Features

The "systems" lens documents Whisper's source modules. This section documents what those modules do for end users — the user-visible capabilities, end to end, with pointers back into the relevant source files.

Feature	What it does
Transcription	Long-form speech-to-text in 99 languages with sliding-window decoding, segment-level timestamps, and temperature fallback.
Translation	X→English speech translation. A multilingual model with task="translate".
Language detection	Pick the spoken language from the first 30 s of audio. Reachable both as a side-effect of transcribe() and as whisper.detect_language().
Word timestamps	Per-word start/end/probability via cross-attention DTW alignment, with optional GPU acceleration and hallucination-silence skipping.

Each page links into the relevant systems/ documents for implementation detail. There is no separate frontend or service for any of these — every feature is exposed by the whisper CLI (whisper/transcribe.py:cli) and the Python API (whisper.transcribe, whisper.decode, whisper.detect_language).