Skip to content
Audio

Before the Implant

крошечный causal-энхансер речи

2 min read

Before the Implant — компактная строго causal модель очистки речи от bmarti44 (Brian Martin). Это waveform-энхансер в стиле Conv-TasNet (CausalTasNetEnh) на 602 510 параметров, 16 kHz mono, созданный для research-сценария кохлеарных имплантов: выделить нужный голос в комнате с толпой. Модель работает без заглядывания вперёд и заявлена с алгоритмической задержкой 5.88 ms.

Ключевые особенности

  • Строго causal архитектура, zero lookahead; причинность проверена suffix-invariance тестом.
  • Всего 602 510 параметров, вход 16 kHz mono.
  • 5.88 ms алгоритмической задержки без учёта вычислений.
  • Обучение: 24 000 шагов на смесях LibriSpeech train-clean-100 — целевой голос плюс толпа, speech-shaped шум или тишина, −5…+10 dB SNR.
  • В карточке есть erratum: default и seed19 выдают инвертированную волну, у весов есть DC offset; исправление кода пока не опубликовано.

Пример

import sys, soundfile as sf
from huggingface_hub import snapshot_download
d = snapshot_download('bmarti44/before-the-implant-causal-enhancer')
sys.path.insert(0, d)
from enhance import load_model, enhance
model = load_model(d + '/model.safetensors')
audio, sr = sf.read('noisy.wav')
sf.write('enhanced.wav', enhance(audio, sr, model=model), sr)
ПодробнееБенчмарк: Crowd ΔSI-SDR +3.51 dB, ΔSTOI +0.035; speech-shaped noise +6.45 dB; 13 из 14 проверок scorecard пройдены.
Слабый случай: Crowd @ −5 dB даёт +1.19 dB при цели +1.5 dB.
Erratum: sign/DC offset затрагивают выход, но не метрики; веса не менялись, кандидатное исправление протестировано, но не опубликовано.
Ограничения: только синтетические смеси; живые слушатели и импланты не тестировались; другие языки, голоса и комнаты не проверены.
Discuss Telegram 0 31

Will it run?

Runs in F32

Fits without quantisation.

  • F32weights 0.0 GB · ready files0.5 GB
  • BF16weights 0.0 GB0.5 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 · F32 — 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.

    # bmarti44/before-the-implant-causal-enhancer
    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 bmarti44/before-the-implant-causal-enhancer --local-dir ~/models/before-the-implant-causal-enhancer
    # 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 before-the-implant-causal-enhancer (quants, finetunes, LoRAs) will show up under the bell.

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

About the model

bmarti44/before-the-implant-causal-enhancer
Author
bmarti44
Category
Audio
Released
Posted

On NeyroLearn

Hugging Face

310

Ask AI about this model

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

Comments