AWS has released a GPU-ready WhisperX container for Amazon SageMaker AI, giving developers a packaged way to produce transcripts with word-level timestamps and speaker labels. WhisperX combines OpenAI’s Whisper speech recognition with wav2vec2 forced alignment and speaker diarization, the process of identifying who spoke when.
The container can return JSON, detailed JSON, SRT captions or WebVTT captions. AWS says it includes the required transcription, alignment and diarization components and does not require a Hugging Face token. Teams can use the output for searchable meetings, call analysis, captions, redaction and audit workflows.
Two SageMaker deployment modes address different workloads. Real-time endpoints are intended for short, interactive clips that finish within SageMaker’s 60-second response limit. Asynchronous endpoints put input and output in Amazon S3, support longer recordings and can scale to zero when idle. Throughput scales by adding instances because each container handles one request at a time.
Deployment still requires a SageMaker role, GPU quota and configuration work. AWS specifically warns that GPU variants must pin the required SageMaker inference AMI or the container will fail to start. The company provides a runnable notebook and suggests a g4dn.xlarge instance for lower cost or g5.2xlarge for more headroom.