Whisper

• Source: https://github.com/openai/whisper
• License: MIT
• Alternatives: Parakeet, Vosk, faster-whisper, Wav2Vec2

Speech-to-text transcription service.

Features

• High-Accuracy Transcription: Handles accents, background noise, and technical vocabulary well
• 99+ Languages: Automatic language detection with multilingual transcription
• Timestamps: Word- and segment-level timestamps for alignment and captioning
• Multiple Model Sizes: Tiny to large variants trade off speed vs. accuracy
• REST API: Various server wrappers (faster-whisper, whisper.cpp) expose an OpenAI-compatible endpoint

References

• Whisper Diarization — speaker identification (who said what) — TBD
• NVIDIA NeMo — alternative speech AI framework