Merge pull request #2413 from birarda/audio-scaling

more audio scaling tweaks
This commit is contained in:
Philip Rosedale 2014-03-20 17:12:28 -07:00
commit 57fb0c9d0f
6 changed files with 70 additions and 60 deletions

View file

@ -63,6 +63,7 @@ void attachNewBufferToNode(Node *newNode) {
AudioMixer::AudioMixer(const QByteArray& packet) :
ThreadedAssignment(packet),
_trailingSleepRatio(1.0f),
_minSourceLoudnessInFrame(1.0f),
_maxSourceLoudnessInFrame(0.0f),
_loudnessCutoffRatio(0.0f),
@ -305,7 +306,7 @@ void AudioMixer::prepareMixForListeningNode(Node* node) {
if ((*otherNode != *node
|| otherNodeBuffer->shouldLoopbackForNode())
&& otherNodeBuffer->willBeAddedToMix()
&& otherNodeClientData->getNextOutputLoudness() > _minRequiredLoudness) {
&& otherNodeBuffer->getAverageLoudness() > _minRequiredLoudness) {
addBufferToMixForListeningNodeWithBuffer(otherNodeBuffer, nodeRingBuffer);
}
}
@ -355,8 +356,7 @@ void AudioMixer::run() {
char* clientMixBuffer = new char[NETWORK_BUFFER_LENGTH_BYTES_STEREO
+ numBytesForPacketHeaderGivenPacketType(PacketTypeMixedAudio)];
int usecToSleep = 0;
bool isFirstRun = true;
int usecToSleep = BUFFER_SEND_INTERVAL_USECS;
while (!_isFinished) {
@ -371,46 +371,48 @@ void AudioMixer::run() {
}
}
if (!isFirstRun) {
const float STRUGGLE_TRIGGER_SLEEP_PERCENTAGE_THRESHOLD = 0.10;
const float BACK_OFF_TRIGGER_SLEEP_PERCENTAGE_THRESHOLD = 0.30;
const float CUTOFF_EPSILON = 0.0001;
float percentageSleep = (usecToSleep / (float) BUFFER_SEND_INTERVAL_USECS);
float lastCutoffRatio = _loudnessCutoffRatio;
bool hasRatioChanged = false;
if (percentageSleep <= STRUGGLE_TRIGGER_SLEEP_PERCENTAGE_THRESHOLD || usecToSleep < 0) {
// we're struggling - change our min required loudness to reduce some load
_loudnessCutoffRatio += (1 - _loudnessCutoffRatio) / 2;
qDebug() << "Mixer is struggling, sleeping" << percentageSleep * 100 << "% of frame time. Old cutoff was"
<< lastCutoffRatio << "and is now" << _loudnessCutoffRatio;
hasRatioChanged = true;
} else if (percentageSleep >= BACK_OFF_TRIGGER_SLEEP_PERCENTAGE_THRESHOLD && _loudnessCutoffRatio != 0) {
// we've recovered and can back off the required loudness
_loudnessCutoffRatio -= _loudnessCutoffRatio / 2;
if (_loudnessCutoffRatio < CUTOFF_EPSILON) {
_loudnessCutoffRatio = 0;
}
qDebug() << "Mixer is recovering, sleeping" << percentageSleep * 100 << "% of frame time. Old cutoff was"
<< lastCutoffRatio << "and is now" << _loudnessCutoffRatio;
hasRatioChanged = true;
}
if (hasRatioChanged) {
// set out min required loudness from the new ratio
_minRequiredLoudness = _loudnessCutoffRatio * (_maxSourceLoudnessInFrame - _minSourceLoudnessInFrame);
qDebug() << "Minimum loudness required to be mixed is now" << _minRequiredLoudness;
}
const float STRUGGLE_TRIGGER_SLEEP_PERCENTAGE_THRESHOLD = 0.10;
const float BACK_OFF_TRIGGER_SLEEP_PERCENTAGE_THRESHOLD = 0.30;
const float CUTOFF_EPSILON = 0.0001;
} else {
isFirstRun = false;
const int TRAILING_AVERAGE_FRAMES = 100;
const float CURRENT_FRAME_RATIO = 1.0f / TRAILING_AVERAGE_FRAMES;
const float PREVIOUS_FRAMES_RATIO = 1 - CURRENT_FRAME_RATIO;
if (usecToSleep < 0) {
usecToSleep = 0;
}
_trailingSleepRatio = (PREVIOUS_FRAMES_RATIO * _trailingSleepRatio)
+ (usecToSleep * CURRENT_FRAME_RATIO / (float) BUFFER_SEND_INTERVAL_USECS);
float lastCutoffRatio = _loudnessCutoffRatio;
bool hasRatioChanged = false;
if (_trailingSleepRatio <= STRUGGLE_TRIGGER_SLEEP_PERCENTAGE_THRESHOLD) {
// we're struggling - change our min required loudness to reduce some load
_loudnessCutoffRatio += (1 - _loudnessCutoffRatio) / 2;
qDebug() << "Mixer is struggling, sleeping" << _trailingSleepRatio * 100 << "% of frame time. Old cutoff was"
<< lastCutoffRatio << "and is now" << _loudnessCutoffRatio;
hasRatioChanged = true;
} else if (_trailingSleepRatio >= BACK_OFF_TRIGGER_SLEEP_PERCENTAGE_THRESHOLD && _loudnessCutoffRatio != 0) {
// we've recovered and can back off the required loudness
_loudnessCutoffRatio -= _loudnessCutoffRatio / 2;
if (_loudnessCutoffRatio < CUTOFF_EPSILON) {
_loudnessCutoffRatio = 0;
}
qDebug() << "Mixer is recovering, sleeping" << _trailingSleepRatio * 100 << "% of frame time. Old cutoff was"
<< lastCutoffRatio << "and is now" << _loudnessCutoffRatio;
hasRatioChanged = true;
}
if (hasRatioChanged) {
// set out min required loudness from the new ratio
_minRequiredLoudness = _loudnessCutoffRatio * (_maxSourceLoudnessInFrame - _minSourceLoudnessInFrame);
qDebug() << "Minimum loudness required to be mixed is now" << _minRequiredLoudness;
}
foreach (const SharedNodePointer& node, nodeList->getNodeHash()) {

View file

@ -40,6 +40,7 @@ private:
// we are MMX adding 4 samples at a time so we need client samples to have an extra 4
int16_t _clientSamples[NETWORK_BUFFER_LENGTH_SAMPLES_STEREO + (SAMPLE_PHASE_DELAY_AT_90 * 2)];
float _trailingSleepRatio;
float _minSourceLoudnessInFrame;
float _maxSourceLoudnessInFrame;
float _loudnessCutoffRatio;

View file

@ -16,8 +16,7 @@
#include "AudioMixerClientData.h"
AudioMixerClientData::AudioMixerClientData() :
_ringBuffers(),
_nextOutputLoudness(0)
_ringBuffers()
{
}
@ -93,16 +92,19 @@ void AudioMixerClientData::checkBuffersBeforeFrameSend(int jitterBufferLengthSam
// set its flag so we know to push its buffer when all is said and done
_ringBuffers[i]->setWillBeAddedToMix(true);
// calculate the average loudness for the next NETWORK_BUFFER_LENGTH_SAMPLES_PER_CHANNEL
// that would be mixed in
_nextOutputLoudness = _ringBuffers[i]->averageLoudnessForBoundarySamples(NETWORK_BUFFER_LENGTH_SAMPLES_PER_CHANNEL);
_ringBuffers[i]->updateAverageLoudnessForBoundarySamples(NETWORK_BUFFER_LENGTH_SAMPLES_PER_CHANNEL);
if (_nextOutputLoudness != 0 && _nextOutputLoudness < currentMinLoudness) {
currentMinLoudness = _nextOutputLoudness;
float ringBufferLoudness = _ringBuffers[i]->getAverageLoudness();
if (ringBufferLoudness != 0 && ringBufferLoudness < currentMinLoudness) {
currentMinLoudness = ringBufferLoudness;
}
if (_nextOutputLoudness > currentMaxLoudness) {
currentMaxLoudness = _nextOutputLoudness;
if (ringBufferLoudness > currentMaxLoudness) {
currentMaxLoudness = ringBufferLoudness;
}
}
}

View file

@ -24,14 +24,11 @@ public:
const std::vector<PositionalAudioRingBuffer*> getRingBuffers() const { return _ringBuffers; }
AvatarAudioRingBuffer* getAvatarAudioRingBuffer() const;
float getNextOutputLoudness() const { return _nextOutputLoudness; }
int parseData(const QByteArray& packet);
void checkBuffersBeforeFrameSend(int jitterBufferLengthSamples, float& currentMinLoudness, float& currentMaxLoudness);
void pushBuffersAfterFrameSend();
private:
std::vector<PositionalAudioRingBuffer*> _ringBuffers;
float _nextOutputLoudness;
};
#endif /* defined(__hifi__AudioMixerClientData__) */

View file

@ -19,7 +19,8 @@ AudioRingBuffer::AudioRingBuffer(int numFrameSamples) :
NodeData(),
_sampleCapacity(numFrameSamples * RING_BUFFER_LENGTH_FRAMES),
_isStarved(true),
_hasStarted(false)
_hasStarted(false),
_averageLoudness(0)
{
if (numFrameSamples) {
_buffer = new int16_t[_sampleCapacity];
@ -55,18 +56,22 @@ int AudioRingBuffer::parseData(const QByteArray& packet) {
return writeData(packet.data() + numBytesPacketHeader, packet.size() - numBytesPacketHeader);
}
float AudioRingBuffer::averageLoudnessForBoundarySamples(int numSamples) {
void AudioRingBuffer::updateAverageLoudnessForBoundarySamples(int numSamples) {
// ForBoundarySamples means that we expect the number of samples not to roll of the end of the ring buffer
float averageLoudness = 0;
float nextLoudness = 0;
for (int i = 0; i < numSamples; ++i) {
averageLoudness += fabsf(_nextOutput[i]);
nextLoudness += fabsf(_nextOutput[i]);
}
averageLoudness /= numSamples;
averageLoudness /= MAX_SAMPLE_VALUE;
nextLoudness /= numSamples;
nextLoudness /= MAX_SAMPLE_VALUE;
const int TRAILING_AVERAGE_FRAMES = 100;
const float CURRENT_FRAME_RATIO = 1.0f / TRAILING_AVERAGE_FRAMES;
const float PREVIOUS_FRAMES_RATIO = 1 - CURRENT_FRAME_RATIO;
return averageLoudness;
_averageLoudness = (_averageLoudness * PREVIOUS_FRAMES_RATIO) + (CURRENT_FRAME_RATIO * nextLoudness);
}
qint64 AudioRingBuffer::readSamples(int16_t* destination, qint64 maxSamples) {

View file

@ -50,7 +50,8 @@ public:
const int16_t* getNextOutput() { return _nextOutput; }
const int16_t* getBuffer() { return _buffer; }
float averageLoudnessForBoundarySamples(int numSamples);
void updateAverageLoudnessForBoundarySamples(int numSamples);
float getAverageLoudness() const { return _averageLoudness; }
qint64 readSamples(int16_t* destination, qint64 maxSamples);
qint64 writeSamples(const int16_t* source, qint64 maxSamples);
@ -85,6 +86,8 @@ protected:
int16_t* _buffer;
bool _isStarved;
bool _hasStarted;
float _averageLoudness;
};
#endif /* defined(__interface__AudioRingBuffer__) */