Automation matters when the repetitive part of voice production starts hiding the creative work. A season of episodes, a localized training library, or a documentation site should not force you to rebuild the same project by hand. It also should not let an agent publish a large batch without a human seeing what will change.
Vois gives an agent a local production surface for the work that happens after the script is approved: it can inspect the running desktop app, read the exact voice IDs, build projects, generate audio, apply export profiles, and report the results. Your script, casting decisions, and final release approval stay with the people responsible for them.
The seven recipes below replace command snippets with a safer pattern: give the agent a constrained brief, get a specific expected output, review it, then grant the next approval. For command details, attach the Vois CLI skill to your agent.
Recipe 1: Batch-generate a podcast season
The problem: You have a folder of approved episode scripts. Building projects, assigning the host, and starting a season-wide render by hand invites small inconsistencies.
Why Vois fits: The agent can create one local project, preserve the same reviewed host voice across the season, and return a production plan before it queues a single episode.
Prompt your agent
Vois is open on this machine. Prepare a batch-production plan for podcast season [SEASON NAME] from the approved scripts in [SCRIPT DIRECTORY].
Verify that Vois is available, inspect the complete voice list, and use only the exact approved host and guest voice IDs in the cast sheet I provide. Identify every script that can be imported, the speaker assignments each one needs, the selected engine, and the proposed project structure.
Return a dry-run report with the episode count, filenames found and skipped, speaker-to-voice mapping, engine, any missing cast entry, and the generation order. Do not create a project, import a script, generate audio, or overwrite an existing episode until I approve the plan.
Review and approval flow
- Compare the dry-run episode count with your release slate and resolve every skipped or missing file.
- Check the exact voice IDs and speaker names against the approved cast sheet.
- Approve the project build. Ask the agent to return project and script IDs plus a per-episode generation report.
- Listen to a representative host and guest exchange before approving the full season export.
Expected output: One local Vois project, one script record per approved episode, a stable speaker map, and a report that labels each episode completed, pending, or failed.
Recipe 2: Change a voice across an established project
The problem: A new voice suits the show better, but a project already has multiple episodes and published material.
Why Vois fits: Voice assignments live with the project, so an agent can show you the scope of a change before it touches an existing render.
Prompt your agent
Review the Vois project [PROJECT NAME] for a proposed cast change: replace [SPEAKER] with the approved exact voice ID [VOICE ID].
List every affected script, its current generation state, the current and proposed voice ID, and whether the script has already been exported. Do not change a speaker assignment or regenerate any audio.
Return an impact report and a recommended review sample: three representative scenes that cover the character's ordinary, emotional, and technical delivery. Wait for my approval after I audition the sample.
Review and approval flow
- Audit the impact report, especially scripts that have already shipped.
- Approve a small sample render first, then listen in context.
- If the change is accepted, approve a targeted regeneration and have the agent report every affected script and output.
- Update the cast record and release notes before replacing published assets.
Expected output: A scoped voice-change report followed, only after approval, by a targeted regeneration record. Nothing should silently re-render the entire project.
Recipe 3: Export one approved mix for each destination
The problem: A finished episode may need separate assets for YouTube, Spotify, Apple Podcasts, or an audiobook workflow. Reusing a generic file can create avoidable delivery problems.
Why Vois fits: Its export profiles pair the approved audio with a platform-targeted export in the same production workspace.
Prompt your agent
For approved Vois project [PROJECT] and script [SCRIPT], prepare an export plan for [DESTINATIONS].
Inspect the available export profiles and map each requested destination to the correct supported profile and output format. Return the source script revision, mastering profile, proposed output filename, destination, and any destination that needs a human delivery-specification check.
Do not export files until I approve the plan. After approval, export only the listed destinations and return the completed paths, profile used, and any failed export.
Review and approval flow
- Confirm the destination list and any distributor-specific specifications before export.
- Approve the filename and output location for each destination.
- Review the export report and spot-check the final files before delivery.
Expected output: A small, traceable export matrix rather than a pile of unlabelled audio files. The audio export feature explains the available Vois production workflow.
Recipe 4: Produce reviewed multilingual versions
The problem: A training module or episode needs several language versions without turning each one into an unrelated production.
Why Vois fits: The same local workspace can keep the project, source revision, casting notes, and language-specific scripts together. Omni provides broader language coverage on eligible plans, while the legacy multilingual engine supports its established language set.
Prompt your agent
Prepare multilingual Vois production for approved source module [MODULE] in [LANGUAGES].
For each language, verify that I supplied a reviewed translation, approved character or narrator voice ID, target engine, language reviewer, and output naming convention. Return a language-by-language readiness table. Flag missing translations, unreviewed text, unsupported voice or engine combinations, and missing reviewers.
Do not generate translations, alter text, or render audio until I approve the readiness table. After approval, generate only the approved language versions and return an output report keyed by locale and source revision.
Review and approval flow
- Have a qualified reviewer approve each translation before it enters the queue.
- Check voice and engine choices against the intended language, tone, and project plan.
- Approve a short sample in every language before approving the full render.
- Keep the locale, translation revision, reviewer, voice ID, and output path with each delivered asset.
Expected output: A localization matrix that shows what is ready, what is blocked, and which Vois output belongs to each language version.
Recipe 5: Monitor a batch without guessing
The problem: A larger generation run can leave you unsure whether every episode or module finished, or whether a problem needs attention.
Why Vois fits: An agent can query the local production state and translate it into a concise report instead of asking you to watch terminal output.
Prompt your agent
Inspect the current Vois generation state for project [PROJECT]. Do not start, cancel, retry, or export anything.
Return a batch-status report with each script's title or ID, state, completed segment count, pending work, failures, output paths when available, and the most likely action for each failure. Summarize total completed, pending, and failed items.
If anything has failed, prepare a targeted retry plan that names only the affected scripts and the suspected prerequisite. Wait for approval before retrying.
Review and approval flow
- Read the totals first, then investigate any item with a mismatch or failure.
- Fix the source text, voice binding, app state, or plan access named in the report.
- Approve a targeted retry and compare the final report with the original batch plan.
Expected output: A human-readable production status and, when needed, a narrow retry plan. A failed item should never disappear inside a batch.
Recipe 6: Maintain pronunciation across a catalogue
The problem: Brand names, technical terms, and proper nouns are easy to mispronounce, and an improvised spelling workaround makes a script hard to maintain.
Why Vois fits: The pronunciation dictionary gives the production team a controlled place to review a spoken form and reuse it across the relevant work.
Prompt your agent
Prepare a pronunciation review for the approved term list in [SOURCE FILE] and the Vois scope [GLOBAL OR PROJECT].
For each term, return the written term, proposed spoken form, intended context, affected projects or modules, and whether it already has a dictionary entry. Do not add, change, import, or export a dictionary entry yet.
After I approve the review, apply only the approved entries and return a change report with the final terms, scope, affected projects, and any conflicts.
Review and approval flow
- Have a subject matter owner or native speaker approve specialist terms.
- Listen to a representative sentence, not an isolated word alone.
- Approve the narrowest appropriate scope, then preserve the resulting dictionary revision with the project.
- Regenerate only assets affected by an approved pronunciation change.
Expected output: A reviewed pronunciation change set, not an opaque global change. See the pronunciation dictionary feature for the Vois workflow.
Recipe 7: Keep documentation audio aligned with approved pages
The problem: Documentation changes regularly. Audio versions drift when updates are treated as a manual afterthought.
Why Vois fits: An agent can prepare a local update set from changed, approved documents while Vois retains the project and voice context. The human release owner decides what is ready to publish.
Prompt your agent
Review the approved documentation changes in [DOCS DIRECTORY] for release [RELEASE ID]. Vois is open on this machine.
Identify only changed documents that have an approved source revision. For each, return the document path, current Vois script or project match, narrator voice ID, engine, proposed output filename, source revision, and whether a refreshed audio asset is required. Do not generate, export, commit, deploy, or replace site assets.
Return a documentation-audio release plan that separates ready, blocked, unchanged, and missing-mapping pages. Wait for my approval before building audio for the ready set.
Review and approval flow
- Confirm that every page in the ready set is final and has the correct title, narrator, and output path.
- Approve the local generation set. Have the agent return the project and script IDs, generated asset paths, and failures.
- Listen to changed sections and approve the audio assets separately from the documentation deployment.
- Let the documentation release owner publish only the reviewed assets and retain the release report.
Expected output: A release-scoped map from approved document revision to reviewed audio asset, with no hidden deployment step.
Use automation to make review easier
These recipes work because the agent has a narrow job: prepare facts from Vois, carry out approved production work, and return a record a person can inspect. The human has the responsibility an agent cannot assume: approving copy, casting, quality, destination, and publication.
Start with one repeatable workflow, such as a season plan or pronunciation review, then reuse the report format as your library grows. For the underlying capabilities, see CLI and AI automation. Get started when you are ready to run your own local workflow, or review current plans.
The Vois Team