guillaumekln/faster-whisper
Faster Whisper
faster-whisper is an automatic speech recognition framework and an optimized implementation of the Whisper speech-to-text engine. It functions as a CTranslate2 inference engine designed to convert spoken audio into written text.
The project serves as a model quantization tool that transforms large audio model weights into lower precision formats. This process reduces memory usage and increases execution speed on hardware by utilizing integer quantized weights.
The framework covers a broad range of capabilities including batch audio transcription for parallel processing and voice activity detection to filter out non-speech audio segments. It also provides utilities for converting original or fine-tuned audio models into formats compatible with the CTranslate2 runtime.
Features
- Speech-to-Text Engines - Provides an optimized engine for converting spoken audio recordings into written text.
- Audio Model Optimization - Optimizes audio models via lower precision formats to improve hardware execution speed and memory requirements.
- CTranslate2 Deployment - Runs audio transcription models through the CTranslate2 engine for higher performance than standard implementations.
- Automatic Speech Recognition - Implements a high-performance system for converting spoken audio into written text.
- Model Quantization Tools - Provides utilities to reduce numerical precision of audio model parameters for improved inference performance.
- Weight Quantization - Implements techniques to compress model weights into lower-precision integer formats for faster inference and reduced memory use.
- Whisper-Based Engines - Implements an optimized version of the Whisper model using CTranslate2 for faster, memory-efficient transcription.
- Transformer Inference Engines - Functions as a high-performance engine optimized for executing transformer-based speech models.
- Batch Transcription - Provides parallel processing of audio segments to maximize transcription throughput and reduce latency.
- Model Format Converters - Transforms PyTorch checkpoints into a proprietary format compatible with the CTranslate2 runtime.
- Voice Activity Detection - Identifies speech boundaries to filter out silent or non-speech segments before transcription.
- Voice Activity Detection - Identify and strip silent or non-speech sections from audio files using a voice activity detection model.
- Model Serving Engines - Optimized C++ inference engine for speech recognition models.
- Model Variants - Optimized reimplementation using CTranslate2 for increased speed.
