Whisper

Whisper is OpenAI's open-source speech recognition model. Because the weights are public, it can be run locally for transcription and translation with no per-minute fee.

Why it is widely used

• Multilingual. Strong transcription across dozens of languages, plus translation to English.
• Noisy audio tolerance. Performs reasonably on imperfect recordings.
• Self-hostable. Suitable for confidential recordings.
• Good ecosystem. Faster runtimes and wrappers exist for most platforms.

Limitations

Timestamps and speaker separation need extra tooling, and accuracy falls on heavy accents or overlapping speech. Long files require chunking.

Best for

Anyone transcribing interviews, meetings or subtitles who wants local processing and no subscription.