| Property | Value |
|---|---|
| Base Model | openai/whisper-small |
| Training Data | 35,141 Galician audio samples |
| Evaluation WER | 13.681% |
| Model Source | Hugging Face |
What is whisper-small-gl?
whisper-small-gl is a specialized speech-to-text model developed by Mozilla.ai, specifically optimized for the Galician language. It's based on OpenAI's Whisper-small architecture and has been finetuned on over 35,000 Galician audio samples from the Common Voice dataset version 17.0.
Implementation Details
The model represents a significant improvement over the baseline Whisper-small model for Galician language processing. Through careful finetuning, the Word Error Rate (WER) was reduced from 40.812% to 13.681%, while the loss decreased from 1.506 to 0.21. This enhancement was achieved using Mozilla.ai's speech-to-text-finetune Blueprint methodology.
- Baseline performance: 40.812% WER, 1.506 loss
- Finetuned performance: 13.681% WER, 0.21 loss
- Training dataset: mozilla-foundation/common_voice_17_0
Core Capabilities
- Accurate transcription of Galician speech
- Significantly improved performance compared to the base model
- Optimized for Galician language nuances and pronunciation
- Suitable for production deployment in Galician-language applications
Frequently Asked Questions
Q: What makes this model unique?
This model stands out for its specialized optimization for the Galician language, achieving a remarkable improvement in Word Error Rate compared to the baseline model. The substantial reduction in WER from 40.812% to 13.681% makes it particularly effective for Galician speech recognition tasks.
Q: What are the recommended use cases?
The model is ideal for applications requiring Galician speech transcription, including: automated subtitling systems, voice assistants for Galician speakers, transcription services for Galician media content, and academic or business applications requiring Galician speech-to-text capabilities.