Recording

The managed capture API collects audio in memory while your application does other work. Native capture buffers are drained on a background thread, so the common API returns one PCM value instead of exposing chunks:

from pyalsoft import start_recording, stop_recording

recording = start_recording()
input("Speak now, then press Enter to stop... ")
captured = stop_recording(recording)

start_recording() uses the default input device and records 48 kHz, mono, 16-bit PCM unless told otherwise. Use list_capture_devices() to select a specific input. The bundled OpenAL Soft runtime accepts unsigned 8-bit, signed 16-bit, and float32 capture in mono, stereo, quad, 5.1, 6.1, and 7.1 layouts; the selected capture backend or hardware can still reject a requested combination. For a known duration, record() is the blocking equivalent.

Bounded incremental capture

Long-running capture and real-time analysis should use a bounded CaptureStream. Readers consume frames incrementally; if they fall behind, the oldest frames are discarded and reported through get_capture_stream_status():

from pyalsoft import (
    get_capture_stream_status,
    read_capture_stream,
    start_capture_stream,
)

with start_capture_stream(capacity_frames=48_000) as stream:
    while should_continue:
        chunk = read_capture_stream(stream, max_frames=4_800, timeout=0.25)
        if chunk is not None:
            process(chunk)

    print("dropped frames:", get_capture_stream_status(stream).overrun_count)

The default capacity is one second at the default sample rate. This API never grows its managed buffer beyond capacity_frames; the original Recording API continues to retain every frame for callers that want one complete PCM value.

Captured and generated PCM values can be passed directly to play():

from pyalsoft import play

sound = play(captured)

See the runnable record_and_play.py example for the complete flow.