Skip to content
Audio

Bayan-Arabic-TTS-v1.0

арабский TTS с голосом «Мурад»

1 min read

Bayan от anwar-dev — арабская TTS-модель для современного литературного арабского (MSA) на базе nvidia/magpie_tts_multilingual_357m. Дообучена через LoRA и отдельный speaker-ряд под мужской голос «Мурад» на 71,6 минуты арабской речи (633 фрагмента).

Ключевые особенности

  • LoRA r=32, α=64, dropout=0.05 на 133 слоях + шестой speaker-ряд; обучено 13,91M параметров.
  • Архитектура Transformer decoder + NanoCodec: 22,05 kHz, 21,5 fps.
  • Файл модели .nemo весит 1,36 GiB; inference на RX 6700 XT 12 GB через ROCm, RTF ≈ 1,0.
  • На 32 отложенных фрагментах: CER 1,39%, WER 8,12%, сходство голоса 0,869.
  • На 32 фразах MSA: CER 1,60% (0,10% после нормализации чисел), WER 2,02%, точных совпадений 28/32.

Пример

from nemo.collections.tts.models import MagpieTTSModel
model = MagpieTTSModel.restore_from("Bayan-Arabic-TTS-v1.0.nemo", map_location="cuda", strict=False).eval()
model.inference_parameters.cfg_scale = 2.5
model.inference_parameters.temperature = 0.6
audio, audio_len = model.do_tts("إِنَّ الْعِلْمَ نُورٌ.", language="ar-MSA", speaker_index=5)
# speaker_index=5 — голос «Мурад»
ПодробнееБенчмарки: CER 1,39%, WER 8,12%, сходство голоса 0,869 на 32 отложенных фрагментах.
Обучение: 47 эпох (лучший результат на 40-й), ~85 с/эпоха, пик VRAM 7,9 GB на AMD RX 6700 XT 12 GB, ROCm 6.3, fp32.
Ограничения: лучше работает с полной огласовкой; максимум ~37 с на фразу; 22,05 kHz моно; данные синтетические из одного TTS-движка.
Лицензия: NVIDIA Open Model License.
Discuss Telegram 2 317

Will it run?

Working out the memory needs

This is a LoRA: it only runs on top of the base model nvidia/magpie_tts_multilingual_357m. The base model's size is still being collected; the estimate will appear within a day.

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.

    # anwar-dev/Bayan-Arabic-TTS-v1.0
    # ===== 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 nvidia/magpie_tts_multilingual_357m --local-dir ~/models/magpie_tts_multilingual_357m
    hf download anwar-dev/Bayan-Arabic-TTS-v1.0 --local-dir ~/models/Bayan-Arabic-TTS-v1.0
    # The model is in ~/models. How to run it is in the model card (link above).
  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

Requires the base model

nvidia/magpie_tts_multilingual_357m

This model is a LoRA / adapter. Download the base model first, then load this one on top.

New posts about versions of magpie_tts_multilingual_357m (quants, finetunes, LoRAs) will show up under the bell.

Quantised versions 5All on Hugging Face
Fine-tunes 12All on Hugging Face
LoRAs and adapters 1All on Hugging Face

About the model

anwar-dev/Bayan-Arabic-TTS-v1.0
Author
anwar-dev
Category
Audio
Released
Posted

On NeyroLearn

Hugging Face

3172

Ask AI about this model

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

Comments