ConvTasNet_Libri2Mix_sepnoisy_8k Open-source Model - Free Deployment to Solve the Problem of Noisy Speech Separation at 8kHz

Convtasnet Libri2Mix Sepnoisy 8k

Developed by JorisCos

ConvTasNet model trained on the Asteroid framework for speech separation tasks at 8kHz sampling rate, specifically designed for noisy mixed speech scenarios.

Sound Separation

PyTorch

#Stereo Speech Separation #Low SNR Enhancement #8kHz Audio Processing

Downloads 31

Release Time : 3/2/2022

Model Overview

This model adopts the ConvTasNet architecture and is trained on the 'sep_noisy' task of the Libri2Mix dataset, capable of separating clear speech signals from noisy mixed audio.

Model Features

Noise Robustness

Specifically optimized for noisy mixed speech, effectively separating background noise and speech signals

Lightweight Architecture

Adopts the ConvTasNet architecture, maintaining high separation performance with low computational complexity

8kHz Optimization

Specifically optimized for 8kHz sampled audio, suitable for low-bandwidth scenarios like telephone speech

Model Capabilities

Speech Separation

Noise Suppression

Audio Enhancement

Use Cases

Voice Communication

Telephone Speech Enhancement

Improves speech clarity in noisy telephone conversations

SI-SDR improvement of 11.94dB, STOI improvement of 0.207

Audio Post-Processing

Recording Denoising

Separates clean speech from recordings with ambient noise

SIR improvement of 22.46dB, SAR improvement of 4.25dB

🚀 Asteroid model `JorisCos/ConvTasNet_Libri2Mix_sepnoisy_8k`

This is an audio model based on ConvTasNet, trained on the Libri2Mix dataset. It can be used for audio separation tasks, providing high - quality audio - to - audio processing capabilities.

🚀 Quick Start

This part mainly focuses on the basic information of the model, including its source and description.

The model is imported from Zenodo.
It was trained by Joris Cosentino using the librimix recipe in [Asteroid](https://github.com/asteroid - team/asteroid) on the sep_noisy task of the Libri2Mix dataset.

📚 Documentation

Training Configuration

The following is the detailed training configuration of the model:

data:
    n_src: 2
    sample_rate: 8000
    segment: 3
    task: sep_noisy
    train_dir: data/wav8k/min/train-360
    valid_dir: data/wav8k/min/dev
filterbank:
    kernel_size: 16
    n_filters: 512
    stride: 8
masknet:
    bn_chan: 128
    hid_chan: 512
    mask_act: relu
    n_blocks: 8
    n_repeats: 3
    skip_chan: 128
optim:
    lr: 0.001
    optimizer: adam
    weight_decay: 0.0
training:
    batch_size: 24
    early_stop: True
    epochs: 200
    half_lr: True
    num_workers: 4

Results

The performance of the model on the Libri2Mix min test set is as follows:

si_sdr: 9.944424856077259
si_sdr_imp: 11.939395359731192
sdr: 10.701526190782072
sdr_imp: 12.481757547845662
sir: 22.633644975545575
sir_imp: 22.45666740833025
sar: 11.131644100944868
sar_imp: 4.248489589311784
stoi: 0.852048619949357
stoi_imp: 0.2071994899565506

Information Table

Property	Details
Model Type	ConvTasNet
Training Data	Libri2Mix (`sep_noisy` task)

📄 License

This work "ConvTasNet_Libri2Mix_sepnoisy_8k" 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](https://creativecommons.org/licenses/by - nc/4.0/) (Research only). "ConvTasNet_Libri2Mix_sepnoisy_8k" is licensed under [Attribution - ShareAlike 3.0 Unported](https://creativecommons.org/licenses/by - sa/3.0/) by Joris Cosentino.

Featured Recommended AI Models

Empowering the Future, Your AI Solution Knowledge Base

English 简体中文繁體中文にほんご