Deploy Piper TTS on a VPS

Piper is a fast, local neural text-to-speech engine designed to run well on modest CPU hardware (it was built with Raspberry Pi in mind), so it's an easy, low-resource deployment on nearly any RamNode VPS.

1. Choose and size the VPS

Piper is lightweight — it is one of the least demanding services you can self-host.

Use case	Recommended RamNode plan specs
Light/personal use, single low-quality voice	1 vCPU, 1–2 GB RAM
Regular use, multiple voices/medium quality	2 vCPU, 2–4 GB RAM
Higher throughput / concurrent requests	4 vCPU, 4–8 GB RAM

Recommended OS: Ubuntu 24.04 LTS.

2. Initial server setup

shell

apt update && apt -y upgrade
apt -y install build-essential python3 python3-venv python3-pip wget unzip ufw ca-certificates

ufw allow OpenSSH
ufw allow 5000/tcp   # the port the wrapper API will use below
ufw enable

Create a dedicated service user and directory layout:

shell

adduser --system --group --home /opt/piper piper
mkdir -p /opt/piper/voices
chown -R piper:piper /opt/piper

3. Install Piper

The simplest path on Linux x86_64 is the prebuilt release binary rather than compiling from source:

shell

su - piper -s /bin/bash
cd /opt/piper

# Check https://github.com/OHF-Voice/piper1-gpl/releases (or rhasspy/piper) for the latest tag
wget https://github.com/rhasspy/piper/releases/latest/download/piper_linux_x86_64.tar.gz
tar -xzf piper_linux_x86_64.tar.gz

This extracts a piper/ directory containing the piper binary and its shared libraries. Verify it before continuing:

shell

cd piper
./piper --help

If you'd rather install as a Python package for easier scripting/integration:

shell

python3 -m venv venv
source venv/bin/activate
pip install --upgrade pip
pip install piper-tts fastapi "uvicorn[standard]"

Both approaches can coexist; the wrapper API below assumes the piper-tts Python package.

4. Download a voice model

Piper voices are distributed as an .onnx model plus a .onnx.json config file, per language/speaker, at varying quality tiers (x_low, low, medium, high).

shell

mkdir -p /opt/piper/voices/en_US-lessac-medium
cd /opt/piper/voices/en_US-lessac-medium

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

Browse the full voice catalog at the piper-voices Hugging Face repo and repeat for any additional languages/speakers you want to serve.

Test synthesis directly from the CLI:

shell

echo "Hello from RamNode." | ./piper/piper \
  --model /opt/piper/voices/en_US-lessac-medium/en_US-lessac-medium.onnx \
  --output_file /opt/piper/test.wav

5. Wrap it in a small HTTP API

Create /opt/piper/server.py:

shell

from fastapi import FastAPI
from fastapi.responses import Response
from pydantic import BaseModel
from piper import PiperVoice
import io, wave, os

VOICE_PATH = os.environ.get(
    "PIPER_VOICE",
    "/opt/piper/voices/en_US-lessac-medium/en_US-lessac-medium.onnx",
)

app = FastAPI()
voice = PiperVoice.load(VOICE_PATH)

class SpeakRequest(BaseModel):
    text: str

@app.post("/speak")
async def speak(req: SpeakRequest):
    buf = io.BytesIO()
    with wave.open(buf, "wb") as wav_file:
        voice.synthesize(req.text, wav_file)
    return Response(content=buf.getvalue(), media_type="audio/wav")

@app.get("/health")
async def health():
    return {"status": "ok", "voice": VOICE_PATH}

Test it:

shell

uvicorn server:app --host 127.0.0.1 --port 5000

6. Run it as a systemd service

As root, create /etc/systemd/system/piper.service:

shell

[Unit]
Description=Piper text-to-speech API
After=network.target

[Service]
Type=simple
User=piper
Group=piper
WorkingDirectory=/opt/piper
Environment=PIPER_VOICE=/opt/piper/voices/en_US-lessac-medium/en_US-lessac-medium.onnx
ExecStart=/opt/piper/venv/bin/uvicorn server:app --host 127.0.0.1 --port 5000
Restart=on-failure
RestartSec=5

[Install]
WantedBy=multi-user.target

shell

systemctl daemon-reload
systemctl enable --now piper
systemctl status piper

7. Put nginx in front (recommended)

shell

server {
    listen 80;
    server_name tts.example.com;

    location / {
        proxy_pass http://127.0.0.1:5000;
        proxy_set_header Host $host;
        proxy_set_header X-Real-IP $remote_addr;
    }
}

shell

apt -y install certbot python3-certbot-nginx
certbot --nginx -d tts.example.com
ufw delete allow 5000/tcp
ufw allow "Nginx Full"

8. Operational notes

Multiple voices: load several PiperVoice instances keyed by name/language in the wrapper if you need more than one voice served from the same API; keep each model in its own subdirectory under /opt/piper/voices/.
Quality vs. speed: medium quality voices are a good default; high sounds noticeably better but is slower per character — test on your actual vCPU count before committing.
Streaming: for long text, synthesize sentence-by-sentence and stream chunks back rather than buffering the whole response, if your client supports chunked audio.
Monitoring: add an HTTP check against /health.
Updates: pip install -U piper-tts (Python route) or re-download the latest release tarball (binary route), then systemctl restart piper.