Piper

Overview

Piper is a fast, local neural text-to-speech system that sounds great while being optimized for the Raspberry Pi 4 and other low-power devices. It uses VITS neural voice models converted to ONNX format for efficient inference, supporting multiple languages and voices without requiring an internet connection. The system is designed to be privacy-focused, with all processing happening locally on your device.

System Requirements

• CPU: Single core for basic use, 2+ cores recommended for better performance
• RAM: 1GB minimum, 2GB+ recommended
• GPU: Optional, can utilize CUDA for faster processing
• Storage: 500MB base + ~50-100MB per voice model
• OS: Linux, Windows, macOS, or any platform supporting Python 3.7+

Installation Guide

Option 1: Python Installation (Recommended)

1. Install with pip:

   pip install piper-tts
   

2. Use the installed command-line tool:

   echo 'Hello, this is a test of text to speech synthesis.' | piper --model en_US-lessac-medium --output_file test.wav
   

Option 2: Binary Release

1. Download a pre-built binary from the GitHub releases page

2. Extract the archive:

   tar -xzf piper_linux_x86_64.tar.gz
   cd piper
   

3. Run Piper:

   echo 'Hello, this is a test of text to speech synthesis.' | ./piper --model en_US-lessac-medium.onnx --output_file test.wav
   

Option 3: Docker Installation

1. Pull the Docker image:

   docker pull rhasspy/piper
   

2. Run Piper in a container:

   echo 'Hello, this is a test.' | docker run -i --rm rhasspy/piper --model en_US-lessac-medium
   

Practical Exercise: Getting Started with Piper

Now that you have Piper installed, let’s walk through a simple exercise to help you get familiar with the basics.

Step 1: Download Voice Models

Piper will automatically download voice models the first time they’re used, but you can explicitly download them first:

# List available voices
piper --list_models

# Download a specific voice
piper --download en_US-lessac-medium

Step 2: Basic Text-to-Speech Conversion

Let’s generate some speech from text input:

1. Create a simple text file named input.txt with some text:

   Welcome to the world of offline text-to-speech.
   Piper makes it easy to convert text to natural-sounding speech
   without sending your data to external services.
   

2. Convert the text to speech:

   cat input.txt | piper --model en_US-lessac-medium --output_file welcome.wav
   

3. Play the audio file using your system’s audio player.

Step 3: Exploring Advanced Features

Once you’re comfortable with the basics, try exploring some of Piper’s more advanced features:

• Streaming Audio: Output audio in real-time with the --output-raw option:

  echo 'This sentence is spoken first. This sentence is synthesized while the first sentence is spoken.' | \
  piper --model en_US-lessac-medium --output-raw | \
  aplay -r 22050 -f S16_LE -t raw -
  

• Multi-Speaker Models: Try models with multiple speakers and switch between them with --speaker <number>:

  echo 'This is speaker 0' | piper --model en_US-multispeaker --speaker 0 --output_file speaker0.wav
  echo 'This is speaker 1' | piper --model en_US-multispeaker --speaker 1 --output_file speaker1.wav
  

• JSON Input: Use structured input with the --json-input flag for more control:

  echo '{"text": "This is JSON formatted input."}' | piper --model en_US-lessac-medium --json-input --output_file json.wav
  

• Integration with Other Systems: Use Piper with Home Assistant, voice assistants, or screenreaders.

Resources

Voice Samples

Listen to samples of the available voices to choose the best one for your project.

Piper Voice Samples

GitHub Repository

Access the source code, report issues, or contribute to development.

GitHub Repository

Training Custom Voices

Learn how to train your own voice models with Piper.

Training Guide