[i]);
if (thisSample >= (32767.0f * CLIPPING_THRESHOLD)) {
_timeSinceLastClip = 0.0f;
// only impose the noise gate and perform tone injection if we sending mono audio
if (!_isStereoInput) {
//
// Impose Noise Gate
//
// The Noise Gate is used to reject constant background noise by measuring the noise
// floor observed at the microphone and then opening the 'gate' to allow microphone
// signals to be transmitted when the microphone samples average level exceeds a multiple
// of the noise floor.
//
// NOISE_GATE_HEIGHT: How loud you have to speak relative to noise background to open the gate.
// Make this value lower for more sensitivity and less rejection of noise.
// NOISE_GATE_WIDTH: The number of samples in an audio frame for which the height must be exceeded
// to open the gate.
// NOISE_GATE_CLOSE_FRAME_DELAY: Once the noise is below the gate height for the frame, how many frames
// will we wait before closing the gate.
// NOISE_GATE_FRAMES_TO_AVERAGE: How many audio frames should we average together to compute noise floor.
// More means better rejection but also can reject continuous things like singing.
// NUMBER_OF_NOISE_SAMPLE_FRAMES: How often should we re-evaluate the noise floor?
float loudness = 0;
float thisSample = 0;
int samplesOverNoiseGate = 0;
const float NOISE_GATE_HEIGHT = 7.0f;
const int NOISE_GATE_WIDTH = 5;
const int NOISE_GATE_CLOSE_FRAME_DELAY = 5;
const int NOISE_GATE_FRAMES_TO_AVERAGE = 5;
const float DC_OFFSET_AVERAGING = 0.99f;
const float CLIPPING_THRESHOLD = 0.90f;
//
// Check clipping, adjust DC offset, and check if should open noise gate
//
float measuredDcOffset = 0.0f;
// Increment the time since the last clip
if (_timeSinceLastClip >= 0.0f) {
_timeSinceLastClip += (float) NETWORK_BUFFER_LENGTH_SAMPLES_PER_CHANNEL / (float) SAMPLE_RATE;
}
loudness += thisSample;
// Noise Reduction: Count peaks above the average loudness
if (_noiseGateEnabled && (thisSample > (_noiseGateMeasuredFloor * NOISE_GATE_HEIGHT))) {
samplesOverNoiseGate++;
}
}
measuredDcOffset /= NETWORK_BUFFER_LENGTH_SAMPLES_PER_CHANNEL;
if (_dcOffset == 0.0f) {
// On first frame, copy over measured offset
_dcOffset = measuredDcOffset;
} else {
_dcOffset = DC_OFFSET_AVERAGING * _dcOffset + (1.0f - DC_OFFSET_AVERAGING) * measuredDcOffset;
}
// Add tone injection if enabled
const float TONE_FREQ = 220.0f / SAMPLE_RATE * TWO_PI;
const float QUARTER_VOLUME = 8192.0f;
if (_toneInjectionEnabled) {
loudness = 0.0f;
for (int i = 0; i < NETWORK_BUFFER_LENGTH_SAMPLES_PER_CHANNEL; i++) {
monoAudioSamples[i] = QUARTER_VOLUME * sinf(TONE_FREQ * (float)(i + _proceduralEffectSample));
loudness += fabsf(monoAudioSamples[i]);
}
}
_lastInputLoudness = fabs(loudness / NETWORK_BUFFER_LENGTH_SAMPLES_PER_CHANNEL);
// If Noise Gate is enabled, check and turn the gate on and off
if (!_toneInjectionEnabled && _noiseGateEnabled) {
float averageOfAllSampleFrames = 0.0f;
_noiseSampleFrames[_noiseGateSampleCounter++] = _lastInputLoudness;
if (_noiseGateSampleCounter == NUMBER_OF_NOISE_SAMPLE_FRAMES) {
float smallestSample = FLT_MAX;
for (int i = 0; i <= NUMBER_OF_NOISE_SAMPLE_FRAMES - NOISE_GATE_FRAMES_TO_AVERAGE; i += NOISE_GATE_FRAMES_TO_AVERAGE) {
float thisAverage = 0.0f;
for (int j = i; j < i + NOISE_GATE_FRAMES_TO_AVERAGE; j++) {
thisAverage += _noiseSampleFrames[j];
averageOfAllSampleFrames += _noiseSampleFrames[j];
}
thisAverage /= NOISE_GATE_FRAMES_TO_AVERAGE;
if (thisAverage < smallestSample) {
smallestSample = thisAverage;
}
measuredDcOffset += networkAudioSamples[i];
networkAudioSamples[i] -= (int16_t) _dcOffset;
thisSample = fabsf(networkAudioSamples[i]);
if (thisSample >= (32767.0f * CLIPPING_THRESHOLD)) {
_timeSinceLastClip = 0.0f;
}
loudness += thisSample;
// Noise Reduction: Count peaks above the average loudness
if (_noiseGateEnabled && (thisSample > (_noiseGateMeasuredFloor * NOISE_GATE_HEIGHT))) {
samplesOverNoiseGate++;
}
averageOfAllSampleFrames /= NUMBER_OF_NOISE_SAMPLE_FRAMES;
_noiseGateMeasuredFloor = smallestSample;
_noiseGateSampleCounter = 0;
}
if (samplesOverNoiseGate > NOISE_GATE_WIDTH) {
_noiseGateOpen = true;
_noiseGateFramesToClose = NOISE_GATE_CLOSE_FRAME_DELAY;
measuredDcOffset /= NETWORK_BUFFER_LENGTH_SAMPLES_PER_CHANNEL;
if (_dcOffset == 0.0f) {
// On first frame, copy over measured offset
_dcOffset = measuredDcOffset;
} else {
if (--_noiseGateFramesToClose == 0) {
_noiseGateOpen = false;
_dcOffset = DC_OFFSET_AVERAGING * _dcOffset + (1.0f - DC_OFFSET_AVERAGING) * measuredDcOffset;
}
// Add tone injection if enabled
const float TONE_FREQ = 220.0f / SAMPLE_RATE * TWO_PI;
const float QUARTER_VOLUME = 8192.0f;
if (_toneInjectionEnabled) {
loudness = 0.0f;
for (int i = 0; i < NETWORK_BUFFER_LENGTH_SAMPLES_PER_CHANNEL; i++) {
networkAudioSamples[i] = QUARTER_VOLUME * sinf(TONE_FREQ * (float)(i + _proceduralEffectSample));
loudness += fabsf(networkAudioSamples[i]);
}
}
if (!_noiseGateOpen) {
memset(monoAudioSamples, 0, NETWORK_BUFFER_LENGTH_BYTES_PER_CHANNEL);
_lastInputLoudness = 0;
_lastInputLoudness = fabs(loudness / NETWORK_BUFFER_LENGTH_SAMPLES_PER_CHANNEL);
// If Noise Gate is enabled, check and turn the gate on and off
if (!_toneInjectionEnabled && _noiseGateEnabled) {
float averageOfAllSampleFrames = 0.0f;
_noiseSampleFrames[_noiseGateSampleCounter++] = _lastInputLoudness;
if (_noiseGateSampleCounter == NUMBER_OF_NOISE_SAMPLE_FRAMES) {
float smallestSample = FLT_MAX;
for (int i = 0; i <= NUMBER_OF_NOISE_SAMPLE_FRAMES - NOISE_GATE_FRAMES_TO_AVERAGE; i += NOISE_GATE_FRAMES_TO_AVERAGE) {
float thisAverage = 0.0f;
for (int j = i; j < i + NOISE_GATE_FRAMES_TO_AVERAGE; j++) {
thisAverage += _noiseSampleFrames[j];
averageOfAllSampleFrames += _noiseSampleFrames[j];
}
thisAverage /= NOISE_GATE_FRAMES_TO_AVERAGE;
if (thisAverage < smallestSample) {
smallestSample = thisAverage;
}
}
averageOfAllSampleFrames /= NUMBER_OF_NOISE_SAMPLE_FRAMES;
_noiseGateMeasuredFloor = smallestSample;
_noiseGateSampleCounter = 0;
}
if (samplesOverNoiseGate > NOISE_GATE_WIDTH) {
_noiseGateOpen = true;
_noiseGateFramesToClose = NOISE_GATE_CLOSE_FRAME_DELAY;
} else {
if (--_noiseGateFramesToClose == 0) {
_noiseGateOpen = false;
}
}
if (!_noiseGateOpen) {
memset(networkAudioSamples, 0, NETWORK_BUFFER_LENGTH_BYTES_PER_CHANNEL);
_lastInputLoudness = 0;
}
}
} else {
float loudness = 0.0f;
for (int i = 0; i < NETWORK_BUFFER_LENGTH_SAMPLES_STEREO; i++) {
loudness += fabsf(networkAudioSamples[i]);
}
_lastInputLoudness = fabs(loudness / NETWORK_BUFFER_LENGTH_SAMPLES_PER_CHANNEL);
}
} else {
// our input loudness is 0, since we're muted
@ -580,19 +601,19 @@ void Audio::handleAudioInput() {
// at this point we have clean monoAudioSamples, which match our target output...
// this is what we should send to our interested listeners
if (_processSpatialAudio && !_muted && _audioOutput) {
QByteArray monoInputData((char*)monoAudioSamples, NETWORK_BUFFER_LENGTH_SAMPLES_PER_CHANNEL * sizeof(int16_t));
if (_processSpatialAudio && !_muted && !_isStereoInput && _audioOutput) {
QByteArray monoInputData((char*)networkAudioSamples, NETWORK_BUFFER_LENGTH_SAMPLES_PER_CHANNEL * sizeof(int16_t));
emit processLocalAudio(_spatialAudioStart, monoInputData, _desiredInputFormat);
}
if (_proceduralAudioOutput) {
processProceduralAudio(monoAudioSamples, NETWORK_BUFFER_LENGTH_SAMPLES_PER_CHANNEL);
if (!_isStereoInput && _proceduralAudioOutput) {
processProceduralAudio(networkAudioSamples, NETWORK_BUFFER_LENGTH_SAMPLES_PER_CHANNEL);
}
if (_scopeEnabled && !_scopeEnabledPause) {
if (!_isStereoInput && _scopeEnabled && !_scopeEnabledPause) {
unsigned int numMonoAudioChannels = 1;
unsigned int monoAudioChannel = 0;
addBufferToScope(_scopeInput, _scopeInputOffset, monoAudioSamples, monoAudioChannel, numMonoAudioChannels);
addBufferToScope(_scopeInput, _scopeInputOffset, networkAudioSamples, monoAudioChannel, numMonoAudioChannels);
_scopeInputOffset += NETWORK_SAMPLES_PER_FRAME;
_scopeInputOffset %= _samplesPerScope;
}
@ -604,9 +625,7 @@ void Audio::handleAudioInput() {
MyAvatar* interfaceAvatar = Application::getInstance()->getAvatar();
glm::vec3 headPosition = interfaceAvatar->getHead()->getPosition();
glm::quat headOrientation = interfaceAvatar->getHead()->getFinalOrientationInWorldFrame();
// we need the amount of bytes in the buffer + 1 for type
// + 12 for 3 floats for position + float for bearing + 1 attenuation byte
quint8 isStereo = _isStereoInput ? 1 : 0;
int numAudioBytes = 0;
@ -615,11 +634,12 @@ void Audio::handleAudioInput() {
packetType = PacketTypeSilentAudioFrame;
// we need to indicate how many silent samples this is to the audio mixer
monoAudioSamples[0] = NETWORK_BUFFER_LENGTH_SAMPLES_PER_CHANNEL;
audioDataPacket[0] = _isStereoInput
? NETWORK_BUFFER_LENGTH_SAMPLES_STEREO
: NETWORK_BUFFER_LENGTH_SAMPLES_PER_CHANNEL;
numAudioBytes = sizeof(int16_t);
} else {
numAudioBytes = NETWORK_BUFFER_LENGTH_BYTES_PER_CHANNEL;
numAudioBytes = _isStereoInput ? NETWORK_BUFFER_LENGTH_BYTES_STEREO : NETWORK_BUFFER_LENGTH_BYTES_PER_CHANNEL;
if (Menu::getInstance()->isOptionChecked(MenuOption::EchoServerAudio)) {
packetType = PacketTypeMicrophoneAudioWithEcho;
@ -628,7 +648,10 @@ void Audio::handleAudioInput() {
}
}
char* currentPacketPtr = monoAudioDataPacket + populatePacketHeader(monoAudioDataPacket, packetType);
char* currentPacketPtr = audioDataPacket + populatePacketHeader(audioDataPacket, packetType);
// set the mono/stereo byte
*currentPacketPtr++ = isStereo;
// memcpy the three float positions
memcpy(currentPacketPtr, &headPosition, sizeof(headPosition));
@ -638,7 +661,7 @@ void Audio::handleAudioInput() {
memcpy(currentPacketPtr, &headOrientation, sizeof(headOrientation));
currentPacketPtr += sizeof(headOrientation);
nodeList->writeDatagram(monoAudioDataPacket, numAudioBytes + leadingBytes, audioMixer);
nodeList->writeDatagram(audioDataPacket, numAudioBytes + leadingBytes, audioMixer);
Application::getInstance()->getBandwidthMeter()->outputStream(BandwidthMeter::AUDIO)
.updateValue(numAudioBytes + leadingBytes);
@ -761,6 +784,24 @@ void Audio::toggleAudioNoiseReduction() {
_noiseGateEnabled = !_noiseGateEnabled;
}
void Audio::toggleStereoInput() {
int oldChannelCount = _desiredInputFormat.channelCount();
QAction* stereoAudioOption = Menu::getInstance()->getActionForOption(MenuOption::StereoAudio);
if (stereoAudioOption->isChecked()) {
_desiredInputFormat.setChannelCount(2);
_isStereoInput = true;
} else {
_desiredInputFormat.setChannelCount(1);
_isStereoInput = false;
}
if (oldChannelCount != _desiredInputFormat.channelCount()) {
// change in channel count for desired input format, restart the input device
switchInputToAudioDevice(_inputAudioDeviceName);
}
}
void Audio::processReceivedAudio(const QByteArray& audioByteArray) {
_ringBuffer.parseData(audioByteArray);
@ -1300,18 +1341,21 @@ bool Audio::switchInputToAudioDevice(const QAudioDeviceInfo& inputDeviceInfo) {
if (adjustedFormatForAudioDevice(inputDeviceInfo, _desiredInputFormat, _inputFormat)) {
qDebug() << "The format to be used for audio input is" << _inputFormat;
_audioInput = new QAudioInput(inputDeviceInfo, _inputFormat, this);
_numInputCallbackBytes = calculateNumberOfInputCallbackBytes(_inputFormat);
_audioInput->setBufferSize(_numInputCallbackBytes);
// how do we want to handle input working, but output not working?
int numFrameSamples = calculateNumberOfFrameSamples(_numInputCallbackBytes);
_inputRingBuffer.resizeForFrameSize(numFrameSamples);
_inputDevice = _audioInput->start();
connect(_inputDevice, SIGNAL(readyRead()), this, SLOT(handleAudioInput()));
supportedFormat = true;
// if the user wants stereo but this device can't provide then bail
if (!_isStereoInput || _inputFormat.channelCount() == 2) {
_audioInput = new QAudioInput(inputDeviceInfo, _inputFormat, this);
_numInputCallbackBytes = calculateNumberOfInputCallbackBytes(_inputFormat);
_audioInput->setBufferSize(_numInputCallbackBytes);
// how do we want to handle input working, but output not working?
int numFrameSamples = calculateNumberOfFrameSamples(_numInputCallbackBytes);
_inputRingBuffer.resizeForFrameSize(numFrameSamples);
_inputDevice = _audioInput->start();
connect(_inputDevice, SIGNAL(readyRead()), this, SLOT(handleAudioInput()));
supportedFormat = true;
}
}
}
return supportedFormat;

2
interface/src/Audio.h
View file

@ -85,6 +85,7 @@ public slots:
void toggleScope();
void toggleScopePause();
void toggleAudioSpatialProcessing();
void toggleStereoInput();
void selectAudioScopeFiveFrames();
void selectAudioScopeTwentyFrames();
void selectAudioScopeFiftyFrames();
@ -127,6 +128,7 @@ private:
QIODevice* _proceduralOutputDevice;
AudioRingBuffer _inputRingBuffer;
AudioRingBuffer _ringBuffer;
bool _isStereoInput;
QString _inputAudioDeviceName;
QString _outputAudioDeviceName;

2
interface/src/Menu.cpp
View file

@ -432,6 +432,8 @@ Menu::Menu() :
SLOT(toggleAudioNoiseReduction()));
addCheckableActionToQMenuAndActionHash(audioDebugMenu, MenuOption::EchoServerAudio);
addCheckableActionToQMenuAndActionHash(audioDebugMenu, MenuOption::EchoLocalAudio);
addCheckableActionToQMenuAndActionHash(audioDebugMenu, MenuOption::StereoAudio, 0, false,
appInstance->getAudio(), SLOT(toggleStereoInput()));
addCheckableActionToQMenuAndActionHash(audioDebugMenu, MenuOption::MuteAudio,
Qt::CTRL | Qt::Key_M,
false,

1
interface/src/Menu.h
View file

@ -402,6 +402,7 @@ namespace MenuOption {
const QString StandOnNearbyFloors = "Stand on nearby floors";
const QString Stars = "Stars";
const QString Stats = "Stats";
const QString StereoAudio = "Stereo Audio";
const QString StopAllScripts = "Stop All Scripts";
const QString SuppressShortTimings = "Suppress Timings Less than 10ms";
const QString TestPing = "Test Ping";

10
libraries/audio/src/PositionalAudioRingBuffer.cpp
View file

@ -20,14 +20,15 @@
#include "PositionalAudioRingBuffer.h"
PositionalAudioRingBuffer::PositionalAudioRingBuffer(PositionalAudioRingBuffer::Type type) :
AudioRingBuffer(NETWORK_BUFFER_LENGTH_SAMPLES_PER_CHANNEL),
PositionalAudioRingBuffer::PositionalAudioRingBuffer(PositionalAudioRingBuffer::Type type, bool isStereo) :
AudioRingBuffer(isStereo ? NETWORK_BUFFER_LENGTH_SAMPLES_STEREO : NETWORK_BUFFER_LENGTH_SAMPLES_PER_CHANNEL),
_type(type),
_position(0.0f, 0.0f, 0.0f),
_orientation(0.0f, 0.0f, 0.0f, 0.0f),
_willBeAddedToMix(false),
_shouldLoopbackForNode(false),
_shouldOutputStarveDebug(true)
_shouldOutputStarveDebug(true),
_isStereo(isStereo)
{
}
@ -40,6 +41,9 @@ int PositionalAudioRingBuffer::parseData(const QByteArray& packet) {
// skip the packet header (includes the source UUID)
int readBytes = numBytesForPacketHeader(packet);
// hop over the channel flag that has already been read in AudioMixerClientData
readBytes += sizeof(quint8);
// read the positional data
readBytes += parsePositionalData(packet.mid(readBytes));
if (packetTypeForPacket(packet) == PacketTypeSilentAudioFrame) {

5
libraries/audio/src/PositionalAudioRingBuffer.h
View file

@ -24,7 +24,7 @@ public:
Injector
};
PositionalAudioRingBuffer(PositionalAudioRingBuffer::Type type);
PositionalAudioRingBuffer(PositionalAudioRingBuffer::Type type, bool isStereo = false);
~PositionalAudioRingBuffer();
int parseData(const QByteArray& packet);
@ -41,6 +41,8 @@ public:
bool shouldLoopbackForNode() const { return _shouldLoopbackForNode; }
bool isStereo() const { return _isStereo; }
PositionalAudioRingBuffer::Type getType() const { return _type; }
const glm::vec3& getPosition() const { return _position; }
const glm::quat& getOrientation() const { return _orientation; }
@ -56,6 +58,7 @@ protected:
bool _willBeAddedToMix;
bool _shouldLoopbackForNode;
bool _shouldOutputStarveDebug;
bool _isStereo;
float _nextOutputTrailingLoudness;
};

3
libraries/networking/src/PacketHeaders.cpp
View file

@ -47,6 +47,9 @@ int packArithmeticallyCodedValue(int value, char* destination) {
PacketVersion versionForPacketType(PacketType type) {
switch (type) {
case PacketTypeMicrophoneAudioNoEcho:
case PacketTypeMicrophoneAudioWithEcho:
return 1;
case PacketTypeAvatarData:
return 3;
case PacketTypeAvatarIdentity: