All work
Desktop appOpen source

Talkr

An open-source desktop app for private, offline speech-to-text and text-to-speech. Whisper, Kokoro and Piper run on your own hardware, with no account, API key or cloud upload.

Product
Offline speech desktop app
Built with
Tauri, whisper.cpp, sherpa-onnx
Talkr model library showing Whisper models and hardware details
100%

offline once a model is downloaded

17

models and voices in the library

8 GB

RAM is enough; no GPU required

Overview

The problem we set out to solve.

Talkr turns text into natural speech and speech into timestamped text, entirely on your own computer. It ships for every platform with signed installers and an in-app updater.

  • Most speech tools send your voice and text to the cloud.
  • Local AI models crash on some GPUs and drivers, freezing the whole app.
  • It had to run on modest hardware, down to a four-core CPU with no GPU.
Talkr Transcribe screen with the record button
Talkr searchable history of speech and transcripts
Talkr Speak screen for text-to-speech
What we built

Every part, built to work in the real world.

Speak

Paste up to 5,000 characters and generate natural speech, with adjustable speed and WAV export.

Transcribe

Record or drop in a file for timestamped text, language detection and TXT/SRT export.

Model library

11 Whisper models and 6 voices, each with size, languages and licence shown before download.

Searchable history

Every generation and transcript stored locally, grouped by day, searchable and filterable.

Hardware aware

Uses the GPU when it works, falls back to the CPU when it does not.

Safe updates

Updates install only after a click, and each one is signature-verified.

How it's built

The speech engine runs as an isolated Rust sidecar that talks to the app over JSON lines. If a model or GPU driver crashes, only the engine fails: the interface stays up, and Talkr retries on the CPU and explains what happened.

  • Tauri 2 & RustDesktop shell and speech engine
  • whisper.cppSpeech-to-text
  • sherpa-onnxKokoro and Piper voices
  • React 19, Vite & ZustandThe desktop interface
  • SQLiteLocal, searchable history
  • Next.js on VercelMarketing and download site
The outcome

A shipped, cross-platform product with a full release pipeline: tagged releases build and sign installers, publish checksums and feed the in-app updater.

Lilac Pearls home page on desktop and mobile
Next project

Lilac Pearls

E-commerce store

Have an idea? Let's build it.

Tell us what you want to build. You'll get a clear plan, a timeline and a fixed price within a week.