Skip to content
Audio

Matoub-4B

TTS для кабильского языка с клонированием голоса

2 min readupdated

Matoub-4B — 4B-параметрическая TTS-модель для кабильского языка (Taqbaylit, kab, латиница) от agbalu. Обучена на 384 960 парах речи из Common Voice Kabyle и синтезирует аудио 24 kHz. Голос задаётся короткой референсной записью и её точным транскриптом; в основе — bosonai/higgs-tts-3-4b с 8-codebook аудиокодеком.

Ключевые особенности

  • Голос клонируется по короткой референсной записи; транскрипт должен точно совпадать с произнесённым и оставаться на исходном языке.
  • Выход — моно-аудио 24 kHz; 8-codebook кодек передаёт фонологию кабильского, включая спирантизированные, фарингализованные, эмфатические и геминированные согласные.
  • Загрузка через Transformers AutoModelForTextToWaveform с trust_remote_code=True.
  • Обучение — 384 960 пар из Common Voice 26.0 Kabyle; validation 29 812, test 5 871.
  • NF4-квантизация снижает пик памяти с 9.94 GB до 4.57 GB, но на замерах ухудшает произношение и завершение относительно BF16.
  • Лицензия — только некоммерческое исследовательское использование, как у базовой модели Boson Higgs TTS 3.

Пример

import soundfile as sf
from transformers import AutoModelForTextToWaveform
model = AutoModelForTextToWaveform.from_pretrained(
    "agbalu/Matoub-4B", trust_remote_code=True
).to("cuda").eval()
speech = model.generate(text="Azul fell-awen!", reference_audio="voice.wav", reference_text="exact transcript")
sf.write("speech.wav", speech.waveform[0].numpy(), speech.sampling_rate)
ПодробнееВалидация: 16 utterances, 6 speakers; 15/16 завершились; средний voice cosine 0.881; CER excess 0.045.
Память: BF16 — 9.94 GB, NF4 — 4.57 GB (H100 MIG, single-candidate); NF4 может ухудшать произношение и завершение.
Обучение: rank-32 LoRA (alpha 64), 12 030 шагов, final loss 3.8997 nats.
Ограничения: только kab; Tarifit и Tashelhit не покрыты; имена и заимствования могут требовать правок; не для биометрии, имитации личности и safety-critical задач.
Discuss Telegram 1 39

Will it run?

Runs in BF16

Fits without quantisation.

  • BF16weights 8.2 GB · ready files9.5 GB
  • FP8 / INT8recommendedweights 4.0 GB4.9 GB
  • 4-bitweights 2.3 GB3.0 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 3080

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.

    # agbalu/Matoub-4B
    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 agbalu/Matoub-4B --local-dir ~/models/Matoub-4B
    # The model is in ~/models. How to run it is in the model card (link above).
    Add more LoRAs to Matoub-4B
  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 Matoub-4B (quants, finetunes, LoRAs) will show up under the bell.

Quantised versions 1All on Hugging Face

About the model

agbalu/Matoub-4B
Author
agbalu
Category
Audio
Released
Posted

On NeyroLearn

Hugging Face

391

Ask AI about this model

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

Comments

Audio

Swanada

Swanada