Skip to content

output-and-files.md

Latest commit

 

History

History
98 lines (68 loc) · 2.33 KB

output-and-files.md

File metadata and controls

98 lines (68 loc) · 2.33 KB
Raw

Output And Files

RealtimeTTS can play audio locally, write a WAV file, or pass raw chunks to your application.

Local Playback

PCM engines use PyAudio and output_device_index:

stream = TextToAudioStream(engine, output_device_index=3)
stream.feed("Speak through a selected PyAudio device.")
stream.play()

Use PyAudio device inspection tools or a small local script to find device indices for your machine.

Compressed-output engines use mpv. Examples include Edge, ElevenLabs, OpenAI when using MP3, MiniMax, and ModelsLab.

stream = TextToAudioStream(engine, mpv_audio_device="auto")

Run this in a terminal to list mpv audio devices:

mpv --audio-device=help

Muted Mode

Muted mode disables local speaker playback but still lets you write files or process chunks.

stream = TextToAudioStream(engine, muted=True)
stream.feed("Generate audio without playing it.")
stream.play(output_wavfile="speech.wav")

You can also pass muted=True to a single play() call.

WAV Output

stream.feed("Save this line.")
stream.play(output_wavfile="speech.wav", muted=True)

The output file is opened when playback begins and finalized when playback finishes. For long-running applications, create a new output path per response.

Audio Chunk Callback

Use on_audio_chunk when another process needs audio bytes immediately.

def handle_chunk(chunk: bytes):
    print("chunk bytes:", len(chunk))


stream.feed("Send chunks to my application.")
stream.play(on_audio_chunk=handle_chunk, muted=True)

Chunk format depends on the engine. Use engine.get_stream_info() to inspect the active engine's format, channel count, and sample rate.

Buffer And Chunk Tuning

TextToAudioStream accepts audio playback sizing controls:

stream = TextToAudioStream(
    engine,
    frames_per_buffer=1024,
    playout_chunk_size=-1,
)

Smaller chunks can reduce latency but increase CPU overhead. Larger chunks can smooth playback but add delay. Keep defaults until you have a concrete latency or stuttering problem.

Volume

The stream exposes clamped volume control from 0.0 to 1.0:

stream.volume = 0.7
# or
stream.set_volume(0.7)

Volume scaling applies inside the playback path. Some compressed or external playout behavior can still depend on the engine and mpv path.