Models
Models control the quality, latency, and language coverage of generated audio.
Overview
Every IntuneLabs capability is powered by a dedicated model. Speech synthesis models turn text into audio, while speech recognition models turn audio into text. Each model trades off expressiveness, latency, and language coverage, so the right choice depends on your workload.
Speech synthesis
Synthesis models generate natural, human-like speech from text. Use a higher-fidelity model for narration and long-form content, and a faster model when real-time latency matters more than maximum expressiveness.
| Model | Best for | Languages |
|---|---|---|
intune_v3 | Expressive, multi-speaker performance | 70+ |
intune_multilingual_v2 | Stable, long-form narration | 29 |
intune_flash_v2_5 | Real-time, low-latency use | 32 |
Speech recognition
Recognition models transcribe spoken audio into accurate, timestamped text. scribe_v2 delivers state-of-the-art accuracy across 90+ languages with speaker diarization and dynamic audio tagging. scribe_v2_realtime targets streaming transcription at low latency.
Choosing a model
- Optimize for quality and emotion: reach for the most expressive synthesis model.
- Optimize for latency: use the flash synthesis model or the realtime recognition model.
- Optimize for cost at scale: the flash model lowers the per-character price for API generations.
Model identifiers are passed wherever a generation is requested, in the dashboard or via the API. See the per-capability guides for the supported model list and parameters.
Built withfern