Amazon Transcribe is the automatic speech recognition service in AWS. It converts audio to text either in batch, from recordings you already have, or as a live stream for real-time transcription, and it is accessed through the AWS Management Console, the AWS SDKs, and the service API. Voice-agent frameworks use the streaming mode as a speech-to-text stage.
Beyond plain transcription, the service adds speaker attribution, timestamps, custom vocabularies and language models, and a set of content-safety controls. Two specialised variants cover contact-centre calls (Call Analytics) and clinical dictation and conversation (Transcribe Medical).
Features
- Streaming and batch: transcribe stored audio files or stream audio over a secure connection for real-time results
- Language identification: detects the dominant language, or several languages within one file, without you specifying a language code
- Speaker diarization: attributes speaker changes in mono audio and identifies channels in multi-channel contact-centre recordings
- Custom vocabulary and language models: add domain terms and product names, or train a custom language model on your own text corpus
- Alternative transcriptions: returns up to 10 alternatives per sentence
- PII redaction and vocabulary filtering: redacts personally identifiable information and removes words on a filter list
- Toxicity detection: flags harmful speech into categories
- Call Analytics: sentiment, issue detection, compliance monitoring, and automated call summaries for contact centres
- Transcribe Medical: dictation and conversational modes across clinical specialties, with speaker diarization
- Formatted output: automatic punctuation, number normalisation, and per-word timestamps for subtitle generation
- Encryption: data at rest in S3 or with KMS keys, and TLS 1.2 in transit