asr-wav2vec2-ctc-french
| Property | Value |
|---|---|
| Parameter Count | 315M |
| License | Apache 2.0 |
| Architecture | Wav2Vec2-CTC |
| Language | French |
What is asr-wav2vec2-ctc-french?
This is a specialized French automatic speech recognition (ASR) model based on the wav2vec2 architecture. It's a fine-tuned version of wav2vec2-FR-7K-large, trained on an extensive dataset of over 2,200 hours of French speech audio from multiple sources including Common Voice, Multilingual LibriSpeech, Voxpopuli, and others.
Implementation Details
The model leverages the wav2vec2 architecture with CTC (Connectionist Temporal Classification) for speech recognition. It operates at a 16kHz sampling rate and includes both standard and language model-enhanced decoding options. The model demonstrates strong performance across various French speech datasets, achieving Word Error Rates (WER) as low as 5.13% on Multilingual LibriSpeech with language model integration.
- Comprehensive training on 6 different French speech datasets
- Supports both standard and language model-enhanced inference
- Optimized for 16kHz audio input
- Implements CTC-based speech recognition
Core Capabilities
- High-accuracy French speech transcription (9.66% WER on Common Voice with LM)
- Handles various French accents including African French
- Efficient processing of long-form audio with chunking support
- Real-time transcription capabilities
Frequently Asked Questions
Q: What makes this model unique?
The model stands out for its comprehensive training on diverse French speech datasets and its dual-mode operation with and without language model enhancement. Its robust performance across different French accents and speech contexts makes it particularly versatile.
Q: What are the recommended use cases?
The model is ideal for French speech transcription tasks, particularly in scenarios requiring high accuracy such as media transcription, voice command systems, and automated subtitling. It's especially effective for applications where varying French accents need to be handled reliably.
