Skip to content
Audio

Pocket TTS language deltas - multilingual delta packages for the speech synthesis engine

2 min readupdated

The repository contains language deltas (small weight changes relative to the base model) for the Kyutai Pocket TTS engine, used in the Frankenstein v2 engine of the mynah-s2s project. Each update is a delta for the English base model in two architectural variants (24-layer and 6-layer). The development makes it possible to adapt compact text-to-speech synthesis to different languages, including Russian, French, German, Spanish, and others. The set will be useful for developers of voice assistants and local speech recognition and generation systems.

Key features

  • Support for multiple languages, including Russian, German, French, Japanese, and Turkish.
  • Availability of variants with different quantization precision: int4, int8, and bf16.
  • Support for two architectural model families: 24-layer (24l) and 6-layer (6l).
  • Integration with the Frankenstein v2 engine of the mynah-s2s project.

What you can do

  • Integrate language packages into Pocket TTS-based projects for voiceover in Russian or other languages.
  • Adjust model precision and size to available resources using int4 or int8 formats.
  • Test multilingual sessions using the base Kyutai Pocket TTS weights.

Example

An example of using the package structure and manifests is described in the documentation of the mynah-s2s repository and Frankenstein v2 engine, which connect JSON language files and weight deltas for the selected language.

Read moreDetails
Model size: from 67.1 MB (int4, 6l) to 660.4 MB (bf16, 24l)
Backend: Kyutai Pocket TTS (base model terms agreement required)
License: CC-BY-4.0, CC-BY-SA-4.0, or MIT depending on the package
Discuss Telegram 1

Will it run?

Runs in BF16

Fits without quantisation.

  • BF16weights 0.6 GB · ready files1.1 GB
  • FP8 / INT8recommendedweights 0.3 GB0.8 GB
  • 4-bitweights 0.2 GB0.7 GB

Tip: go 8-bit.FP8 / INT8 is within about a percent of full precision in quality, takes half the memory and runs faster on RTX 40/50 and H100. Full precision only pays off for fine-tuning.

What to get for this model

Massed ComputeOpen Massed Compute

Your own GPU: FP8 / INT8 — from RTX 2060 · BF16 — from RTX 2060

Massed Compute hourly prices, November 2026. The discounted price applies with the coupon at checkout.

Estimate: weights plus working memory. Long videos, high resolutions and long contexts need more.

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.

    # gabrione/pocket-tts-deltas
    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 gabrione/pocket-tts-deltas --local-dir ~/models/pocket-tts-deltas
    # The model is in ~/models. How to run it is in the model card (link above).
    Add more LoRAs to pocket-tts-deltas
  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 pocket-tts-deltas (quants, finetunes, LoRAs) will show up under the bell.

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

About the model

gabrione/pocket-tts-deltas
Author
gabrione
Category
Audio
Released
Posted

On NeyroLearn

Hugging Face

01

Ask AI about this model

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

Comments