API reference

`KittenTTS`

KittenTTS(
    model_name="KittenML/kitten-tts-nano-0.8",
    cache_dir=None,
    backend=None,
)

Argument	Type	Meaning
`model_name`	`str`	Hugging Face repo ID or a name within `KittenML`
`cache_dir`	`str \| None`	Optional Hugging Face cache root
`backend`	`"cpu" \| "cuda" \| "amd_gpu" \| None`	Requested ONNX execution provider

audio = tts.generate(text, voice="expr-voice-5-m", speed=1.0, clean_text=False)

Returns a NumPy array. Set clean_text=True for normal input containing non-spoken forms.

for audio_chunk in tts.generate_stream(text, voice="Bella", speed=1.0, clean_text=True):
    ...

Returns a generator of NumPy arrays, one per text chunk.

tts.generate_to_file(text, output_path, voice="Luna", speed=1.0, sample_rate=24_000)

Generates all audio, then writes it with SoundFile.

voices = tts.available_voices

Returns the eight display-name aliases.

Available both as a module function and instance method:

normalize_text(text, locale="en-US", return_spans=False)
tts.normalize_text(text, locale="en-US", return_spans=False)

When return_spans=False, returns str. When true, returns a NormalizedTextResult with .text and .spans.

get_model(repo_id, cache_dir=None, backend=None) constructs KittenTTS for older integrations. New code should use KittenTTS(...) directly.