Vercel Integrates Microsoft AI Audio and Transcription Models into AI Gateway
Vercel has added Microsoft's MAI-Voice-2.1 and MAI-Transcribe-2 Streaming models to its AI Gateway, enabling highly secure and low-latency audio processing in 23 languages.
Vercel has expanded its AI Gateway capabilities by integrating the latest audio models from Microsoft AI. This release introduces MAI-Voice-2.1, which supports consistent voice characteristics across 23 languages, alongside the lower-latency MAI-Voice-2.1-Flash optimized for real-time voice agents. For audio-to-text workflows, developers can now leverage MAI-Transcribe-2 Streaming to achieve continuous, near-instantaneous speech transcription.
Primary source
Open Vercel Changelog ↗This article summarizes the source. Check the original for full details and conditions.
Comparison
| Aspect | Before / Alternative | After / This |
|---|---|---|
| Supported Modalities | Primarily text-based models | Text, multilingual speech synthesis, and real-time streaming transcription |
| Audio Latency | Standard API routing with potential overhead | Flash models optimized specifically for low-latency voice agents |
| Data Retention | Standard provider policies requiring individual configuration | Zero Data Retention (ZDR) policy automatically enforced for training data |
| Gateway Management | Separate endpoints and distinct credential management systems | Unified routing, rate limiting, and monitoring under Vercel AI Gateway |
