Reads to you. Tells no one.
SoftSpoken is a free Windows tray app that reads any text you select — in any
program — out loud with offline neural voices. Select text, click the floating
play button (or press Ctrl+Alt+R), and listen. Nothing you read ever leaves
your PC.
| SoftSpoken | Typical alternatives | |
|---|---|---|
| Price | Free, forever | $100+/year subscriptions |
| Privacy | 100% offline — no account, no cloud, no telemetry | Text is sent to cloud servers |
| Where it works | Any app: browser, PDF reader, Word, chat, code editor | Browser-only extensions, or one vendor's apps |
| Voices | Neural (Piper) + Windows natural voices + classic SAPI, merged in one list | Locked behind the paywall |
| Control | Sentence-level skip back/forward, floating transport bar | Play/stop only |
| Sound | Built-in effects chain: de-esser, pitch, EQ, reverb, presets | None |
| Footprint | Portable folder, no admin, no runtime to install | Installers, background services |
- Fully offline. The neural voices run on your CPU. Aeroplane mode works.
- Genuinely private. No account, no analytics, no network calls. What you read is nobody's business — the code is right here to check.
- System-wide. Reads selections from any application, not just a browser tab: the floating ▶ button appears wherever you select text.
- Listenable for hours. The "Soft and natural" preset tames the sibilance and flat delivery that make synthetic speech tiring.
- A real transport. Skip back a sentence when your mind wandered — like a podcast player, not a tape recorder.
- Save narration. Render any text to a
.wavwith your chosen voice and effects — offline, faster than realtime. - Yours to keep. Remappable hotkeys, portable install, MIT licensed.
Two options on the Releases page:
| Size | What you get | |
|---|---|---|
| SoftSpoken-2.0.0-win64.zip | 291 MB | Everything — app, engine, four British neural voices. Unzip and run. |
| SoftSpoken-2.0.0-lite.zip | 148 KB | App + full source. Run fetch-runtime.ps1 once to pull the engine and voices. |
Unzip anywhere and run SoftSpoken.exe. Optionally run install.cmd for a
Start-menu shortcut (per-user, no admin).
SmartScreen note: the binary is unsigned (signing certificates cost money; this app is free). The first launch may show "Windows protected your PC" — click More info → Run anyway, or build from source below and the warning never appears. A winget package is prepared, which avoids the prompt entirely.
No SDK, no Visual Studio, no NuGet — the build uses the C# compiler that ships inside Windows:
git clone https://github.com/Prateek-LN2002/SoftSpoken.git
cd SoftSpoken
powershell -ExecutionPolicy Bypass -File fetch-runtime.ps1 # engine + voices (~310 MB)
build.cmd
Three ways to read:
- Select text with the mouse → click the play button that appears.
- Select, then right-click → the play button appears beside the menu.
- Select and press
Ctrl+Alt+R.
While reading: Ctrl+Alt+P pause/resume · Ctrl+Alt+S stop ·
Ctrl+Alt+←/→ previous/next sentence — or use the floating bar. Every
shortcut is remappable in the app window (tray icon → Open SoftSpoken).
A voice is two files, name.onnx + name.onnx.json, dropped into Voices\.
The library (50+ languages): huggingface.co/rhasspy/piper-voices.
Windows' own natural voices install under Settings → Accessibility → Narrator.
test.cmd
94 checks: sentence chunking, settings round-trips and corruption handling, WAV parsing, the DSP chain (NaN/peak safety on every preset), engine clamps, hotkey parsing, and live integration against the real Piper engine — including a regression for the UTF-8 bug that once stopped smart-quoted text after two sentences. Exit code 0 means green.
SoftSpoken makes no network connections. The only file it writes outside
its own folder is settings.ini in %APPDATA%\SoftSpoken. The optional
fetch-runtime.ps1 downloads the Piper engine and voices from GitHub and
Hugging Face — once, at your command, and that is the only download in the
project.
- Piper — the offline neural TTS engine
- rhasspy/piper-voices — the voice models
MIT — use it, learn from it, fork it. This repository is maintained by its owner; contributions are welcome as pull requests but only the owner can change what's published here.