DPRNNTasNet-ks2 Audio Enhancement Model - An Open-source and Free Sharp Tool for Single-channel Speech Enhancement

Dprnntasnet Ks2 Libri1Mix Enhsingle 16k

Developed by JorisCos

An audio enhancement model trained on the Asteroid framework, specifically designed for single-channel speech enhancement tasks, trained on the Libri1Mix dataset.

Audio Enhancement

PyTorch

#Single-channel speech enhancement #16kHz audio processing #DPRNN architecture

Downloads 4,859

Release Time : 3/2/2022

Model Overview

This model adopts the DPRNNTasNet architecture, focusing on single-channel speech enhancement tasks, capable of separating and enhancing target speech signals from mixed audio.

Model Features

Efficient speech separation

Utilizes the DPRNNTasNet architecture to effectively separate target speech from mixed audio.

High-quality enhancement

Achieves a 14.72dB SI-SDR improvement on the Libri1Mix test set.

Real-time processing capability

The model is designed with real-time processing needs in mind, featuring chunk processing capability.

Model Capabilities

Single-channel speech enhancement

Speech signal separation

Audio quality improvement

Use Cases

Voice communication

Noisy environment speech enhancement

Improves voice call quality in background noise environments

SI-SDR improvement of 11.27dB, STOI improvement of 0.134

Audio post-processing

Recording quality improvement

Enhances low-quality recordings

SDR improvement of 11.85dB

🚀 Asteroid model `JorisCos/DPRNNTasNet_Libri1Mix_enhsignle_16k`

This model is an audio - related model focusing on asteroid, DPRNNTasNet, and audio - to - audio tasks. It was trained on specific datasets and has achieved certain results.

🚀 Quick Start

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.

✨ Features

Trained on the enh_single task of the Libri1Mix dataset.
Utilizes a well - defined training configuration for better performance.

📦 Installation

No installation steps are provided in the original document, so this section is skipped.

💻 Usage Examples

No code examples are provided in the original document, so this section is skipped.

📚 Documentation

Training Config

data:
  n_src: 1
  sample_rate: 16000
  segment: 1
  task: enh_single
  train_dir: data/wav16k/min/train-360
  valid_dir: data/wav16k/min/dev
filterbank:
  kernel_size: 2
  n_filters: 64
  stride: 1
masknet:
  bidirectional: true
  bn_chan: 128
  chunk_size: 250
  dropout: 0
  hid_size: 128
  hop_size: 125
  in_chan: 64
  mask_act: sigmoid
  n_repeats: 6
  n_src: 1
  out_chan: 64
optim:
  lr: 0.001
  optimizer: adam
  weight_decay: 1.0e-05
training:
  batch_size: 2
  early_stop: true
  epochs: 200
  gradient_clipping: 5
  half_lr: true
  num_workers: 4

Results

On Libri1Mix min test set :

si_sdr: 14.7228101708889
si_sdr_imp: 11.2730288650292
sdr: 15.35661405197161
sdr_imp: 11.853951252758595
sir: Infinity
sir_imp: NaN
sar: 15.35661405197161
sar_imp: 11.853951252758595
stoi: 0.9300461826351578
stoi_imp: 0.13412635909461715

📄 License

This work "DPRNNTasNet_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](https://creativecommons.org/licenses/by - nc/4.0/) (Research only). "DPRNNTasNet_Libri1Mix_enhsignle_16k" 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