where possible audio devices that disappear are not forgotten until
the user selects another device, this should allow temporarily missing
devices or forgetting to switch on devices before starting WSJT-X to
be handled more cleanly. If all else fails, visiting the Settings
dialog and clicking OK should get things going again. Note that we
still do not have a reliable way of detecting failed audio out
devices, in that case selecting another device and then returning to
the original should work.
Enumerating audio devices is expensive and on Linux may take many
seconds per device. To avoid lengthy blocking behaviour until it is
absolutely necessary, audio devices are not enumerated until one of
the "Settings->Audio" device drop-down lists is opened. Elsewhere when
devices must be discovered the enumeration stops as soon as the
configured device is discovered. A status bar message is posted when
audio devices are being enumerated as a reminder that the UI may block
while this is happening.
The message box warning about unaccounted-for input audio samples now
only triggers when >5 seconds of audio appears to be missing or over
provided. Hopefully this will make the warning less annoying for those
that are using audio sources with high and/or variable latencies. A
status bar message is still posted for any amount of audio input
samples unaccounted for >1/5 second, this message appearing a lot
should be considered as notification that there is a problem with the
audio sub-system, system load is too high, or time synchronization is
stepping the PC clock rather than adjusting the frequency to maintain
monotonic clock ticks.
Fixed buffer sizes are used. Rx use s 3456 x 1st downsample rate x 5
audio frames of buffer space. On Windows this means that each
chunk (periodSize()) delivered from the audio stream is our initial
DSP processing chunk size, thus matching audio buffer latency exactly
with WSJT-X's own front end latency. This should result in optimal
resilience to high system loads that might starve the soundcard ADC of
buffers to fill and case dropped audio frames.
For Tx a buffer sufficient for 1 s of audio is used at present, on
Windows the period size will be set to 1/40 of that which gives
reasonably low latency and plenty of resilience to high system loads
that might starve the soundcard DAC of audio frames to render. Note
that a 1 s buffer will make the "Pwr" slider slow to respond, we may
have to reduce the Tx audio buffer size if this is seen as a problem.
This is a first cut at this to evaluate buffer size adjustments on
supported platforms. A final version might limit status bar reports to
>1000 dropped frames or similar.
This enforces an audio input device in the settings dialog since we
can't do anything without an input device. A nil audio output device
is allowed with a warning.
Adjusting these may help with audio drop-outs, particularly on slower
CPU systems or heavily loaded systems. Smaller buffer sizes leave less
margin for process interruptions, larger sizes waste resources that
could impact other processes.
