ConvTasNet Libri1Mix enhsingle 16k

ProviderJorisCos
Categoryaudio-to-audio
Licensecc-by-sa-4.0
Downloads11.2K
Stars0

Overview

ConvTasNet Libri1Mix enhsingle 16k is a specialized deep learning model designed for single-channel speech enhancement. Unlike traditional spectral subtraction methods, this model operates in the time domain using a convolutional encoder-decoder architecture to isolate target speech from background noise. It is specifically optimized for 16kHz audio, making it a practical choice for developers building voice assistants, transcription pre-processing pipelines, or communication tools where signal-to-noise ratio (SNR) improvement is critical. Integration is straightforward for those working with PyTorch or similar tensor-based frameworks, providing a lightweight alternative to heavier transformer-based separation models while maintaining low latency for real-time applications.

Highlights

  • Time-domain speech enhancement for 16kHz audio signals
  • Effective noise reduction for single-channel voice recordings
  • Low-latency architecture suitable for real-time processing
  • Ideal pre-processing step for ASR and transcription engines
  • Open-source CC-BY-SA 4.0 license for flexible deployment

Usage

Install
# Install Hugging Face transformers
pip install transformers torch
SDK Usage
# Load model with transformers
from transformers import AutoModel, AutoTokenizer

model = AutoModel.from_pretrained("JorisCos/ConvTasNet_Libri1Mix_enhsingle_16k")
tokenizer = AutoTokenizer.from_pretrained("JorisCos/ConvTasNet_Libri1Mix_enhsingle_16k")

Hugging Face Download

We recommend downloading the model via the Hugging Face CLI or Hub SDK.

Guidance：Before downloading, install huggingface_hub with:

Guidance 
pip install -U huggingface_hub

CLI Download

Download the full repository

Download the full repository 
huggingface-cli download JorisCos/ConvTasNet_Libri1Mix_enhsingle_16k

Download a single file to a local folder (e.g. config.json into ./dir)

Download a single file to a local folder (e.g. config.json into ./dir) 
huggingface-cli download JorisCos/ConvTasNet_Libri1Mix_enhsingle_16k config.json --local-dir ./dir

See the official docs for more CLI options

SDK Download

SDK Download 
# 模型下载
from huggingface_hub import snapshot_download
model_dir = snapshot_download('JorisCos/ConvTasNet_Libri1Mix_enhsingle_16k')

Git Download

Make sure git-lfs is installed first

Git Download 
git lfs install
git clone https://huggingface.co/JorisCos/ConvTasNet_Libri1Mix_enhsingle_16k

To skip LFS large-file downloads, use:

Skip LFS 
GIT_LFS_SKIP_SMUDGE=1 git clone https://huggingface.co/JorisCos/ConvTasNet_Libri1Mix_enhsingle_16k

Model files are hosted on the Hugging Face Hub — download directly via HF CLI / SDK / Git, not through this site.

PyTorch / Transformers Usage

Install Transformers

Install Transformers 
pip install -U transformers torch

Load the model and run inference

Load the model and run inference 
from transformers import AutoModelForCausalLM, AutoTokenizer

model = AutoModelForCausalLM.from_pretrained('JorisCos/ConvTasNet_Libri1Mix_enhsingle_16k')
tokenizer = AutoTokenizer.from_pretrained('JorisCos/ConvTasNet_Libri1Mix_enhsingle_16k')

Full Documentation

来源: HuggingFace

---
tags:

  • asteroid

  • audio

  • ConvTasNet

  • audio-to-audio

datasets:
  • Libri1Mix

  • enh_single

license: cc-by-sa-4.0
---

Asteroid model JorisCos/ConvTasNet_Libri1Mix_enhsignle_16k

Description:

This model was trained by Joris Cosentino using the librimix recipe in Asteroid.
It was trained on the enh_single task of the Libri1Mix dataset.

Training config:

yml
data:
  n_src: 1
  sample_rate: 16000
  segment: 3
  task: enh_single
  train_dir: data/wav16k/min/train-360
  valid_dir: data/wav16k/min/dev
filterbank:
  kernel_size: 32
  n_filters: 512
  stride: 16
masknet:
  bn_chan: 128
  hid_chan: 512
  mask_act: relu
  n_blocks: 8
  n_repeats: 3
  n_src: 1
  skip_chan: 128
optim:
  lr: 0.001
  optimizer: adam
  weight_decay: 0.0
training:
  batch_size: 6
  early_stop: true
  epochs: 200
  half_lr: true
  num_workers: 4

Results:

On Libri1Mix min test set :

yml
si_sdr: 14.743051006476085
si_sdr_imp: 11.293269700616385
sdr: 15.300522933671061
sdr_imp: 11.797860134458015
sir: Infinity
sir_imp: NaN
sar: 15.300522933671061
sar_imp: 11.797860134458015
stoi: 0.9310514162434267
stoi_imp: 0.13513159270288563

License notice:

This work "ConvTasNet_Libri1Mix_enhsignle_16k" is a derivative of LibriSpeech ASR corpus by Vassil Panayotov,
used under CC BY 4.0; of The WSJ0 Hipster Ambient Mixtures
dataset by Whisper.ai, used under CC BY-NC 4.0 (Research only).
"ConvTasNet_Libri1Mix_enhsignle_16k" is licensed under Attribution-ShareAlike 3.0 Unported by Joris Cosentino

View official source
Join our Telegram