wav2vec2-large-xlsr-53-faroese-100h
| Property | Value |
|---|---|
| License | cc-by-4.0 |
| Paper | ASR Language Resources for Faroese |
| WER (Test) | 7.6% |
| WER (Dev) | 5.5% |
What is wav2vec2-large-xlsr-53-faroese-100h?
This is a specialized automatic speech recognition (ASR) model designed specifically for the Faroese language. It was developed by fine-tuning the facebook/wav2vec2-large-xlsr-53 model using 100 hours of Faroese speech data from the Ravnur Project. The model represents a significant advancement in Faroese language technology, achieving impressive word error rates of 7.6% on test data.
Implementation Details
The model was fine-tuned at the Language and Voice Lab at Reykjavík University using the Ravnursson Faroese Speech and Transcripts dataset. It implements the wav2vec2 architecture, which has proven highly effective for low-resource languages like Faroese.
- Base Architecture: wav2vec2-large-xlsr-53
- Training Data: 100 hours of Faroese audio
- Sampling Rate: 16kHz
- Evaluation Metric: Word Error Rate (WER)
Core Capabilities
- Automatic speech recognition for Faroese language
- High accuracy with 7.6% WER on test set
- Support for 16kHz audio input
- Batch processing capabilities
- Integration with Hugging Face Transformers library
Frequently Asked Questions
Q: What makes this model unique?
This model is one of the first high-performing ASR models specifically trained for the Faroese language, making it a crucial resource for Faroese language technology development.
Q: What are the recommended use cases?
The model is ideal for Faroese speech transcription tasks, language documentation efforts, and building Faroese language applications requiring speech recognition capabilities.
