← All models

Amazon Transcribe

Audio Available Full comparison ↗

amazon/transcribe · by Amazon · end-to-end-asr

Pricing — 1 offering(s)

Amazon Web Services

About this offering: Amazon Transcribe automatic speech recognition. Deeply integrated with AWS services (Lambda triggers, S3 event notifications, Glue ETL pipelines). Supports 100+ languages with custom vocabulary and custom language model training. Speaker diarisation, automatic content redaction (PII), and medical transcription (Medical tier at $0.0750/min) available. Volume pricing kicks in after 250K minutes ($0.015/min) and 1M minutes ($0.0102/min) per month. Streaming via WebSocket. New account free tier: 60 minutes/month for 12 months.

Standard transcription (first 250K min/month)

  • $0.024 / minute Historical 2018-04-01 → 2026-09-19

    Free tier 60 min/month for first 12 months (new accounts). Volume discounts at 250K+ and 1M+ minutes/month.

  • $0.006 / minute Current 2026-09-20 → present
    US East (N. Virginia) batch rate, cut from $0.024 to $0.006 per minute (checked 2026-09-20). Amazon's pricing…

    US East (N. Virginia) batch rate, cut from $0.024 to $0.006 per minute (checked 2026-09-20). Amazon's pricing example now states a flat $0.006/min; the earlier 250K- and 1M-minute volume bands may still appear in the regional pricing table, so check it for very high volumes.

Real-time streaming

  • $0.024 / minute Historical 2018-04-01 → 2026-09-19
  • $0.01 / minute Current 2026-09-20 → present

    US East (N. Virginia) streaming rate, cut from $0.024 to $0.01 per minute (checked 2026-09-20).

Showing the active price and any recorded history. Full pricing history is available via the paid API — see API docs.

See how Amazon Transcribe fits into a cost-aware routing setup

See how →

Capability profile

How Amazon Transcribe rates across core capability dimensions, with the task-level evidence behind each rating.

transcription accuracy Strong

Strong on general English; Custom Language Model training significantly improves domain-specific accuracy. Medical tier optimised for clinical vocabulary.

language support Strong

100+ languages with automatic language identification. Multi-language transcription identifies and transcribes multiple languages in a single audio file.

speaker diarisation Strong

Built-in speaker identification. Call Analytics feature adds turn-by-turn analysis with sentiment, issues, and interruption detection.

real time streaming Strong

WebSocket and HTTP/2 streaming API. Stable for production workloads; deep integration with AWS EventBridge and Lambda.

custom vocabulary Strong

Custom Vocabulary lists and Custom Language Model training both supported. Industry-specific vocabulary packages available (medical, legal, financial).

Benchmarks

Benchmark Score Config Source
WER (LibriSpeech) — Amazon does not publish LibriSpeech WER benchmarks. Internal benchmarks indicate competitive accuracy; independent comparisons suggest parity with Azure on clean English audio. —

Operator guidance

The best choice for AWS-native stacks with call centre analytics, medical transcription, or heavy PII redaction needs. At $0.024/min ($1.44/hr) it is the most expensive mainstream STT option — justified by the rich AWS ecosystem integration and Call Analytics feature. For cost, Universal-2 ($0.15/hr) is 10× cheaper. For latency, Deepgram Nova-3 is significantly faster. Volume pricing kicks in at 250K minutes/month.

Use cases

Limitations

Citations