Model ID
whisper-large-v3
Whisper Large V3
Model Overview
Section titled “Model Overview”Base Architecture
Whisper Large V3
Model Highlights
Section titled “Model Highlights”- Handles automatic speech recognition across 50+ languages with improved accuracy over earlier Whisper releases.
- Supports English speech translation and zero-shot transcription for low-resource languages.
- Optimized for long-form audio; robust to accents, background noise, and code-switching.
Specifications
Section titled “Specifications”| Specification | Value |
|---|---|
| Parameters | ~1.55B |
| Encoder-Decoder Layers | 32 encoder / 32 decoder |
| Audio Frontend | 80-channel log-Mel spectrogram (25 ms window, 10 ms stride) |
| Max Audio Duration | ~30 seconds per chunk (sliding window supported) |
Recommended Use Cases
Section titled “Recommended Use Cases”- High-fidelity meeting and podcast transcription.
- Real-time captioning with batch segmentation.
- Multilingual customer-support call analytics.