Whisper Large v3
OpenAI
OpenAI's state-of-the-art automatic speech recognition (ASR) model with 1.55B parameters supporting 99+ languages. Trained on 1M hours of weakly labeled audio plus 4M hours of pseudo-labeled audio from Whisper large-v2. Uses 128 Mel frequency bins (vs 80 in previous versions) and trained for 2.0 epochs. Achieves 7.4% WER average and 97.9% word accuracy on LibriSpeech. Shows 10-20% error reduction vs large-v2 and 72% WER reduction vs prior MLPerf ASR model (RNN-T). Performs automatic language identification, generates phrase-level timestamps, and handles punctuation/capitalization. Strongest on high-resource languages (English, Spanish, French, German). Open source enables self-hosting and commercial use.
Strengths
- Supports 99+ languages - exceptional multilingual coverage
- 7.4% WER and 97.9% word accuracy demonstrate state-of-the-art ASR
- 10-20% error reduction vs Whisper large-v2
- Automatic language ID, phrase-level timestamps, punctuation/capitalization
- Trained on 5M hours of audio - massive scale
- Apache 2.0 license enables full commercial use and modification
- Efficient deployment - RTX 3060 transcribes ~200 hours per dollar
Caveats
- Strongest on high-resource languages - lower accuracy on rare languages
- Requires ~10GB VRAM for inference
- No real-time streaming support in base model
- Accuracy degrades on accented speech and domain-specific terminology
- Consider fine-tuning for specialized domains (medical, legal, etc.)
- Vendor
- OpenAI
- License
- Apache-2.0
- Release Date
- 2023-11-01
- Modalities
- audio
| Benchmark | Score |
|---|---|
| LibriSpeech WER | 7.4 |
| LibriSpeech Word Accuracy | 97.9 |
Capabilities
Pricing
Open source (Apache 2.0) - free. Requires ~10GB VRAM. OpenAI API: $0.006 per minute