AWS Machine Learning Blog•September 24, 2026•Published September 24, 2026

Speaker-labeled transcription with WhisperX on SageMaker AI

WhisperX on SageMaker gives speaker‑labeled, per‑word timestamps for real‑time or batch audio, fixing common transcription gaps.

Why Now

AWS released a WhisperX Deep Learning Container for SageMaker, enabling production‑ready speech‑to‑text with speaker diarization and precise timestamps.

What Happened

WhisperX wraps OpenAI’s Whisper, adds wav2vec2 forced alignment for per‑word timestamps, and includes speaker diarization. It ships as a GPU‑ready Docker image that can be deployed to SageMaker real‑time or asynchronous endpoints. The container exposes POST /invocations and supports json, verbose_json, srt, and vtt outputs.

Why It Matters

Accurate speaker labels and timestamps let contact centers measure talk time, run sentiment analysis, and meet compliance in regulated fields. Media teams can auto‑generate captions in SRT/VTT, and developers can build searchable meeting notes or audit trails.

The Limitation

The real‑time endpoint is limited to 60‑second clips; longer audio requires the asynchronous pattern, which adds S3 overhead and longer latency.

What You Can Do

Deploy the WhisperX DLC to a SageMaker endpoint and test it on a short audio clip to see the speaker‑labeled transcript output.

Source

Read original source

Why we picked this

Detailed technical guide on deploying WhisperX for speaker-labeled transcription, core AI tooling.