Speech-to-Text
Converting audio of speech into written text—transcribe voice calls, podcasts, lectures.
Think of it like
A court reporter typing out everything a witness says.
Example
Whisper transcribes English, Spanish, and Japanese from audio files with high accuracy in seconds per minute of audio.
How it actually works
Sequence-to-sequence model: encoder processes audio spectrograms, decoder generates text. Training on multilingual data (Whisper: 680k hours) enables zero-shot generalization. Attention layers learn to align audio frames with text. Inference faster than real-time.
For product teams
Unlock content: search podcasts by topic, caption videos, transcribe meetings.
For engineers
Audio preprocessed to Mel spectrogram; encoder extracts features; attention-based decoder generates tokens; beam search finds best hypothesis.
Related
- Text-to-Speech — Inverse of.
- Used in meeting transcription.
Read anything AI without the jargon
Look up any term in plain English, or save terms as you read with the free Chrome extension.
Open DecoderAdd to Chrome