When code needs to work with audio device info, it keeps index, max_input_channels, max_output_channels, default_sample_rate here instead of scattering it across globals. Initialization establishes the handles and sizes; teardown must release any allocation, lock, thread, or callback context recorded in those fields. More...

#include <include/ascii-chat/audio/audio.h>

Data Fields
PaStream *	duplex_stream
	PortAudio full-duplex stream (simultaneous input+output).
PaStream *	input_stream
	Separate input stream (when full-duplex unavailable).
PaStream *	output_stream
	Separate output stream (when full-duplex unavailable).
audio_ring_buffer_t *	raw_capture_rb
	Callback → Worker: Raw mic samples (no processing).
audio_ring_buffer_t *	raw_render_rb
	Callback → Worker: Raw speaker samples (AEC3 reference).
audio_ring_buffer_t *	processed_playback_rb
	Worker → Callback: Ready for speakers.
audio_ring_buffer_t *	capture_buffer
	OLD: Worker → Network (will become processed_capture_rb).
audio_ring_buffer_t *	playback_buffer
	OLD: Network → Worker (will become network_playback_rb).
audio_ring_buffer_t *	render_buffer
	Ring buffer for render reference (separate streams mode).
platform_thread_t	worker_thread
	Worker thread for audio processing.
mutex_t	audio_processing_wait_guard
	Coordinates device-worker sleep with shutdown.
cond_t	audio_processing_wakeup
	Capture/AEC work, monitor refill, or shutdown notification.
bool	worker_running
	True if worker thread is running.
atomic_t	worker_should_stop
	Signal worker thread to exit.
float *	worker_capture_batch
	Pre-allocated buffer for batch capture processing.
float *	worker_render_batch
	Pre-allocated buffer for batch render processing.
float *	worker_playback_batch
	Pre-allocated buffer for batch playback processing.
bool	initialized
	True if context has been initialized.
bool	running
	True if duplex stream is active.
bool	separate_streams
	True if using separate input/output streams.
bool	playback_only
	True for playback-only mode (mirror), skip microphone capture.
atomic_t	shutting_down
	True when shutdown started - callback outputs silence.
mutex_t	state_mutex
	Mutex protecting context state.
void *	audio_pipeline
	Client audio pipeline for AEC3 echo cancellation (opaque pointer).
bool	monitor_local_media
	Play local file audio in solo sessions.
uint64_t	test_audio_playback_position
	Output callback's synthesized monitor sample clock.
void *	capture_media_source
	Borrowed local media audio; configure before starting duplex.
double	sample_rate
	Actual sample rate of streams (48kHz).
unsigned	input_device_channels
unsigned	output_device_channels
audio_resampler_t	capture_resampler
audio_resampler_t	playback_resampler
double	input_device_rate
	Native sample rate of input device.
double	output_device_rate
	Native sample rate of output device.
double	input_resample_phase
	Source position carried between native-rate input blocks.
float	input_resample_previous
	Last input sample used at a block boundary.
bool	input_resample_has_previous
	Whether a prior input block has been processed.
double	output_resample_phase
	48 kHz source position carried between output callbacks
float	output_resample_previous
	Last source sample used at an output boundary.
bool	output_resample_has_previous
	Whether a prior output block has been processed.

Detailed Description

When code needs to work with audio device info, it keeps index, max_input_channels, max_output_channels, default_sample_rate here instead of scattering it across globals. Initialization establishes the handles and sizes; teardown must release any allocation, lock, thread, or callback context recorded in those fields.

Audio context for full-duplex capture and playback

Manages a single PortAudio full-duplex stream for simultaneous audio capture and playback. Uses a single callback that receives both render (speaker output) and capture (microphone input) at the EXACT same instant, enabling perfect timing for WebRTC AEC3 echo cancellation.

WORKER THREAD ARCHITECTURE: The audio system now uses a dedicated worker thread to move ALL heavy processing (AEC3 and RMS calculations) out of the real-time PortAudio callbacks. This ensures callbacks complete in <2ms (was 50-80ms).

Architecture:

PortAudio Callback (Real-Time, <2ms): Lock-free ring buffer copies only
Audio callbacks: Bounded stateful sample-rate conversion when device rates differ
Worker Thread (Non-Real-Time): Heavy processing (AEC3, filters)

Ring Buffers:

raw_capture_rb: Callback → Worker (raw mic samples)
raw_render_rb: Callback → Worker (raw speaker samples for AEC3 reference)
processed_capture_rb: Worker → Network (after AEC3/filters, ready for encoder)
processed_playback_rb: Worker → Callback (ready for speakers)
network_playback_rb: Network → Worker (decoded Opus from network)

Note: The audio context must be initialized with audio_init() before use.; Use audio_start_duplex()/audio_stop_duplex() to control the stream.; State is protected by state_mutex for thread-safe operations.

Definition at line 117 of file include/ascii-chat/audio/audio.h.