Start with an interview recording

A creator wants to adapt part of an interview into a two-voice explanatory program. The work includes transcription, checking meaning, writing for listening, casting voices, generating, reviewing and revising. A single text-to-speech call covers only one step.

This is a scenario for discussion, not an executed demonstration. A synthetic adaptation is not a recording of words the interviewee actually spoke. Source material, voice permissions and the distinction between quotation and adaptation belong in the production process.

The Agent needs to discover its tools

A person sees model, language and voice lists in an app. An Agent needs the equivalent information: discover tools, check ready models and confirm the required inputs and outputs. A catalog entry alone does not mean the model is downloaded or suitable for the recording.

In Sonano, MCP exposes audio tools and the Skill explains workflows. Tool registration, model readiness and file access are separate conditions. Permission to submit transcription does not grant access to arbitrary host paths. External media still depends on host permissions and the available transport.

Create objects that can be revised

After transcription, the creator checks names, quotations and conclusions. An Agent can draft dialogue from the confirmed material, distinguishing source text from adaptation and leaving uncertainty open for review. The useful connection between steps keeps editable objects in view.

A dialogue needs more than speaker names inside one paragraph. Roles need stable identity, lines need addresses and voices need explicit assignments. Sonano’s Agent dialogue projects persist roles, cues and generation candidates, with one generation model per project. That structure does not automatically identify original speakers or guarantee a faithful adaptation.

Make a voice retrievable across tasks

“Use last time’s narrator for the second character” needs an identifiable voice asset. Rewriting a description on every attempt may produce a changing character. Saving the sample and identifier gives the instruction something concrete to reference.

Sonano can save reviewed design candidates and import authorized recordings into the Voice Library. Shared reference voices are not bound to their original design engine, but model and language compatibility still apply. An engine change calls for another listening review. Reuse preserves material; it does not choose a performance for the scene.

A changed line should lead to the affected audio

After the first version, the creator finds that line seven overstates the original claim. The Agent should read the current script, revise that line and generate a new candidate for the affected material. Compatible existing results can be reused, then selected audio can be combined for export.

This requires script versions, chosen takes and job states to correspond. Sonano’s persistent dialogue tools provide cue editing, candidate selection, local regeneration and export from selected audio. An accepted request starts a job; completion, failure and partial delivery still need checking. A disconnection is not automatically cancellation, and an interrupted job after restart is not automatic continuation.

Keep editorial decisions visible

An Agent can check a file and a terminal job state, and can assist with transcript comparison. Those signals alone cannot settle whether a performance respects the interview’s meaning, whether a character is convincing or whether a pause works for the audience.

Three review points suit this scenario: the adaptation’s accuracy, the casting and delivery, and the final release decision. These are proposed human checkpoints, not a claim that every Agent automatically enforces an approval workflow.

Local audio is one part of the route

Running the voice model on a Mac keeps that inference local. If the Agent uses a hosted language model, transcript or script text may still go to its provider. Review the complete material flow for confidential work.

Sonano’s current MCP tools cover speech, transcription, conversion, voice creation and persistent dialogue projects. Audiobooks have separate app and HTTP workflows; a video dubbing timeline is not implemented. This scenario concerns sequential dialogue audio, not promised automatic video dubbing. See the local voice connection guide for setup.

Sources & further reading

Sources checked on November 3, 2026. Creative scenarios and listening examples are illustrative, not executed model tests. App capabilities depend on the current version.