melotts-mblt

Run MeloTTS text-to-speech on Mobilint NPUs. melotts-mblt ships pre-quantized MeloTTS synthesizers for English and Korean. It provides a Python API, a command line, and a Gradio WebUI. The acoustic model runs on the NPU, and its Mobilint BERT prosody encoder loads through transformers-mblt.

Models run on Mobilint ARIES and REGULUS boards. Supported target-device identifiers are aries-rb, regulus-ra, regulus-rb, regulus-ra-usb, and regulus-rb-usb.

Version 0.0.0 is the first standalone release. It was extracted from mblt-model-zoo 2.11.0.

Installation

pip install melotts-mblt
# One-time download of the NLTK tagger (English G2P) and the UniDic dictionary
melotts-mblt download

NPU execution requires a supported Mobilint NPU driver and device. Python packaging cannot run post-install steps, so run melotts-mblt download once after installing.

Quick start

from melotts_mblt import TTS

model = TTS(language="EN_NEWEST", device="cpu", trust_remote_code=True)
try:
    speaker_ids = model.hps.data.spk2id
    model.tts_to_file("Did you ever hear a folk tale about a giant turtle?", speaker_ids["EN-Newest"], "en.wav", speed=1.0)
finally:
    model.dispose()  # release the NPU backends

trust_remote_code=True lets the Mobilint BERT prosody encoder load its Hub remote code; the melotts-mblt CLI and WebUI pass it for you. Use language="KR" with speaker "KR" for Korean. Each language downloads its model from the Mobilint Hugging Face organization: mobilint/MeloTTS-English-v3 and mobilint/MeloTTS-Korean.

These TTS(...) arguments select the NPU: dev_no, target_core, target_device, encoder_mxq_path, and decoder_mxq_path. The API reference has more examples.

Command line

The melotts-mblt command provides three subcommands:

• tts synthesizes speech with the MeloTTS CLI.
• ui launches the Gradio WebUI.
• download fetches the NLTK and UniDic resources.

melotts-mblt tts "Text to read" output.wav --language EN_NEWEST --speed 1.2
melotts-mblt tts "text-to-speech 안녕하세요" kr.wav --language KR
melotts-mblt tts file.txt out.wav --file
melotts-mblt ui --host 0.0.0.0 --port 7860
melotts-mblt download

python -m melotts_mblt.cli is equivalent to melotts-mblt. Run melotts-mblt tts --help for every TTS option.

Using with mblt-model-zoo

This package replaces mblt_model_zoo.MeloTTS. The modules have the same contents, and only the import prefix and the command names differ:

Development

Use uv to manage the development environment:

git clone https://github.com/mobilint/MeloTTS-mblt.git
cd MeloTTS-mblt
uv venv --python 3.12
source .venv/bin/activate
uv pip install -e . --group dev
pre-commit install

The test guide explains the hardware-free and NPU test runs.

Support and issues

For installation, model, or runtime support, visit the Mobilint forum or contact tech-support@mobilint.com. Report reproducible package issues in the MeloTTS-mblt issue tracker.

License

Mobilint's code is distributed under the BSD 3-Clause License. The MeloTTS-derived code in melotts_mblt remains under MyShell.ai's MIT License, and that license ships in the wheel. MeloTTS was created by Wenliang Zhao, Xumin Yu, and Zengyi Qin; see the API reference for the original authors and the citation.