Skip to content
Other

WeSpeaker ResNet221-LM

Core ML для эмбеддингов диктора

1 min read

Это Core ML-конверсия модели WeSpeaker ResNet221-LM от GoodnotesLtd, основанной на весах VoxCeleb2. Модель извлекает 256-мерные эмбеддинги диктора из 16 кГц mono-аудио и рассчитана на фиксированные 10-секундные окна. Артефакты включают NeuralNetwork с fp16-весами, fp32 I/O и динамический ONNX-граф.

Ключевые особенности

  • 256-мерные эмбеддинги; перед сравнением по cosine применяйте L2-нормализацию.
  • Core ML NeuralNetwork: входы fbank_features[1,1,80,998] и weights[1,589], выходы normalized/raw embedding[1,256].
  • 16 кГц mono, kaldi-fbank80-v1: 80-bin, 25/10 мс, Hamming, без dither, per-mel CMN, PCM16.
  • Есть динамический ONNX opset 14 и статический unmasked ONNX; Core ML — отдельная фиксированная ветка.
  • Хранимые веса — fp16, входы/выходы и pooling-арифметика — fp32; на macOS используется CPUAndNeuralEngine.

Пример

Сценарий: на macOS или iOS подать 10-секундный фрагмент 16 кГц mono, посчитать 80-мерный fbank, получить normalized embedding[1,256] или L2-нормализовать raw-выход и сравнить с эталоном по cosine.

ПодробнееБенчмарки: NN vs PyTorch — min/median cosine 0.999878/0.999977; три gate-проверки конверсии пройдены (≥200 клипов, cosine ≥0.999).
Core ML vs dynamic ONNX на 70 сегментах — min/mean cosine 0.890734/0.987775; минимум на 1 с не проходит порог 0.95.
На M2 Max warm median RTF: Core ML 0.005877 против dynamic ONNX CPU-2 0.230177.
Ограничения: фиксированные 10 с, Windows/iOS/Android локально не проверялись; iOS runtime не квалифицирован сборкой под macOS. Лицензия upstream — Apache-2.0 с оговоркой по атрибуции VoxCeleb2.
Discuss Telegram 0 11

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.

    # GoodnotesLtd/wespeaker-resnet221-lm-coreml
    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 GoodnotesLtd/wespeaker-resnet221-lm-coreml --local-dir ~/models/wespeaker-resnet221-lm-coreml
    # 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

New posts about versions of wespeaker-resnet221-lm-coreml (quants, finetunes, LoRAs) will show up under the bell.

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

About the model

GoodnotesLtd/wespeaker-resnet221-lm-coreml
Author
GoodnotesLtd
Category
Other
Released
Posted

On NeyroLearn

Hugging Face

110

Ask AI about this model

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

Comments

Other

CiteNet

CiteNet