Bulletins at 4:20am
Draw your coverage area, set your dayparts, and traffic bulletins write, voice and deliver themselves. An empty incident list gets the clear-roads line you wrote — it never invents one.
Our own speech engine
Every voice your work needs. In seconds.
Generate lifelike speech, clone voices from a single clip, convert one performance into another, transcribe, dub and publish — from script to finished, mixed audio.
- 10,000
- free characters a month
- 1 clip
- to clone a voice
- 44.1 kHz
- WAV masters, no watermark
The Working Parts
What it does, and what it refuses to fake
We run the speech engine ourselves, so a clone works the moment it uploads and no vendor takes a cut of every render after. These cards name the numbers — and the limits that come with them.
Warnings lead first
Weather bulletins lead on the active warning rather than the weekend outlook, in your market's own units and clock.
Five ways into playout
Bulletins land by pull URL, webhook, FTP, SFTP or Dropbox — overwrite one filename for a watch folder, or stamp each one to archive it.
Cuts through the bed
Voice and bed through a real broadcast chain — 75 Hz shelf −6, 400 Hz −3, 6 kHz +4, then compressor and limiter — ported from one already going to air. Mixing costs no characters.
Fix one word
Click the word that came out wrong, retype it, and it re-renders in the same voice and splices back with a crossfade. The rest of the take is untouched.
Undo never re-bills
Cuts, fades and gain are free and endlessly reversible, and undoing a re-rendered word re-splices rather than paying twice. Saving writes a new clip, so the master survives forty edits.
Names said right
Write the respelling once — Decatur becomes Deekayder — and studio takes, API calls, project blocks and the 4am bulletin all inherit it.
Voice a fix
Flag a mangled word, say it properly into your mic, and it comes back in the voice that got it wrong. The engine's mistake, so it is on us rather than your quota.
Written to the slot
Copy cut to a real word budget — 72 to 84 words for a :30 — and measured again on the way back, from an email, a web page or a document.
Forty tags, one script
Put a tag in the copy, paste the rows out of your sheet, and get one take each — up to 500, priced before you commit and delivered as a project you can retry row by row.
Fix one paragraph
A book is chapters and paragraphs, each rendered on its own, so a typo in chapter forty re-reads one paragraph. It renders on a queue, so you can close the tab.
Rescue a bad recording
Four purpose-built chains rather than one Enhance button — the interview chain keeps the pauses, because in an interview the pause is often the answer. Costs no characters.
The tags we refuse
A pause marker gives you exactly the silence you ask for, to a tenth of a second. We do not offer laugh or sigh tags — we measured them on this engine, and they do not work.
Pin the take
Every take shows the seed it came from. Pin the one the client approved and it comes back identically on any later render.
No vendor per take
Zero-shot cloning on an engine we run ourselves: one clean clip, usable the moment it uploads, with nobody taking a cut of every render after. 200 voices ship with it.
Failed takes cost nothing
Quota is claimed before a render starts and refunded the moment one fails, so you pay for finished work. Transcription bills at a tenth of the synthesis rate.
It asks first
Describe the job and the agent drives the same API you can, at your own role, proposing anything destructive for you to confirm. Every call is logged, including the refused ones.
One spec, no drift
The reference, the code samples, the playground and the OpenAPI document all generate from one spec, so they cannot disagree. One key drives both REST and MCP.
How it works
Three steps, and the first one is optional
- 1
Choose or clone a voice
Start from the premade library, or upload a clip and have your own voice ready before the kettle boils.
- 2
Write, upload or point at a file
Type a script, drop in a recording to re-voice, or push text through the REST API from your own pipeline.
- 3
Ship it
Download the master, export a whole project, or let a webhook tell your system the render has landed.
For creators
The publishing work that isn’t performing
Sponsor reads, pickups, dubs, back-catalogue narration — the audio work between recordings. Clone your own voice once and the workspace handles the takes you’d rather not re-record, for a podcast feed, a channel or an audiobook.
Clone yourself once
One clean clip and your voice lives in the workspace — sponsor reads, intros and corrections without setting the mic up for a two-line fix.
Fix the flub, keep the take
Retype the word that came out wrong and it re-renders in the same voice. The episode keeps its energy; you keep your afternoon.
Publish past your language
Send an English episode, audio or video, and dubbing returns it re-voiced in another language. Same cut, new audience.
Ship the back catalogue
Projects split a manuscript into chapters, render while the tab is closed, and export one clean master.
For broadcasters
Traffic on the fives, with nobody in the building
A live incident feed for the roads your listeners drive, ranked down to the few that are worth airing, written in your station’s wording and read in your station’s voice — on the schedule you set, whether or not anyone is on shift.
Select your coverage area
Draw the circle your signal covers and name the roads in and out of it. Those are the names a bulletin leads on.
Choose a voice, or clone your talent
Any voice in the workspace reads it — including a clone of the presenter who would have read it live.
Customize it for your station
Your intro, outro and sponsor line, how many incidents make the cut, and what to say when the roads are clear. Paste in old bulletins as house style.
Set your schedule
Dayparts rather than one cadence — every ten minutes through drive, hourly overnight — each built from what the feed says at that moment.
And the weather, on the same rails
Pin a market and the forecast gets the same treatment: warnings first, your units, your dayparts. Traffic and weather share the delivery machinery, so setting up one sets up the other.
Bulletins are metered like everything else: the finished script's characters come off the same monthly allowance as the studio and the API, and a failed render is never billed. No per-minute rate, nothing extra per coverage area.
Set up a coverage areaFor publishers
The manuscript is the input
An ebook, a course, a back catalogue nobody ever recorded — long-form projects turn a manuscript into a narrated master without booking a booth, and every step of it is also reachable from the API.
Split it into chapters
Paste the manuscript and it becomes chapters and paragraph blocks, each rendered on its own.
Cast the narrator
Any voice reads it — a premade narrator, a designed one, or a clone of the author — and blocks can switch voices.
Teach it the names
The pronunciation dictionary applies to every block, so the protagonist's name is right in chapter one and chapter forty alike.
Render while the tab is closed
Batches run on the server and a webhook says when the book is done. Export one master, or pull chapters through the API.
Pricing
Pay for characters, not for seats
Every plan includes the whole studio. What changes is how much you can generate, how many voices you can keep, and — for batches, projects, mixes and dubs — where that work sits in the queue when the platform is busy. Invite the whole team on any of them.
Free
Free
Try the whole platform on a small monthly allowance.
10,000
characters / month
- 10,000 characters per month
- 3 custom voices
- Instant voice cloning
- Speech to text
- Non-commercial use
Starter
$5/month
For solo creators shipping regular work.
30,000
characters / month
- 30,000 characters per month
- 10 custom voices
- Commercial licence
- Dubbing
- API access
Creator
Popular$22/month
Higher quality settings and professional voice cloning.
100,000
characters / month
- 100,000 characters per month
- 30 custom voices
- Professional voice cloning
- Long-form projects
- 44.1 kHz WAV downloads
Pro
$99/month
For studios and production teams with a queue to clear.
500,000
characters / month
- 500,000 characters per month
- 160 custom voices
- Everything in Creator
- Priority in the render queue
Scale
$330/month
Volume production with room to grow.
2,000,000
characters / month
- 2,000,000 characters per month
- 660 custom voices
- Everything in Pro
- Top priority in the render queue
Prices in USD. Characters are counted on the text you send; failed renders are never billed.
Queue priority is a head start, not a separate lane, and it applies to work that goes on the background queue — batches, long-form projects, mixes, dubs, multivoice documents and voice previews. A render you start in the studio, and a call to the text-to-speech API, run inside their own request and never queue, so those are the same speed on every plan.
Where it does apply it is bounded on purpose: within a class of work, no job is ever overtaken by work queued more than 8 minutes after it, so free-plan work still drains rather than waiting behind an endless paid queue. Class comes first and plan only breaks ties inside one, so nothing anybody pays lets a background sweep outrank somebody else’s live render. It orders work and nothing else: no plan on this page carries an uptime commitment, a service-level credit or a support response time. (Account credit, the dollar balance a promotional code grants, is a different thing entirely — it offsets a charge and promises nothing about availability.) The Terms of Service are what govern.
For developers
One key, one endpoint, an MP3 back
The API is the same engine the studio uses — same voices, same settings, same quota. Issue a key from the workspace, keep it in your environment, and render from anywhere.
- Keys are hashed at rest and scoped per surface — tts, sts, stt, voices.
- Every request is metered against the workspace, whoever made it.
- Webhooks fire when a long render finishes, so nothing has to poll.
POST /api/v1/text-to-speech/{voice_id}
curl -X POST https://voicebuddy.radioworkflow.com/api/v1/text-to-speech/{voice_id} \
-H "xi-api-key: $VOICEBUDDY_API_KEY" \
-H "Content-Type: application/json" \
-d '{
"text": "Episode three is live. [pause:0.8] Here is what changed.",
"voice_settings": { "stability": 0.5, "similarity_boost": 0.75 },
"output_format": "mp3"
}' \
--output take-01.mp3
Hear your first line in about a minute
Ten thousand characters a month on the free plan. No card, no sales call, no watermark on the audio.