Amazon: Amazon Transcribe

Amazon Transcribe's standard batch (asynchronous job) speech-to-text: the media is read from an S3 object, StartTranscriptionJob returns a job that is polled with GetTranscriptionJob, and the transcript JSON is fetched from a presigned URL. Word-level timestamps and confidences are always returned; speaker diarization (up to 30 speakers) and automatic language identification are opt-in and included in the base batch rate. 113 documented batch locales. NO NATIVE MODEL ID EXISTS: StartTranscriptionJob has no model, engine or version parameter (the only model-selecting field is ModelSettings.LanguageModelName for account-trained custom language models, not exposed), so native_model_ids is empty and the route's provider_model_id `amazon-transcribe` is a synthetic router-side id that is never sent on the wire. The real-time streaming surface (different host, price and language list) is not part of this card.