AIbase
EN
Home
Convtasnet Libri2Mix Sepclean 8k
C

Convtasnet Libri2Mix Sepclean 8k

Developed by JorisCos
ConvTasNet model trained on the Asteroid framework for speech separation tasks at 8kHz sampling rate, trained on the Libri2Mix dataset.
Sound Separation
PyTorch
#Bilingual speech separation#8kHz audio processing#ConvTasNet architecture
Downloads 179
Release Time : 3/2/2022

Model Overview

This model is used to separate clean speech signals from mixed audio, particularly suitable for dual-speaker scenarios.

Model Features

Efficient speech separation
Effectively separates speakers in mixed speech at 8kHz sampling rate
Optimized ConvTasNet architecture
Uses a deep network structure with 8 blocks and 3 repetitions, featuring 128 skip connection channels
Lightweight training
Efficient training with a batch size of 24 and audio segment length of 3 seconds

Model Capabilities

Dual-speaker speech separation
Audio signal enhancement
Speech signal processing

Use Cases

Speech processing
Meeting recording separation
Separates individual speaker voices from multi-person meeting recordings
SI-SDR improvement of 14.76dB
Speech enhancement
Extracts target speech from background noise or other speakers
STOI improvement of 0.218
Featured Recommended AI Models
AIbase
Empowering the Future, Your AI Solution Knowledge Base
EnglishįŽ€äŊ“中文įšéĢ”中文ãĢãģんご
Š 2025AIbase