git clone https://github.com/gafiatulin/tada-mlx && cd tada-mlx
uv sync

# Encode a voice
uv run tada-mlx --encode-voice ref.wav --voice-text "transcript" --save-voice voice.safetensors

# Synthesize
uv run tada-mlx --text "Hello!" --prompt voice.safetensors --model gafiatulin/tada-1b-mlx

Or as a library:

from tada_mlx import load_model, generate, InferenceOptions

Requires gafiatulin/tada-codec-mlx for the decoder/encoder (downloaded automatically).