IndexTTS-1.5-MLX

IndexTeam/IndexTTS-1.5 converted to MLX safetensors (fp32) for native Apple Silicon inference, with mlx-indextts.

IndexTTS-1.5 is a zero-shot voice-cloning TTS system for Chinese and English. It clones a voice from a single short reference clip (5–15 s).

Model files

File	Description
gpt.safetensors	GPT semantic-code generator (autoregressive)
bigvgan.safetensors	BigVGAN vocoder
dvae.pth	Discrete VAE (kept in PyTorch format, loaded by the runtime)
tokenizer.model / unigram_12000.vocab	Text front end
config.json	Runtime config

Use with mlx-indextts

git clone https://github.com/Jup33Q/mlx-indextts
cd mlx-indextts && uv sync

hf download Jup33QE/IndexTTS-1.5-MLX --local-dir models/IndexTTS-1.5-MLX

Generation with this package has not been end-to-end verified yet — the weights were converted with mlx_indextts/convert.py and pass structural checks, but only IndexTTS-2 has a verified synthesis path at the moment. Status updates are tracked in the inference repo. For the fully verified package, see IndexTTS-2-MLX.

License

IndexTTS-1.5 is released under the Apache-2.0 license by the upstream authors. The same terms apply to this converted copy.