Giving written policies a voice

A Healthcare & Life Sciences business working with Cloud Combinator on AWS. The client is anonymised at their request.

The challenge had three focus areas, all about closing the gap between a written document and usable spoken content.

The workflow starts from a source document. A large language model within Amazon Bedrock reads the document and produces a clean narration script, rewriting dense policy language into text that reads well aloud while preserving the substance of the original.

The proof of concept demonstrated the full path from document to spoken script on a real the client's policy. Using the platform, Health and Environment policy as the source, the language model produced a clean, spoken-word narration transcript ready for Amazon Polly to voice, showing that the approach works on the client's own material.

• Real policy proven — The workflow was demonstrated on the client's own the platform, Health and Environment policy, not a synthetic example.
• Natural narration script — The language model turned formal policy text into a clean, flowing narration transcript suited to being spoken aloud.
• Studio-free voice — With Amazon Polly synthesising the script, the client can produce professional narration without booking studio time or a voice artist.
• Repeatable by design — Because the pipeline is automated end to end, new and updated documents can be turned into audio by re-running the workflow.

With the document-to-script step proven on a real policy, the natural next step is to connect Amazon Polly to complete the audio pipeline, tune the voice to the client's preference and roll the approach across the wider library of policies and training material. The proof of concept shows the path is viable on AWS.