Skip to content
Audio

Visteon MIO TTS en-jp 0.6B

потоковая озвучка для Jetson Orin

1 min readupdated

Visteon MIO TTS en-jp 0.6B — production-модель синтеза речи для английского и японского, рассчитанная на on-device запуск на NVIDIA Jetson Orin. Основа — 0.6B decoder-only LLM, которая выдаёт 8 параллельных акустических кодбуков на 24 кГц; акустика уточняется TensorRT fp16-движком, аудио декодирует нейросетевой вокодер на ONNX Runtime.

Ключевые особенности

  • Потоковый режим: TTFA 608 мс mean / 875 мс p95 на 20k предложений; RTF 0.157 mean / 0.189 p95 на Orin в fp16.
  • Zero-shot voice design: голос задаётся reference-audio промптом в .bin и выбирается на каждый запрос.
  • Проверка разборчивости: ASR round-trip на 20k предложений и 5k ответах ассистента; в аудите не нашлось строк, где модель не произнесла входной текст.
  • Детерминированность: seed на запрос, стабильность после перезапусков; fast-start разбиение предложений снижает задержку первого чанка.
  • Автоопределение языка en/ja, нормализация чисел, единиц и URL, обработка смешанных скриптов.

Пример

tar xzf visteon-mio-tts-runtime.tar.gz
cd visteon-mio-tts-runtime
./tts_inference no_bundle.trtfb assets_empty 18892 --mio-only &
ПодробнееTTFA на 5k ответах ассистента: p50 692 мс, p95 2166 мс; RTF ~0.5 при трёх инстансах.
Железо: Jetson Orin 64 GB, fp16; ауд��о 24 кГц mono PCM16.
Голосовой промпт — 5–15 с чистой речи с дословной транскрипцией.
Не медицинское/защитное устройство; синтетическую речь нужно маркировать.
Discuss Telegram 0 8

Will it run?

Working out the memory needs

The bot is still collecting this model's weight sizes; the estimate will appear here in a few minutes. Meanwhile you can rent a GPU with headroom.

Massed ComputeOpen Massed Compute

Run it on Massed Compute

The script downloads the model to the rented machine. How to run it is in the model card.

  1. 1. Rent a machine with a fitting GPU

    Open Massed Compute
  2. 2. Paste the script into the machine's terminal

    Connect over SSH (the Massed Compute panel shows the command) or open a terminal on the machine's desktop.

    HF_TOKEN / CIVITAI_TOKENThe script starts with token lines: paste yours between the quotes. The Hugging Face token is for gated models, the Civitai one for Civitai models.

    # visteon/mio-tts-en-jp-0.6b
    # ===== Tokens: paste yours between the quotes =====
    export HF_TOKEN=""        # huggingface.co/settings/tokens, for gated models
    # ==========================================
    mkdir -p ~/models
    curl -LsSf https://astral.sh/uv/install.sh | sh
    source "$HOME/.local/bin/env"
    uv tool install "huggingface_hub[cli]"
    hf download visteon/mio-tts-en-jp-0.6b --local-dir ~/models/mio-tts-en-jp-0.6b
    # The model is in ~/models. How to run it is in the model card (link above).
    Add more LoRAs to mio-tts-en-jp-0.6b
  3. 3. Use it

    The model is in ~/models. How to run it is in the model card (link above).

Stop the machine in the panel when you are done: it bills by the minute while it runs.

How to speed it up

Picked for this model type. Choose your hardware above to see what fits you exactly.

  • FlashAttention 44.0.0b40from November 18, 2026new release

    fastest exact attention on B200

    Kernels for Blackwell (beta). Exact attention without quantisation — fine for everything, LLMs included.

    How to enable: pip install flash-attn-4; vLLM and SGLang pick it up themselves.

    Earlier generations (2)
    • FlashAttention 3 — The Hopper (H100/H200) version with FP8.
    • FlashAttention 2 — The standard for RTX 30/40/50 and A100: exact attention without the memory cost of long contexts.
  • torch.compile2.15.0from October 28, 2026new release

    10–30% for free

    Compiles the model for your GPU. Slower first run, faster afterwards.

    How to enable: model = torch.compile(model); in ComfyUI the TorchCompileModel node.

Model family

New posts about versions of mio-tts-en-jp-0.6b (quants, finetunes, LoRAs) will show up under the bell.

No known relatives of this model yet. All derivatives on Hugging Face

About the model

visteon/mio-tts-en-jp-0.6b
Author
visteon
Category
Audio
Released
Posted

On NeyroLearn

Hugging Face

80

Ask AI about this model

Answered by Gemini 4 · the chat is kept only in this tab

Comments