Skip to content
Audio

indicF5-best

исследовательский Telugu-TTS с голосом Swamiji

2 min read

Модель sw-voice/indicF5-best — исследовательский TTS для телугу на базе семейства F5. Это F5TTS_Base (~0,34 млрд параметров) с vocab IndicF5, дообученный под голос Swamiji: нужен референсный клип 10 с и его транскрипт. Язык — только телугу (te).

Ключевые особенности

  • Few-shot TTS: копирует тембр и темп по референсному клипу; в репозитории лежит reference/p3R1wZLty6A_029.wav (10 с).
  • Двухэтапный файнтюн: IndicF5 + 10 эпох на 402 клипах Navaratri, затем 1 эпоха на 83 ч телугу и ещё 5 эпох на Navaratri.
  • Обязателен float32: float16 превращает веса IndicF5-семейства в шум; для телугу нужен патч токенизатора.
  • Длинный текст F5 режет на фрагменты примерно по 12 с и склеивает с cross-fade 0.15 с.
  • Темп берётся из референса: при этом клипе speed=0.92 даёт темп Mahabharatam; 1.0 звучит примерно на 16% быстрее.

Пример

Фрагмент инференса; переменные model, vocoder, ref_audio и ref_text — из полного скрипта карточки.

import f5_tts.infer.utils_infer as ui
wav, sr, _ = ui.infer_process(ref_audio, ref_text, TEXT, model, vocoder, 'vocos',
                              nfe_step=16, cfg_strength=2,
                              sway_sampling_coef=-1, speed=0.92,
                              cross_fade_duration=0.15, device=device)
ПодробнееБенчмарки: voice similarity 96.6% от real vs real; CER/WER 1.84/8.4% на 10 ответах; в Telugu Q&A CER 1.25%, WER 5.6%.
Железо: MacBook Air M5 16 ГБ — около 2 мин на минуту речи при 16 steps; A100 — около 26 с при 32 steps; RTX 3090 — один вызов за раз и 1.7 с до первого аудио.
Ограничения: нужен референсный клип и транскрипт; float32; патч токенизатора для телугу; 32 steps — полное качество, 16 — быстрее.
Discuss Telegram 0 80

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.

    # sw-voice/indicF5-best
    # ===== 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 sw-voice/indicF5-best --local-dir ~/models/indicF5-best
    # The model is in ~/models. How to run it is in the model card (link above).
    Add more LoRAs to indicF5-best
  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 indicF5-best (quants, finetunes, LoRAs) will show up under the bell.

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

About the model

sw-voice/indicF5-best
Author
sw-voice
Category
Audio
Released
Posted

On NeyroLearn

Hugging Face

800

Ask AI about this model

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

Comments

Audio

Swanada

Swanada