MILabs
Tools

Kokoro text-to-speech for small business voice output

Kokoro is an open-weight text-to-speech model that small businesses can self-host for spoken notices, replies and reminders.

Updated 2026-06-23

What Kokoro is

Kokoro is an open-weight text-to-speech model published by the hexgrad project on GitHub. It takes written text and produces spoken audio, and it runs on hardware the business controls rather than on a hosted service. Because the model can be downloaded, a business can run it on a local computer or a private server and keep its text and audio inside its own environment. Kokoro does not handle speech recognition, so it forms one part of a voice workflow. It is usually paired with components that listen for input, decide what to say and route the result to a phone line, an app or a file. Its output suits read-aloud content, spoken notices and automated voice replies. For teams that want voice output without relying on a hosted voice provider, Kokoro is a practical starting point, and quality, latency and language coverage are the details that shape the final choice. Teams also gain a clear picture of where their voice content lives, which makes ownership easier to assign.

hexgrad: official site

Kokoro in plain language

Speech synthesis, often called text-to-speech, is the step where a computer reads written words aloud. A system takes a sentence, works out how each word should sound, and produces an audio file or a live stream a listener can hear. The quality of that audio depends on the voice model behind it, on how the text is prepared, and on the hardware that runs the model.

Kokoro sits in this category as a model that can be installed on equipment a business already owns or rents. Once it is running, a script can send it text and receive audio back, and the words need not leave the local network. That property matters for teams that handle client names, medical details or contract terms they would rather keep inside their own environment.

Speech synthesis is only one half of a spoken interaction. A phone assistant also needs to hear the caller, understand the request and decide what to say. Kokoro covers the final step, the voice that delivers the reply, so knowing that split helps a business judge what the model can do alone.

What businesses use it for

  • Appointment reminders that clinics and trades play down a phone line
  • Read-aloud copies of quotes, invoices and written notices
  • Voice replies on a phone line that answers callers the team cannot reach
  • Audio versions of staff procedures and training notes
  • Spoken status updates for field teams who cannot read a screen

Who it suits

  • Clinics that want spoken reminders running on infrastructure they control
  • Trades and field businesses that send spoken notices to crews
  • Teams with private documents that should stay on their own servers

How MILabs sets it up

The easiest first step is to say hello through the enquiry form, and we'll take it from there. From that first conversation, the owner chooses where Kokoro will run, either on a computer the business already has or on a private server it controls. We then connect Kokoro to the phone line, booking calendar or document store that holds the text to be spoken. Each connection is tested with real examples so you can hear the voice and check how names, suburbs and product codes are pronounced. Access is set so only named staff can change the voice, the templates or the audio files. Before handover, you and one team lead agree in writing who looks after updates, the server and changes to the spoken scripts.

Builds it fits

Industries it fits

Kokoro in practice

In a working business, this usually starts with text that already exists. A booking system holds appointment times and patient names, a job management tool holds site addresses and task notes, and a document store holds the wording of a standard notice. A small script pulls the relevant fields, fills them into an approved template and passes the finished sentence to Kokoro. The output is an audio clip that a phone system or crew app can play when needed.

Review sits in the middle of that chain. Before a template goes live, someone on the team listens to several rendered examples, checks how surnames, street names and product codes are spoken, and corrects the text where the voice stumbles. Small fixes, such as spacing out a code or writing an abbreviation in full, can change every future clip that uses the same template.

Over time the work settles into a few stable outputs, each with a named owner. A clinic front desk may maintain the reminder wording, while a trades lead maintains the site notice wording. Each change passes through the same review, so the spoken version of a message stays close to the written version the team has approved.

good questions, straight answers

Frequently asked questions

What is Kokoro?
Kokoro is an open-weight text-to-speech model from the hexgrad project. It turns written text into spoken audio. Because the model can run locally, a business can produce speech without sending its text to a hosted voice service. It handles voice output only, so for most teams it is usually combined with other tools for phone lines, scripts and storage.
How does MILabs set up Kokoro?
The engagement opens with a conversation with the owner about where Kokoro should be hosted. We then link it to the phone line, calendar or document store your team already uses. Sample recordings are reviewed against real names and terms, and only named staff can reach the settings. You and a team lead agree who maintains the model and server. We deliver this remotely from Sydney by video call.
Where does Kokoro store data?
Kokoro runs on the machine or server where the business installs it, so the text it reads and the audio it creates remain under that environment's control. Whether anything is saved afterwards depends on the surrounding workflow, such as call logs or a file folder. During setup we record each storage location with the owner, so the business always knows where its copies sit and who can reach them.
When is Kokoro the wrong choice?
Kokoro is a poor fit when speech recognition is the main requirement, because it only produces audio. Another voice model or a full voice platform may serve better when a team needs a specific cloned voice, a language or accent the model covers poorly, or live conversation with very little delay. We compare those options with the owner before any build begins.
Can my team use Kokoro without training?
Yes, most staff never touch the model itself. They hear or download the voice output the workflow produces, such as a played message or an audio file, and an approved script decides what is said. Anyone who edits scripts or voice settings receives a short walkthrough from MILabs, and the owner keeps a written record of who holds those permissions.
let's make something happen

Workflow automation: how we can help

Choose how you want to work on this with MILabs. Both options start with an enquiry.

Discovery is free, no obligation from the first chat.

Let us build it for you

We build it and look after it.

Build it yourself, we'll help

You build it with our direction, teaching and a weekly check-in.