OpenAI Whisper Integration for Speech Recognition OpenAI Whisper is an open-source speech recognition model trained on 680,000 hours of multilingual audio. Its WER on the English Libri
Speech dataset is 2.7%, which is on par with professional transcriptionists. For Russian on clean audio, the WER is 8–12%. ### What you get with Whisper integration - Local processing without sending data to third-party clouds - Support for 99 languages out of the box - Works with MP3, WAV, FLAC, M4A, OGG, WebM formats - Automatic language detection - Output of word-level timestamps (with --word_timestamps True) ### Deployment Options | Model | Parameters | VRAM | Speed (RTX 3090) |
|--------|-----------|-------|----------------------| | tiny | 39M | 1 GB | ~32x realtime | | base | 74M | 1 GB | ~16x realtime | | small | 244M | 2 GB | ~6x realtime | | medium | 769M | 5 GB | ~2x realtime | | large-v3 | 1550M | 10 GB | ~1x realtime | For most production tasks, small or medium are sufficient - acceptable quality with reasonable resources. ### Integration stack Connect via openai-whisper (PyPI) or via the OpenAI HTTP API (/v1/audio/transcriptions). For high loads, use faster-whisper based on CTranslate2: 4x speedup with the same quality.```python
from faster_whisper import WhisperModel
model = WhisperModel("medium", device="cuda", compute_type="float16") segments, info = model.transcribe("audio.mp3", beam_size=5) for segment in segments: print(f"[{segment.start:.2f}s] {segment.text}")