from mlx_audio.tts.utils import load_model
import soundfile as sf

model = load_model("mlx-community/kitten-tts-nano-0.8")

# The generator yields one segment at a time.
for out in model.generate("This model runs with MLX.", voice="Jasper"):
    sf.write("out.wav", out.audio, model.sample_rate)

Notes:

phonemizer (espeak backend) is required for generate().
Available voices: Bella, Jasper, Luna, Bruno, Rosie, Hugo, Kiki, Leo.

Conversion Details

Converted from ONNX weights.
Output shapes and durations match the reference ONNX model for test inputs.