A small self‑contained web app for offline text‑to‑speech playback on Linux.
It runs a local HTTP server, generates speech using Sherpa‑ONNX, and plays audio through aplay on Linux.
The UI lets you paste text, adjust speaker and speed, and monitor playback progress in real time.
This tool is designed to be simple, lightweight, and easy to hack.
- Fully offline TTS
- Simple web interface
- Sentence‑by‑sentence playback
- Adjustable speaker ID
- Adjustable speech speed
- Live progress bar
- Shows currently spoken sentence
- Optional startup text file
- No database, frameworks, or external services
- Linux
- Go 1.22+
alsa-utils(providesaplay)- Sherpa‑ONNX Go bindings
- A compatible TTS model
# Valiadte sha256sum
echo "$(cat go-speak-1.0.4-linux-amd64.deb.sha256)" | sha256sum --check
# go-speak-1.0.4-linux-amd64.deb: OK
# install
sudo dpkg -i go-speak-1.0.4-linux-amd64.deb
Clone the repository:
git clone git@github.com:wasmup/go-speak.git
cd go-speak
go install -trimpath -ldflags="-s -w"export GOAMD64=v1
echo $GOAMD64
go clean
./build-deb.sh
cd ./build/
echo "$(cat go-speak-1.0.4-linux-amd64.deb.sha256)" | sha256sum --check
ls -lh go-speak-1.0.4-linux-amd64.deb
dpkg-deb -f go-speak-1.0.4-linux-amd64.deb
# install
sudo dpkg -i go-speak-1.0.4-linux-amd64.deb
# remove
# sudo dpkg --purge go-speak
# run
/opt/go-speak/go-speak
# run with model dir:
/opt/go-speak/go-speak -m /opt/go-speak
go version -m /opt/go-speak/go-speak
Select/test your model here
Download TTS model here, for example:
wget -c https://github.com/k2-fsa/sherpa-onnx/releases/download/tts-models/vits-piper-en_US-libritts_r-medium.tar.bz2
tar -xjf vits-piper-en_US-libritts_r-medium.tar.bz2
Expected root model directory layout:
/opt/go-speak/
└── vits-piper-en_US-libritts_r-medium
├── en_US-libritts_r-medium.onnx
├── tokens.txt
└── espeak-ng-data/
└── vits-piper-en_US-amy-medium
├── en_US-amy-medium.onnx
├── tokens.txt
└── espeak-ng-data/
Default model directory:
/opt/go-speak
You can override it with -m.
/opt/go-speak/go-speak -m /opt/go-speak -name "vits-piper-en_US-libritts_r-medium"
Open your browser:
http://127.0.0.1:8080
-addr HTTP listen address
-i Startup text file (optional)
-m TTS model root directory which may contain more the one model
-name pre selected TTS model name ("vits-piper-en_US-libritts_r-medium")
-sid Default speaker ID
-speed Default speech speed
Example:
/opt/go-speak/go-speak -addr 127.0.0.1:8080 -m ~/tts -sid 7 -speed 1.0
Load a startup text:
/opt/go-speak/go-speak -i speech.txt
Controls:
Play
Start speaking the text.
Stop
Stop playback immediately.
Speaker ID
Select the voice speaker (depends on the model).
Speed
Adjust speech speed (0.2 – 4.0).
- The text is split into sentences.
- Each sentence is synthesized using Sherpa‑ONNX.
- Generated audio is streamed to
aplay. - The browser polls
/progressto update UI state.
Playback runs in a background goroutine with cancellation support.
Exclude chars
Characters entered here are removed from the text before playback.
Example: *#
Start playback.
Form parameters:
text=Hello world.
exclude=*#
Response:
{
"text": "Hello world."
}
Stop playback.
Returns playback status.
Example:
{
"done": 2,
"total": 5,
"playing": true,
"percent": 40,
"current": "This is the sentence being spoken."
}
Set speaker ID.
sid=7
Set speech speed.
speed=1.2
Allowed range:
0.2 – 4.0
main.go server startup
app.go application state and playback loop
handlers.go HTTP handlers
progress.go progress tracking
audio.go audio playback via aplay
tts.go Sherpa TTS initialization
config.go CLI flags and configuration
sentences.go sentence splitting
index.html embedded web UI
- Minimal dependencies
- Easy to understand
- Easy to modify
- Fully offline
- Small codebase
- Only tested with Piper VITS models.
- Playback currently depends on
aplay(ALSA). - Sentence splitting is simple regex/dictionary cleaner then sentence splitter and optional LLM cleaner per sentence.
MIT License