piper-local-tts-demo

Local text-to-speech. Your machine, your voice, no cloud required.

Run Piper — a fast, offline TTS engine — on your own hardware. Speak any text aloud without sending a single byte to the internet.

---

Why this matters

Most text-to-speech today routes through cloud APIs — every word goes to Google, Amazon, or Microsoft. Piper flips this:

• Runs entirely on your machine — laptop, desktop, Raspberry Pi
• Works offline — no internet after setup
• No ongoing costs — nothing per word, per month, ever
• Private by default — your text stays on your hardware

This is the difference between renting a voice and owning one.

---

Quick start

No repo to clone — install straight from GitHub:

pip install git+https://github.com/arthursilas-ai/piper-local-tts-demo.git
curl -sO https://raw.githubusercontent.com/arthursilas-ai/piper-local-tts-demo/main/download_voice.sh && bash download_voice.sh
piper-speak "The future belongs to those who own their tools."

Installs five commands: piper-speak, piper-server, piper-announce, piper-batch, piper-voices.

Or as an agent skill: npx skills add arthursilas-ai/piper-local-tts-demo

Working from a clone instead:

# 1. Install Piper
pip install piper-tts

# 2. Download a voice (~60MB, one time)
./download_voice.sh

# 3. Speak something
python speak.py "The future belongs to those who own their tools."

That is the whole thing. No account, no key, no internet after step 2.

---

What is in this repo

File	What it does
speak.py	Speak any text from the command line or stdin
home_announcer.py	Practical home announcement system with templates
batch.py	Speak or export a text file line by line
server.py	Local HTTP server — curl to speak, works with Home Assistant / Node-RED
list_voices.py	Browse and download Piper voices by language
download_voice.sh	Download the default British English voice

---

speak.py — basic usage

# Say something
python speak.py "Dinner is ready."

# Save to a WAV file instead of playing
python speak.py --output announcement.wav "Meeting in five minutes."

# Read from stdin (pipe-friendly)
echo "Task complete." | python speak.py

# Use a different voice
python speak.py --voice voices/en_US-lessac-medium.onnx "Hello from the US."

---

server.py — HTTP endpoint for home automation

Run a local HTTP server so any script, smart home system, or browser tab can trigger speech:

python server.py              # starts on port 5000
python server.py --port 8080  # different port

Then from anywhere on the same machine (or your local network):

# Speak something
curl "http://localhost:5000/speak?text=Dinner+is+ready"

# POST with JSON
curl -X POST http://localhost:5000/speak \
  -H "Content-Type: application/json" \
  -d '{"text": "Motion detected in the kitchen."}'

# Built-in templates
curl http://localhost:5000/template/time     # "The time is 3:45 PM."
curl http://localhost:5000/template/washing  # "The washing machine has finished."
curl http://localhost:5000/template/morning  # "Good morning. Today is Saturday..."

# Download the WAV instead of playing (useful for remote clients)
curl "http://localhost:5000/speak?text=Hello&download=1" -o hello.wav

# Health check
curl http://localhost:5000/health

Browser UI: open http://localhost:5000 for a simple web interface.

Home Assistant integration

# configuration.yaml
shell_command:
  announce: 'curl -s "http://localhost:5000/speak?text={{ message }}"'

# Then in an automation:
service: shell_command.announce
data:
  message: "Front door opened."

Node-RED integration

Use the HTTP Request node: POST http://localhost:5000/speak with JSON body {"text": "{{ msg.payload }}"}.

---

home_announcer.py — smart home announcer

A class-based announcer with templates, urgent repeat, and a real attention chime:

python home_announcer.py "Washing machine finished"
python home_announcer.py --urgent "Front door opened"   # says it twice
python home_announcer.py --chime "Package delivered"    # beep then speak
python home_announcer.py --time                         # "The time is 3:45 PM."
python home_announcer.py --template washing
python home_announcer.py --list-templates               # show all templates

Available templates: time, reminder, timer, door, washing, weather.

---

batch.py — process a text file

Speak or export multiple lines in sequence:

# Speak every line in a file
python batch.py announcements.txt

# Export as numbered WAV files
python batch.py announcements.txt --export wav_output/

# Named exports: prefix each line with "name: "
echo "morning: Good morning, it is eight o clock." >> named.txt
echo "wash: The washing machine has finished."     >> named.txt
python batch.py named.txt --export wav_output/
# produces: wav_output/morning.wav, wav_output/wash.wav

# Pipe from stdin
cat script.txt | python batch.py

# Built-in demo
python batch.py --demo

---

list_voices.py — browse available voices

Piper has voices in 30+ languages:

python list_voices.py                              # all English voices
python list_voices.py --lang en_GB                 # British English only
python list_voices.py --lang de                    # German
python list_voices.py --lang all                   # every language
python list_voices.py --download en_US-ryan-medium # print download command

Recommended voices:

Voice key	Description
en_GB-alan-medium	British English, male, general purpose
en_GB-alba-medium	British English, female, clear
en_US-lessac-medium	American English, female, neutral
en_US-ryan-medium	American English, male, natural
de_DE-thorsten-medium	German, male, clear
fr_FR-siwis-medium	French, female, natural

Full list and audio samples: https://rhasspy.github.io/piper-samples/

---

Hardware performance

Device	Short phrase	Notes
MacBook Pro M1	< 50ms	Near-instant
ThinkPad T480 (i5)	~80ms	Fast enough
Raspberry Pi 4 (4GB)	~200ms	Good for announcements
Raspberry Pi 3	~500ms	Noticeable delay, still works

Disk: ~50–100MB per voice. RAM: ~200MB during inference.

---

What this connects to

Piper is one piece of a stack that runs entirely on your own hardware:

• Whisper (speech-to-text) + Piper (text-to-speech) = local voice interface
• Add a local LLM (Llama, Mistral, Phi) = voice assistant with no cloud at all
• No subscriptions, no one listening, no vendor lock-in

---

Learn more

• Piper GitHub
• Voice samples
• Rhasspy project

---

A Solystopia tool showcase — AI tools that run on hardware you own.