S

Sew Tiny 100k Ft Ls100h

Developed by asapp
SEW (Squeezed and Efficient Wav2vec) is a speech recognition pre-trained model developed by ASAPP Research, outperforming wav2vec 2.0 in both performance and efficiency.
Downloads 736
Release Time : 3/2/2022

Model Overview

A speech recognition model pre-trained on 16kHz sampled audio, requiring fine-tuning for downstream tasks.

Model Features

Efficient Performance
Achieves 1.9x inference speedup compared to wav2vec 2.0 with a 13.5% reduction in word error rate.
Compressed Architecture
Optimized model architecture reduces computational resource requirements while maintaining performance.
Multi-task Adaptation
Can be fine-tuned for various speech tasks such as ASR, speaker recognition, and intent classification.

Model Capabilities

Speech Recognition
Speech-to-Text
Audio Feature Extraction

Use Cases

Speech Transcription
LibriSpeech Transcription
Transcribing English audiobook content into text.
Achieves WER 10.61 on LibriSpeech clean test set and WER 23.74 on other test set.
Speech Application Development
Voice Assistant
Serving as the speech recognition component for voice assistants.
Featured Recommended AI Models
AIbase
Empowering the Future, Your AI Solution Knowledge Base
Š 2025AIbase