Skip to content
Audio

Whiscribe Diarization

ONNX-модели диаризации речи для Android

2 min readupdated

Две ONNX-модели от s1lvrn0 для Android-приложения Whiscribe, которое транскрибирует аудио на телефоне. Они определяют, кто и когда говорит: сегментация ищет речи в 10-секундных окнах до тр��х спикеров, а эмбеддинг превращает голос в 192 числа для сравнения. Всё работает на CPU provider ONNX Runtime.

Ключевые особенности

  • pyannote segmentation 3.0 ONNX: вход [batch, 1, 160000] — 10 с при 16 kHz; выход [batch, 589, 7] — 7 powerset-классов: никто, спикеры 1–3 и их пары.
  • ReDimNet2-B3: вход wav [1, samples] 16 kHz mono в диапазоне [-1, 1] любой длины; выход emb [1, 192], голоса сравниваются по cosine similarity.
  • Обе модели — ONNX Runtime CPU: размеры 5 992 913 и 18 041 088 байт, без GPU.
  • Пайплайн Whiscribe: сегментация окнами 10 с каждые 2,5 с, эмбеддинг на участках, где говорит только один спикер, average-linkage clustering по cosine distance с порогом 0,6 и слияние мелких кластеров.
  • Модели не дообучались: pyannote 3.0 — неизменённая ONNX-конверсия из sherpa-onnx, ReDimNet2-B3 обучен только на VoxCeleb2; репозиторий под MIT.

Пример

Экспорт ReDimNet2 в ONNX из карточки:

python export_redimnet2_onnx.py path/to/redimnet2 b3-vox2-lm.pt redimnet2-b3-vox2-lm.onnx
ПодробнееСегментация: 5 992 913 байт; эмбеддинг: 18 041 088 байт. Обе модели — ONNX Runtime CPU.
pyannote 3.0 — неизменённая ONNX-конверсия из sherpa-onnx; ReDimNet2-B3 обучен только на VoxCeleb2 и экспортирован в opset 17 без дообучения.
Замеры в карточке — на AMI и выдуманных диалогах реальных голосов на восьми языках, без публикации метрик.
Ограничение сегментации — до трёх говорящих в окне; эмбеддинг принимает аудио любой длины, приложение использует от 0,5 с.
Discuss Telegram 0

Will it run?

Runs in BF16

Fits without quantisation.

  • BF16weights 0.0 GB · ready files0.5 GB
  • FP8 / INT8recommendedweights 0.0 GB0.5 GB
  • 4-bitweights 0.0 GB0.5 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.

    # s1lvrn0/whiscribe-diarization
    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 s1lvrn0/whiscribe-diarization --local-dir ~/models/whiscribe-diarization
    # The model is in ~/models. How to run it is in the model card (link above).
    Add more LoRAs to whiscribe-diarization
  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 whiscribe-diarization (quants, finetunes, LoRAs) will show up under the bell.

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

About the model

s1lvrn0/whiscribe-diarization
Author
s1lvrn0
Category
Audio
Released
Posted

On NeyroLearn

Hugging Face

00

Ask AI about this model

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

Comments

Audio

Dragonfly

Dragonfly