oscarstories/Voxtral-Mini-3B-2507-executorch

Voxtral-Mini-3B-2507 Fine-tuned Model

Model description

This model is a derivative of mistralai/Voxtral-Mini-3B-2507, adapted for speech transcription tasks.
It retains the base model architecture and is intended for automatic speech recognition (ASR) workflows.

Intended use

• Speech-to-text transcription
• Multilingual audio transcription (depending on fine-tuning setup)
• Real-time or batch transcription pipelines

Out-of-scope use

• High-stakes decision-making without human review
• Use cases requiring guaranteed factual correctness
• Audio domains not represented in training/adaptation data

Base model

• mistralai/Voxtral-Mini-3B-2507

Limitations

• Performance can degrade with noisy audio, strong accents, overlapping speech, or domain-specific jargon
• Possible transcription bias across languages, dialects, and speaker demographics
• May hallucinate or omit words in challenging acoustic conditions

Ethical considerations

Use responsibly and with user consent, especially for personal or sensitive recordings.
Ensure compliance with local privacy and data protection regulations.

License

This model is released under the MIT License.