jbpiper

A local text-to-speech server powered by Piper TTS and FastAPI. Sends synthesized audio to a PulseAudio device.

Prerequisites

• Python 3.14+
• PulseAudio (for paplay)
• espeak-ng (required by Piper)

Setup

1. Clone the repo and create a virtual environment:

git clone https://github.com/JamesBlackman-SLG/jbpiper.git
cd jbpiper
python -m venv venv
source venv/bin/activate

2. Install dependencies:

pip install fastapi uvicorn piper-tts

3. Download a Piper voice model. The default config expects en_US-lessac-medium:

curl -LO https://huggingface.co/rhasspy/piper-voices/resolve/main/en/en_US/lessac/medium/en_US-lessac-medium.onnx
curl -LO https://huggingface.co/rhasspy/piper-voices/resolve/main/en/en_US/lessac/medium/en_US-lessac-medium.onnx.json

4. Update the DEVICE variable in server.py to match your PulseAudio sink. List available sinks with:

pactl list short sinks

Usage

Server

Start the FastAPI server:

source venv/bin/activate
uvicorn server:app --host 0.0.0.0 --port 8000

Then send a POST request:

curl -X POST http://localhost:8000/speak \
  -H "Content-Type: application/json" \
  -d '{"text": "Hello, how are you doing?"}'

Standalone script

For quick testing without the server:

source venv/bin/activate
python speak.py

Edit the TEXT variable in speak.py to change what is spoken.