whisper

Audio transcription converts spoken recordings into readable text and timestamped subtitles. You can process meetings or podcasts by selecting a Whisper model, applying optional language hints, and running CPU or GPU inference to output segment-level timing for captions.

Multilingual transcription supports 99 languages and can translate non-English audio to English while preserving meaning. The process identifies the spoken language before or during transcription, allowing you to generate English translations from foreign language recordings.

Speech recognition handles noisy recordings and podcast episodes by preprocessing audio before transcription. You select a Whisper model and apply audio preprocessing to isolate speech, generating accurate text transcripts even from recordings with background noise.

Creating timestamped meeting notes requires segment-level or word-level timing outputs from audio transcription. By processing meeting recordings with language identification and Whisper model selection, you generate searchable transcripts with precise timestamps for review and archiving.

Whisper transcription supports both CPU and GPU inference. You select your preferred hardware environment during setup, allowing you to process audio files and generate text transcripts regardless of whether you have GPU acceleration available.

Language identification detects the spoken language before or during transcription of multilingual audio. You can provide optional language hints to improve accuracy, and the system automatically identifies languages across 99 supported options to generate accurate text and English translations.