lite-whisper-large-v3 Open-Source Speech Recognition Model - High Accuracy, Compact, and Easy to Deploy

Lite Whisper Large V3

Developed by efficient-speech

Lite-Whisper is a compressed version based on OpenAI Whisper, utilizing LiteASR technology to reduce model size while maintaining high accuracy.

Speech Recognition

Transformers

Open Source License:Apache-2.0 #Compressed Speech Recognition #Low Parameter Count #Multilingual Support

Downloads 70

Release Time : 2/26/2025

Model Overview

This is a compressed version of the Whisper speech recognition model, designed for efficient speech recognition and suitable for automatic speech-to-text tasks in multiple languages.

Model Features

Efficient Compression

Uses LiteASR technology to significantly reduce model size while maintaining recognition accuracy close to the original model.

Fast Inference

Optimized model structure supports faster speech recognition processing speed.

Multilingual Support

Supports speech recognition tasks in multiple languages.

Model Capabilities

Speech-to-text

Multilingual speech recognition

Real-time speech transcription

Use Cases

Speech Transcription

Meeting Minutes

Automatically converts meeting recordings into text transcripts

Average WER 12.6%

Podcast Transcription

Automatically converts podcast content into text scripts

Voice Assistants

Voice Command Recognition

Recognizes user voice commands and converts them into executable commands

🚀 Model Card for Lite-Whisper large-v3

Lite-Whisper large-v3 is a compressed version of OpenAI Whisper, leveraging LiteASR technology to optimize performance and resource utilization.

🚀 Quick Start

Here's a code snippet to help you get started with Lite-Whisper large-v3:

Basic Usage

import librosa 
import torch
from transformers import AutoProcessor, AutoModel

device = "cuda:0"
dtype = torch.float16

# load the compressed Whisper model
model = AutoModel.from_pretrained(
    "efficient-speech/lite-whisper-large-v3-turbo", 
    trust_remote_code=True, 
)
model.to(dtype).to(device)

# we use the same processor as the original model
processor = AutoProcessor.from_pretrained("openai/whisper-large-v3")

# set the path to your audio file
path = "path/to/audio.wav"
audio, _ = librosa.load(path, sr=16000)

input_features = processor(audio, sampling_rate=16000, return_tensors="pt").input_features
input_features = input_features.to(dtype).to(device)

predicted_ids = model.generate(input_features)
transcription = processor.batch_decode(
    predicted_ids, 
    skip_special_tokens=True
)[0]

print(transcription)

📚 Documentation

Model Information

Property	Details
Base Model	openai/whisper-large-v3
Library Name	transformers
License	apache-2.0
Pipeline Tag	automatic-speech-recognition
Tags	audio, automatic-speech-recognition, whisper, hf-asr-leaderboard

Benchmark Results

The following table presents the average word error rate (WER) evaluated on the ESB datasets:

Model	Average WER (↓)	Encoder Size	Decoder Size
whisper-large-v3	10.1	635M	907M
lite-whisper-large-v3-acc	10.1	429M	907M
lite-whisper-large-v3	10.2	377M	907M
lite-whisper-large-v3-fast	11.3	308M	907M

whisper-large-v3-turbo	10.1	635M	172M
lite-whisper-large-v3-turbo-acc	10.2	421M	172M
lite-whisper-large-v3-turbo	12.6	374M	172M
lite-whisper-large-v3-turbo-fast	20.1	313M	172M

whisper-medium	14.8	306M	457M

Additional Resources

For more detailed information about Lite-Whisper and LiteASR, please refer to our GitHub repository and paper.

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