Skip to content
Other

TitaNet-S Optimized ONNX

эмбеддинги диктора прямо на CPU

2 min read

Два FP32-энкодера голоса на базе NVIDIA NeMo TitaNet-S: один — оригинальная модель с оптимизациями под CPU, второй — дообученный под короткие фразы. Каждый возвращает один вектор размерности 192 на весь клип. Для инференса нужны только NumPy, ONNX Runtime и SoundFile — PyTorch и NeMo не требуются.

Ключевые особенности

  • Два варианта: titanet-s-optimized.onnx (lossless-экспорт) и titanet-s-tuned.onnx (файнтюн под верификацию короткой речи).
  • Встроенный waveform-фронтенд, на входе — mono 16 kHz float32 PCM.
  • Без квантизации, прунинга и временного субсэмплинга: сохранены все временные фреймы и все 3072 финальных канала.
  • Есть версии для Core ML (FP32/FP16) с нативным инференсом на Swift и отчётами по качеству и таймингам.
  • Размер каждого графа — 23,84 МБ, ONNX opset 17.

Пример

from inference import SpeakerEmbedder, load_audio

encoder = SpeakerEmbedder("titanet-s-tuned.onnx", threads=4)
enrollment = encoder(load_audio("enroll.wav"))
query = encoder(load_audio("query.wav"))
similarity = float(query @ enrollment)
Подробнее Задержка на Ryzen 9 5950X в WSL2, 4 потока: 1 с аудио — 2,65–3,00 мс, 5 с — 11,30 мс.
EER на LibriSpeech при 5 с: 0,000% у optimized и tuned против 0,100% у B0.
Tuned не хуже B0 в 19 из 20 ячеек; качество меряли на запросах длиной 1–5 с.
Файнтюн на VoxCeleb, Common Voice и LibriSpeech, лицензия Apache-2.0.
VAD, downmixing и ресемплинг нужны заранее, порог сходства подбирается на своих данных.
Discuss Telegram 0 10

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.

    # notaneimu/titanet-s-optimized-onnx
    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 notaneimu/titanet-s-optimized-onnx --local-dir ~/models/titanet-s-optimized-onnx
    # 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 titanet-s-optimized-onnx (quants, finetunes, LoRAs) will show up under the bell.

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

About the model

notaneimu/titanet-s-optimized-onnx
Author
notaneimu
Category
Other
Released
Posted

On NeyroLearn

Hugging Face

100

Ask AI about this model

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

Comments

Other

AISniff

AISniff