Amazon Transcribe
An AWS service that listens to audio and writes down what was said.
Amazon Transcribe is speech-to-text as an API. You give it an audio or video file stored in S3, or a live audio stream, and it returns a transcript: the words spoken, with a timestamp for each one and a confidence score. It works in two modes: batch, for recordings you already have, and streaming, for audio arriving in real time, like a live call.
Beyond plain transcription, it can tell speakers apart ("speaker 1", "speaker 2"), learn your jargon through a custom vocabulary so product names come out right, and hide personal data like card numbers in the text. A variant called Call Analytics is built for contact centers and adds things like sentiment and call summaries.
Common uses are subtitles for videos, searchable archives of meetings or podcasts, notes from customer calls, and voice input inside apps. It is the mirror image of Amazon Polly, which turns text into speech. You pay per second of audio transcribed.
Keep in mind
- Turns recorded or live audio into text with timestamps.
- Batch mode for files, streaming mode for real time.
- Can separate speakers, learn custom vocabulary and redact personal data.
- Billed per second of audio processed.
