Mistral-Common v1.9.0 - Audio Streaming Support
AI Impact Summary
Mistral-Common has gained the ability to process audio streams directly, enabling real-time transcription capabilities. This enhancement leverages the `Voxtral-Mini-4B-Realtime-2602` model via a MistralTokenizer, allowing for online or offline transcription requests. The updated code includes improvements to padding and audio encoding, specifically targeting streaming workflows, and introduces a new audio encoder within the tokenizer.
Affected Systems
- Date
- Date not specified
- Change type
- capability
- Severity
- info