mirror of
https://github.com/overte-org/overte.git
synced 2025-04-20 02:03:57 +02:00
Fix heap corruption.
This commit is contained in:
parent
8661bdf788
commit
c3934d9e86
2 changed files with 78 additions and 49 deletions
|
@ -95,6 +95,7 @@ void Audio::reset() {
|
|||
QAudioDeviceInfo getNamedAudioDeviceForMode(QAudio::Mode mode, const QString& deviceName) {
|
||||
QAudioDeviceInfo result;
|
||||
foreach(QAudioDeviceInfo audioDevice, QAudioDeviceInfo::availableDevices(mode)) {
|
||||
qDebug() << audioDevice.deviceName() << " " << deviceName;
|
||||
if (audioDevice.deviceName().trimmed() == deviceName.trimmed()) {
|
||||
result = audioDevice;
|
||||
}
|
||||
|
@ -162,6 +163,8 @@ QAudioDeviceInfo defaultAudioDeviceForMode(QAudio::Mode mode) {
|
|||
qDebug() << "output device:" << woc.szPname;
|
||||
deviceName = woc.szPname;
|
||||
}
|
||||
qDebug() << "DEBUG [" << deviceName << "] [" << getNamedAudioDeviceForMode(mode, deviceName).deviceName() << "]";
|
||||
|
||||
return getNamedAudioDeviceForMode(mode, deviceName);
|
||||
#endif
|
||||
|
||||
|
@ -321,10 +324,12 @@ QVector<QString> Audio::getDeviceNames(QAudio::Mode mode) {
|
|||
}
|
||||
|
||||
bool Audio::switchInputToAudioDevice(const QString& inputDeviceName) {
|
||||
qDebug() << "DEBUG [" << inputDeviceName << "] [" << getNamedAudioDeviceForMode(QAudio::AudioInput, inputDeviceName).deviceName() << "]";
|
||||
return switchInputToAudioDevice(getNamedAudioDeviceForMode(QAudio::AudioInput, inputDeviceName));
|
||||
}
|
||||
|
||||
bool Audio::switchOutputToAudioDevice(const QString& outputDeviceName) {
|
||||
qDebug() << "DEBUG [" << outputDeviceName << "] [" << getNamedAudioDeviceForMode(QAudio::AudioOutput, outputDeviceName).deviceName() << "]";
|
||||
return switchOutputToAudioDevice(getNamedAudioDeviceForMode(QAudio::AudioOutput, outputDeviceName));
|
||||
}
|
||||
|
||||
|
@ -343,7 +348,7 @@ void Audio::handleAudioInput() {
|
|||
|
||||
QByteArray inputByteArray = _inputDevice->readAll();
|
||||
|
||||
if (Menu::getInstance()->isOptionChecked(MenuOption::EchoLocalAudio) && !_muted) {
|
||||
if (Menu::getInstance()->isOptionChecked(MenuOption::EchoLocalAudio) && !_muted && _audioOutput) {
|
||||
// if this person wants local loopback add that to the locally injected audio
|
||||
|
||||
if (!_loopbackOutputDevice && _loopbackAudioOutput) {
|
||||
|
@ -356,7 +361,7 @@ void Audio::handleAudioInput() {
|
|||
_loopbackOutputDevice->write(inputByteArray);
|
||||
}
|
||||
} else {
|
||||
static float loopbackOutputToInputRatio = (_outputFormat.sampleRate() / (float) _inputFormat.sampleRate())
|
||||
float loopbackOutputToInputRatio = (_outputFormat.sampleRate() / (float) _inputFormat.sampleRate())
|
||||
* (_outputFormat.channelCount() / _inputFormat.channelCount());
|
||||
|
||||
QByteArray loopBackByteArray(inputByteArray.size() * loopbackOutputToInputRatio, 0);
|
||||
|
@ -381,9 +386,6 @@ void Audio::handleAudioInput() {
|
|||
// zero out the monoAudioSamples array and the locally injected audio
|
||||
memset(monoAudioSamples, 0, NETWORK_BUFFER_LENGTH_BYTES_PER_CHANNEL);
|
||||
|
||||
// zero out the locally injected audio in preparation for audio procedural sounds
|
||||
memset(_localProceduralSamples, 0, NETWORK_BUFFER_LENGTH_BYTES_PER_CHANNEL);
|
||||
|
||||
if (!_muted) {
|
||||
// we aren't muted, downsample the input audio
|
||||
linearResampling((int16_t*) inputAudioSamples,
|
||||
|
@ -492,28 +494,8 @@ void Audio::handleAudioInput() {
|
|||
_lastInputLoudness = 0;
|
||||
}
|
||||
|
||||
// add procedural effects to the appropriate input samples
|
||||
addProceduralSounds(monoAudioSamples,
|
||||
NETWORK_BUFFER_LENGTH_SAMPLES_PER_CHANNEL);
|
||||
|
||||
if (!_proceduralOutputDevice && _proceduralAudioOutput) {
|
||||
_proceduralOutputDevice = _proceduralAudioOutput->start();
|
||||
}
|
||||
|
||||
// send whatever procedural sounds we want to locally loop back to the _proceduralOutputDevice
|
||||
QByteArray proceduralOutput;
|
||||
proceduralOutput.resize(NETWORK_BUFFER_LENGTH_SAMPLES_PER_CHANNEL * _outputFormat.sampleRate() *
|
||||
_outputFormat.channelCount() * sizeof(int16_t) / (_desiredInputFormat.sampleRate() *
|
||||
_desiredInputFormat.channelCount()));
|
||||
|
||||
linearResampling(_localProceduralSamples,
|
||||
reinterpret_cast<int16_t*>(proceduralOutput.data()),
|
||||
NETWORK_BUFFER_LENGTH_SAMPLES_PER_CHANNEL,
|
||||
proceduralOutput.size() / sizeof(int16_t),
|
||||
_desiredInputFormat, _outputFormat);
|
||||
|
||||
if (_proceduralOutputDevice) {
|
||||
_proceduralOutputDevice->write(proceduralOutput);
|
||||
if (_proceduralAudioOutput) {
|
||||
processProceduralAudio(monoAudioSamples, NETWORK_BUFFER_LENGTH_SAMPLES_PER_CHANNEL);
|
||||
}
|
||||
|
||||
NodeList* nodeList = NodeList::getInstance();
|
||||
|
@ -593,9 +575,43 @@ void Audio::addReceivedAudioToBuffer(const QByteArray& audioByteArray) {
|
|||
}
|
||||
}
|
||||
|
||||
if (_audioOutput) {
|
||||
// Audio output must exist and be correctly set up if we're going to process received audio
|
||||
processReceivedAudio(audioByteArray);
|
||||
}
|
||||
|
||||
Application::getInstance()->getBandwidthMeter()->inputStream(BandwidthMeter::AUDIO).updateValue(audioByteArray.size());
|
||||
|
||||
_lastReceiveTime = currentReceiveTime;
|
||||
}
|
||||
|
||||
bool Audio::mousePressEvent(int x, int y) {
|
||||
if (_iconBounds.contains(x, y)) {
|
||||
toggleMute();
|
||||
return true;
|
||||
}
|
||||
return false;
|
||||
}
|
||||
|
||||
void Audio::toggleMute() {
|
||||
_muted = !_muted;
|
||||
muteToggled();
|
||||
}
|
||||
|
||||
void Audio::toggleAudioNoiseReduction() {
|
||||
_noiseGateEnabled = !_noiseGateEnabled;
|
||||
}
|
||||
|
||||
void Audio::processReceivedAudio(const QByteArray& audioByteArray)
|
||||
{
|
||||
_ringBuffer.parseData(audioByteArray);
|
||||
|
||||
static float networkOutputToOutputRatio = (_desiredOutputFormat.sampleRate() / (float) _outputFormat.sampleRate())
|
||||
float outSampleRate = _outputFormat.sampleRate();
|
||||
float desSampleRate = _desiredOutputFormat.sampleRate();
|
||||
int outChannelCount = _outputFormat.channelCount();
|
||||
int desChannelCount = _desiredOutputFormat.channelCount();
|
||||
float myOutputRatio = desSampleRate / outSampleRate * (desChannelCount / outChannelCount);
|
||||
float networkOutputToOutputRatio = (_desiredOutputFormat.sampleRate() / (float) _outputFormat.sampleRate())
|
||||
* (_desiredOutputFormat.channelCount() / (float) _outputFormat.channelCount());
|
||||
|
||||
if (!_ringBuffer.isStarved() && _audioOutput && _audioOutput->bytesFree() == _audioOutput->bufferSize()) {
|
||||
|
@ -650,33 +666,38 @@ void Audio::addReceivedAudioToBuffer(const QByteArray& audioByteArray) {
|
|||
}
|
||||
delete[] ringBufferSamples;
|
||||
}
|
||||
|
||||
}
|
||||
|
||||
Application::getInstance()->getBandwidthMeter()->inputStream(BandwidthMeter::AUDIO).updateValue(audioByteArray.size());
|
||||
|
||||
_lastReceiveTime = currentReceiveTime;
|
||||
}
|
||||
|
||||
bool Audio::mousePressEvent(int x, int y) {
|
||||
if (_iconBounds.contains(x, y)) {
|
||||
toggleMute();
|
||||
return true;
|
||||
void Audio::processProceduralAudio(int16_t* monoInput, int numSamples) {
|
||||
|
||||
// zero out the locally injected audio in preparation for audio procedural sounds
|
||||
// This is correlated to numSamples, so it really needs to be numSamples * sizeof(sample)
|
||||
memset(_localProceduralSamples, 0, NETWORK_BUFFER_LENGTH_BYTES_PER_CHANNEL);
|
||||
// add procedural effects to the appropriate input samples
|
||||
addProceduralSounds(monoInput, NETWORK_BUFFER_LENGTH_SAMPLES_PER_CHANNEL);
|
||||
|
||||
if (!_proceduralOutputDevice) {
|
||||
_proceduralOutputDevice = _proceduralAudioOutput->start();
|
||||
}
|
||||
|
||||
// send whatever procedural sounds we want to locally loop back to the _proceduralOutputDevice
|
||||
QByteArray proceduralOutput;
|
||||
proceduralOutput.resize(NETWORK_BUFFER_LENGTH_SAMPLES_PER_CHANNEL * _outputFormat.sampleRate() *
|
||||
_outputFormat.channelCount() * sizeof(int16_t) / (_desiredInputFormat.sampleRate() *
|
||||
_desiredInputFormat.channelCount()));
|
||||
|
||||
linearResampling(_localProceduralSamples,
|
||||
reinterpret_cast<int16_t*>(proceduralOutput.data()),
|
||||
NETWORK_BUFFER_LENGTH_SAMPLES_PER_CHANNEL,
|
||||
proceduralOutput.size() / sizeof(int16_t),
|
||||
_desiredInputFormat, _outputFormat);
|
||||
|
||||
if (_proceduralOutputDevice) {
|
||||
_proceduralOutputDevice->write(proceduralOutput);
|
||||
}
|
||||
return false;
|
||||
}
|
||||
|
||||
void Audio::toggleMute() {
|
||||
_muted = !_muted;
|
||||
muteToggled();
|
||||
}
|
||||
|
||||
void Audio::toggleAudioNoiseReduction() {
|
||||
_noiseGateEnabled = !_noiseGateEnabled;
|
||||
}
|
||||
|
||||
|
||||
|
||||
// Take a pointer to the acquired microphone input samples and add procedural sounds
|
||||
void Audio::addProceduralSounds(int16_t* monoInput, int numSamples) {
|
||||
float sample;
|
||||
|
|
|
@ -165,9 +165,17 @@ private:
|
|||
// Audio callback in class context.
|
||||
inline void performIO(int16_t* inputLeft, int16_t* outputLeft, int16_t* outputRight);
|
||||
|
||||
// Process procedural audio by
|
||||
// 1. Echo to the local procedural output device
|
||||
// 2. Mix with the audio input
|
||||
void processProceduralAudio(int16_t* monoInput, int numSamples);
|
||||
|
||||
// Add sounds that we want the user to not hear themselves, by adding on top of mic input signal
|
||||
void addProceduralSounds(int16_t* monoInput, int numSamples);
|
||||
|
||||
// Process received audio
|
||||
void processReceivedAudio(const QByteArray& audioByteArray);
|
||||
|
||||
bool switchInputToAudioDevice(const QAudioDeviceInfo& inputDeviceInfo);
|
||||
bool switchOutputToAudioDevice(const QAudioDeviceInfo& outputDeviceInfo);
|
||||
};
|
||||
|
|
Loading…
Reference in a new issue