mirror of
https://github.com/saitohirga/WSJT-X.git
synced 2024-11-21 19:55:20 -05:00
60792182ad
The two environment variables: WSJT_RX_AUDIO_BUFFER_FRAMES WSJT_TX_AUDIO_BUFFER_FRAMES each can be defined to an integer number which will be used as the suggested audio buffer size for Rx and Tx respectively. Not setting the variable or setting it to zero or less will cause the default buffer size to be used, which should be a good choice for most, if not all, systems.
213 lines
5.1 KiB
C++
213 lines
5.1 KiB
C++
#include "soundout.h"
|
|
|
|
#include <QDateTime>
|
|
#include <QAudioDeviceInfo>
|
|
#include <QAudioOutput>
|
|
#include <QSysInfo>
|
|
#include <qmath.h>
|
|
#include <QDebug>
|
|
|
|
#include "Logger.hpp"
|
|
#include "Audio/AudioDevice.hpp"
|
|
|
|
#include "moc_soundout.cpp"
|
|
|
|
bool SoundOutput::checkStream () const
|
|
{
|
|
bool result {false};
|
|
|
|
Q_ASSERT_X (m_stream, "SoundOutput", "programming error");
|
|
if (m_stream) {
|
|
switch (m_stream->error ())
|
|
{
|
|
case QAudio::OpenError:
|
|
Q_EMIT error (tr ("An error opening the audio output device has occurred."));
|
|
break;
|
|
|
|
case QAudio::IOError:
|
|
Q_EMIT error (tr ("An error occurred during write to the audio output device."));
|
|
break;
|
|
|
|
case QAudio::UnderrunError:
|
|
Q_EMIT error (tr ("Audio data not being fed to the audio output device fast enough."));
|
|
break;
|
|
|
|
case QAudio::FatalError:
|
|
Q_EMIT error (tr ("Non-recoverable error, audio output device not usable at this time."));
|
|
break;
|
|
|
|
case QAudio::NoError:
|
|
result = true;
|
|
break;
|
|
}
|
|
}
|
|
return result;
|
|
}
|
|
|
|
void SoundOutput::setFormat (QAudioDeviceInfo const& device, unsigned channels, int frames_buffered)
|
|
{
|
|
Q_ASSERT (0 < channels && channels < 3);
|
|
m_device = device;
|
|
m_channels = channels;
|
|
m_framesBuffered = frames_buffered;
|
|
}
|
|
|
|
void SoundOutput::restart (QIODevice * source)
|
|
{
|
|
if (!m_device.isNull ())
|
|
{
|
|
QAudioFormat format (m_device.preferredFormat ());
|
|
// qDebug () << "Preferred audio output format:" << format;
|
|
format.setChannelCount (m_channels);
|
|
format.setCodec ("audio/pcm");
|
|
format.setSampleRate (48000);
|
|
format.setSampleType (QAudioFormat::SignedInt);
|
|
format.setSampleSize (16);
|
|
format.setByteOrder (QAudioFormat::Endian (QSysInfo::ByteOrder));
|
|
if (!format.isValid ())
|
|
{
|
|
Q_EMIT error (tr ("Requested output audio format is not valid."));
|
|
}
|
|
else if (!m_device.isFormatSupported (format))
|
|
{
|
|
Q_EMIT error (tr ("Requested output audio format is not supported on device."));
|
|
}
|
|
else
|
|
{
|
|
// qDebug () << "Selected audio output format:" << format;
|
|
m_stream.reset (new QAudioOutput (m_device, format));
|
|
checkStream ();
|
|
m_stream->setVolume (m_volume);
|
|
m_stream->setNotifyInterval(1000);
|
|
error_ = false;
|
|
|
|
connect (m_stream.data(), &QAudioOutput::stateChanged, this, &SoundOutput::handleStateChanged);
|
|
connect (m_stream.data(), &QAudioOutput::notify, [this] () {checkStream ();});
|
|
|
|
// qDebug() << "A" << m_volume << m_stream->notifyInterval();
|
|
}
|
|
}
|
|
if (!m_stream)
|
|
{
|
|
if (!error_)
|
|
{
|
|
error_ = true; // only signal error once
|
|
Q_EMIT error (tr ("No audio output device configured."));
|
|
}
|
|
return;
|
|
}
|
|
else
|
|
{
|
|
error_ = false;
|
|
}
|
|
|
|
// we have to set this before every start on the stream because the
|
|
// Windows implementation seems to forget the buffer size after a
|
|
// stop.
|
|
//qDebug () << "SoundOut default buffer size (bytes):" << m_stream->bufferSize () << "period size:" << m_stream->periodSize ();
|
|
if (m_framesBuffered > 0)
|
|
{
|
|
m_stream->setBufferSize (m_stream->format().bytesForFrames (m_framesBuffered));
|
|
}
|
|
m_stream->setCategory ("production");
|
|
m_stream->start (source);
|
|
LOG_DEBUG ("Selected buffer size (bytes): " << m_stream->bufferSize () << " period size: " << m_stream->periodSize ());
|
|
}
|
|
|
|
void SoundOutput::suspend ()
|
|
{
|
|
if (m_stream && QAudio::ActiveState == m_stream->state ())
|
|
{
|
|
m_stream->suspend ();
|
|
checkStream ();
|
|
}
|
|
}
|
|
|
|
void SoundOutput::resume ()
|
|
{
|
|
if (m_stream && QAudio::SuspendedState == m_stream->state ())
|
|
{
|
|
m_stream->resume ();
|
|
checkStream ();
|
|
}
|
|
}
|
|
|
|
void SoundOutput::reset ()
|
|
{
|
|
if (m_stream)
|
|
{
|
|
m_stream->reset ();
|
|
checkStream ();
|
|
}
|
|
}
|
|
|
|
void SoundOutput::stop ()
|
|
{
|
|
if (m_stream)
|
|
{
|
|
m_stream->reset ();
|
|
m_stream->stop ();
|
|
}
|
|
m_stream.reset ();
|
|
}
|
|
|
|
qreal SoundOutput::attenuation () const
|
|
{
|
|
return -(20. * qLn (m_volume) / qLn (10.));
|
|
}
|
|
|
|
void SoundOutput::setAttenuation (qreal a)
|
|
{
|
|
Q_ASSERT (0. <= a && a <= 999.);
|
|
m_volume = qPow(10.0, -a/20.0);
|
|
// qDebug () << "SoundOut: attn = " << a << ", vol = " << m_volume;
|
|
if (m_stream)
|
|
{
|
|
m_stream->setVolume (m_volume);
|
|
}
|
|
}
|
|
|
|
void SoundOutput::resetAttenuation ()
|
|
{
|
|
m_volume = 1.;
|
|
if (m_stream)
|
|
{
|
|
m_stream->setVolume (m_volume);
|
|
}
|
|
}
|
|
|
|
void SoundOutput::handleStateChanged (QAudio::State newState)
|
|
{
|
|
switch (newState)
|
|
{
|
|
case QAudio::IdleState:
|
|
Q_EMIT status (tr ("Idle"));
|
|
break;
|
|
|
|
case QAudio::ActiveState:
|
|
Q_EMIT status (tr ("Sending"));
|
|
break;
|
|
|
|
case QAudio::SuspendedState:
|
|
Q_EMIT status (tr ("Suspended"));
|
|
break;
|
|
|
|
#if QT_VERSION >= QT_VERSION_CHECK (5, 10, 0)
|
|
case QAudio::InterruptedState:
|
|
Q_EMIT status (tr ("Interrupted"));
|
|
break;
|
|
#endif
|
|
|
|
case QAudio::StoppedState:
|
|
if (!checkStream ())
|
|
{
|
|
Q_EMIT status (tr ("Error"));
|
|
}
|
|
else
|
|
{
|
|
Q_EMIT status (tr ("Stopped"));
|
|
}
|
|
break;
|
|
}
|
|
}
|