mp.tasks.audio.TextToSpeech

On-device speech synthesis via MediaPipe CalculatorGraph.

mp.tasks.audio.TextToSpeech(
    handle: ctypes.c_void_p
)

Args
handle	Opaque C pointer created by MpTextToSpeechCreate.

Methods

cancel

View source

cancel() -> None

Cancels any in-flight synthesis or streaming session.

close

View source

close() -> None

Closes the synthesizer and releases native C++ resources.

create_from_model_path

View source

@classmethod
create_from_model_path(
    model_path: str,
    delegate: mp.tasks.BaseOptions.Delegate = _BaseOptions.Delegate.CPU,
    default_voice_id: str = '',
    speaking_rate: float = 1.0
) -> 'TextToSpeech'

Creates a TextToSpeech instance from a .litertlm model path.

create_from_options

View source

@classmethod
create_from_options(
    options: mp.tasks.audio.TextToSpeechOptions
) -> 'TextToSpeech'

Creates a TextToSpeech instance from TextToSpeechOptions.

list_voices

View source

list_voices() -> list[VoiceInfo]

Lists all voices available in the loaded model bundle.

start_stream

View source

start_stream(
    initial_text: str = '',
    voice_id: Optional[str] = None,
    locale: Optional[str] = None,
    finish_input: bool = False
) -> 'TtsStream'

Starts an incremental or single-utterance streaming synthesis session.

Args
initial_text	Optional initial text fragment to enqueue immediately.
voice_id	Optional speaker voice ID override.
locale	Optional BCP-47 locale override.
finish_input	If False, additional tokens/words can be appended via TtsStream.push_text() followed by TtsStream.finish_input().

Returns
An active TtsStream handle supporting push_text(), finish_input(), and iteration over TtsAudioResult chunks.

synthesize

View source

synthesize(
    text: str, voice_id: Optional[str] = None, locale: Optional[str] = None
) -> mp.tasks.audio.TtsAudioResult

Synthesizes text synchronously and returns mono float32 PCM audio.

Args
text	Input text to synthesize.
voice_id	Optional speaker voice ID override; if None, falls back to default_voice_id configured in TextToSpeechOptions.
locale	Optional BCP-47 locale override; if None, falls back to default_locale configured in TextToSpeechOptions.

Returns
A TtsAudioResult containing the full synthesized audio waveform.

synthesize_stream

View source

synthesize_stream(
    text: str, voice_id: Optional[str] = None, locale: Optional[str] = None
) -> Iterator[mp.tasks.audio.TtsAudioResult]

Streams synthesized audio chunk-by-chunk.

Args
text	Input text to synthesize into clause-level audio chunks.
voice_id	Optional speaker voice ID override; if None, falls back to default_voice_id configured in TextToSpeechOptions.
locale	Optional BCP-47 locale override; if None, falls back to default_locale configured in TextToSpeechOptions.

Yields
TtsAudioResult chunks as each clause finishes acoustic + vocoder synthesis.

__enter__

View source

__enter__() -> 'TextToSpeech'

__exit__

View source

__exit__(
    exc_type, exc_val, exc_tb
) -> None