Most podcast producers are still moving the same episode through three or four applications: one for voices, another for editing, another for mastering. Each handoff creates another place for a file, level, or version to go wrong.
Vois is built for the full job. It gives a scripted show one local project for speaker-tagged writing, voice assignment, timeline arrangement, mastering, and export. That means the production decisions stay with the episode instead of being scattered across folders and subscriptions.
With a prepared script, a 30-minute show can be produced in under an hour on suitable hardware. The valuable part is not the clock. It is having a repeatable workflow that leaves room to listen, revise, and approve the episode before it ships.
Write a script that knows who is speaking
Open a Vois project and write in the script editor. Insert a speaker tag with the slash menu, give it a stable name such as "Host" or "Guest," and write the dialogue after the tag. The tag connects the text to its assigned voice when you generate.
Host Welcome back to Tech Breakdown. Today we're looking at why local AI tools are eating cloud services alive.
Guest Thanks for having me. This is a topic I've been following for two years now, and the shift is accelerating.
Host Let's start with the obvious question. Why would anyone run AI locally when cloud services are so convenient?
Guest Convenience is real. But convenience has a price tag, and most people don't realize how steep it gets at scale.
Write the way people talk. Let one person give a full answer, then let the other interrupt or react. That uneven rhythm does more for a scripted conversation than a pile of stage directions.
Assign a stable voice to every role
Open the voice library and assign one voice to each speaker tag. Vois includes 100+ production voices across 21 categories, so the host and guest can be clearly different without making either one a caricature. Test the pair with a minute of your real dialogue, not a generic demo sentence.
If you have permission to use a recorded voice, Vois also supports local voice cloning from a clean sample. Whether you choose a library voice or a clone, keep the assignment stable across the season. A listener should not have to work out who is speaking in episode six.
Generate, listen, and revise the line that needs it
Generate the script in the project. Each tagged section uses the voice assigned to that role. If emphasis lands in the wrong place or a line feels rushed, review the line in context and regenerate that section rather than settling for it because a quota is running down.
For a 30-minute script, generation time depends on the selected engine and your computer. Use that waiting time to prepare show notes or review the opening and closing copy. The finished audio still deserves a listening pass.
Shape the conversation on the timeline
Generated sections land in Vois's multi-track timeline, where you can trim clips, set gaps, and add crossfades. The timeline's undo history makes small experiments safe.
Use gaps to support the conversation:
- Quick exchanges can have short gaps, around 0.1 to 0.2 seconds.
- A substantial point benefits from a slightly longer beat.
- An occasional overlap can make a reaction feel less mechanical.
Keep each adjustment in service of intelligibility. If a listener cannot tell who spoke or where a thought ended, the edit is too aggressive.
Master once for the platform you chose
Select the Vois export preset for the platform that will carry the show. The professional mastering controls handle loudness normalization, de-essing, EQ, and limiting, then export in WAV, MP3, FLAC, or AAC as needed. Spotify and YouTube commonly target -14 LUFS; Apple Podcasts commonly targets -16 LUFS. Check the current requirements for your host before final export.
You can adjust the controls manually, but the preset gives a consistent starting point from episode to episode. Listen to the finished export on headphones before uploading. A correct loudness number is not a substitute for a clean listening experience.
Prompt your agent for the repeatable work
If you use an external coding agent with the Vois CLI, ask it to prepare work for you, not to publish unattended:
Prompt your agent: "For the current Vois podcast project, inspect the speaker tags and list any unassigned speakers, repeated pronunciation risks, and sections longer than two minutes. Prepare a proposed generation order and a review checklist. Do not generate, export, upload, or change project settings until I approve the plan."
Then keep the human approval points explicit:
- Review the proposed speakers, pronunciation list, and generation order.
- Approve or adjust voice assignments in Vois.
- Generate the approved sections and listen to the flagged passages.
- Approve the timeline and mastering preset after a full episode listen.
- Export, upload, and schedule the episode yourself.
Turn the workflow into a weekly show
Keep a reference episode with the approved speaker assignments, opening and closing material, pacing notes, and mastering choice. Before each new episode, compare its setup against that reference. The effort moves from rebuilding the production chain to making the next script worth hearing.
| Stage | What you review |
|---|---|
| Script | Speaker labels, names, and conversational rhythm |
| Voices | Clear contrast and stable assignments |
| Generation | Pacing, emphasis, and pronunciation |
| Timeline | Transitions, silence, and interruptions |
| Mastering | Platform preset and final listening pass |
| Export | Format, metadata, and upload details |
The full studio matters because a podcast is not simply a generated voice file. It is an edited, mastered, and recognizable show. Get started with Vois, then see the pricing page when you are ready to make a repeatable production workflow part of your schedule.
The Vois Team