For a lightweight text-to-speech (TTS) setup on a Raspberry Pi 4 or 5, Piper TTS is the practical default because its neural voices run offline on the CPU. You can generate speech locally, keep text on your device, and avoid a network connection for playback.
Your Pi’s operating system and audio output need a little setup, and the voice model takes disk space and memory while it runs. For a maker project that needs spoken prompts or replies, Piper gives you a useful balance of natural-sounding speech and manageable setup.
Which Lightweight TTS Engine Fits Your Project?
Choose Piper for offline neural speech on a Raspberry Pi; choose eSpeak NG for a very small, fast voice. Newer options such as KittenTTS and XTTS-v2 have different hardware demands, while ElevenLabs provides a hosted option.
| Option | Offline use | Pi fit | Voice cloning |
| Piper TTS | Yes | Strong on Pi 4 or 5 | No built-in cloning |
| eSpeak NG | Yes | Very light | No |
| KittenTTS | Yes | Promising on Pi 5 | Depends on model |
| XTTS-v2 | Yes | Heavy for a Pi | Supported |
| ElevenLabs | Cloud service | Uses an internet connection | Available with supported features |
Why Is Piper TTS a Strong Default?
Piper TTS combines a compact neural TTS model with ONNX Runtime, which runs the model on your Pi’s CPU. You download a voice once, then generate speech offline. That makes Piper useful for announcements, accessibility prompts, and a local voice assistant that needs consistent spoken output.
You choose a voice model to set its language, voice, and quality level. Larger or higher-quality models use more storage and memory, so check the model’s details before adding several voices to a smaller Pi. Piper focuses on ready-made voices; voice cloning is not part of its standard voice selection process.
When Are eSpeak and eSpeak NG Better Choices?
Use eSpeak NG when your project needs brief, clear prompts and a small software footprint. Its synthetic voice sounds more robotic than a neural voice, but it starts quickly and suits status messages such as “Door open” or “Battery low.” The older eSpeak engine is another compact option; eSpeak NG is its maintained successor.
Both engines work well when your priority is low resource use or fast startup. If you need a more natural voice for longer speech, Piper is the stronger starting point. You can install eSpeak NG from your distribution’s package manager with sudo apt install espeak-ng.
Can KittenTTS or XTTS-v2 Run Well on a Pi?
KittenTTS is a compact neural model worth exploring on a Raspberry Pi 5 when you want offline speech and are comfortable checking its setup requirements. Model packaging and software support can change, so confirm that the current release supports your Pi’s operating system and CPU architecture before building your project around it.
XTTS-v2 supports voice cloning and offers more expressive speech, with higher compute and memory demands. That makes it a poor first choice for a small Pi deployment, especially if the same board also runs speech recognition or a language model. ElevenLabs provides hosted speech features, including voice options; your project needs an internet connection to use a cloud service.
What Hardware and Operating System Do You Need?
A Raspberry Pi 4 or 5 running a 64-bit operating system is a suitable base for Piper, with a speaker or audio output device for playback. Give the Pi reliable power, cooling, and enough free storage for the operating system, voice files, and generated audio.
Is a Raspberry Pi 4 or 5 Fast Enough?
Yes. A Raspberry Pi 4 can run Piper’s CPU-based voice models for speech generation, and a Raspberry Pi 5 offers more processing headroom. Your actual response time depends on the voice model, the length of the text, and what else is running on the board.
For a TTS-only project, you can start with the Pi you already have. If you plan to run a voice assistant with local speech recognition or other AI software at the same time, keep those workloads in mind when choosing a model and memory size. A fan or heatsink also helps the board maintain performance during longer periods of use.
Which Raspberry Pi OS Version and Audio Output Should You Use?
Use 64-bit Raspberry Pi OS, including Raspberry Pi OS Bookworm, for a straightforward setup on a Pi 4 or 5. Debian or Ubuntu can also work when you have compatible packages for your board and Python version. Check your architecture with uname -m; aarch64 indicates a 64-bit ARM system.
The Pi 4 has a 3.5 mm audio output, and both Pi models can send audio over HDMI or USB. The Raspberry Pi 5 has no 3.5 mm audio jack, so use HDMI audio, a USB speaker, or a USB audio adapter for analog speakers. Install ALSA’s playback tools with sudo apt install alsa-utils, then check available devices with aplay -l.
Can a Raspberry Pi 3 Handle the Same Workload?
A Raspberry Pi 3 can handle small speech tasks, though it has less processing headroom than a Pi 4 or 5. For short prompts, eSpeak NG is a lighter fit. Piper may work on a compatible 64-bit setup, but model choice and response time become more important on older hardware.
Check that your Pi 3 is running a 64-bit operating system before following the Piper installation steps. If you use the board for other services too, test with one voice model and short phrases first. A Pi 3 is less suited to running speech generation alongside a speech-recognition model.
Install Piper and Choose an Offline Voice
Install Piper inside a Python virtual environment on 64-bit Raspberry Pi OS, then download one compatible voice model. Keeping the software in its own environment helps you manage Python packages without changing system files.
How Do You Install Piper on 64-Bit Raspberry Pi OS?
Update your packages and install Python’s virtual-environment support and ALSA tools:
sudo apt update
sudo apt install python3-venv python3-pip alsa-utils
Create a project folder and activate a virtual environment:
mkdir -p ~/piper-project
cd ~/piper-project
python3 -m venv .venv
source .venv/bin/activate
Install Piper from the active environment:
python -m pip install –upgrade pip
python -m pip install piper-tts
The package installs Piper and its ONNX Runtime dependency. Check that the command is available:
piper –help
If installation reports that no compatible package is available, verify that you are using 64-bit Raspberry Pi OS and a supported Python version. Keep using the same virtual environment for the voice download and your project scripts.
How Do You Download and Test a Piper Voice?
Download a voice into a dedicated folder. The voice name below selects the US English Lessac medium model:
mkdir -p voices
cd voices
python -m piper.download_voices en_US-lessac-medium
cd ..
The download provides an .onnx model file and a matching .onnx.json configuration file. Generate a test WAV file by piping text into Piper:
printf ‘Your Raspberry Pi is ready to speak.’ | \
.venv/bin/piper –model voices/en_US-lessac-medium.onnx \
–output_file test.wav
Play the file with ALSA:
aplay test.wav
If playback fails, run aplay -l to see whether your intended audio device is available. You can also play the WAV file through HDMI or a USB audio adapter by selecting the matching output device in your audio settings.
Why Choose the Lessac Medium Model?
The lessac-medium voice is a useful starting point when you want a ready-to-use US English voice with a balance of file size and speech quality. “Medium” describes the model variant, so it gives you a practical starting point for comparing voice quality and resource use on your Pi.
Piper voices are separate downloads, which lets you keep only the languages and speakers your project needs. Check the voice model’s license before distributing it with a product or sharing it as part of a public project. Keep the .onnx and .onnx.json files together so Piper can load the voice settings.
Connect Speech Output to Your Project
Use Piper to create WAV audio, then send that file to an available playback device with aplay. For a Python project, call Piper as a subprocess so you can pass text and file paths directly.
How Do You Generate and Play a WAV File?
Generate a WAV file whenever you want a saved announcement or a sound file for later playback. With the virtual environment active, run:
printf ‘The front door is open.’ | \
piper –model voices/en_US-lessac-medium.onnx \
–output_file announcement.wav
aplay announcement.wav
For repeated playback, choose clear file names or store temporary output in a dedicated folder. If your program generates speech while another sound is playing, manage playback so the two audio streams do not overlap.
How Can Python Trigger Speech for Home Automation?
You can call Piper from Python after a home automation event, such as a temperature alert or a sensor update. This example generates a WAV file and plays it through the default ALSA device:
import subprocess
text = “The garage door is open.”
model = “voices/en_US-lessac-medium.onnx”
output_file = “/tmp/garage-alert.wav”
subprocess.run(
[
“.venv/bin/piper”,
“–model”, model,
“–output_file”, output_file,
],
input=text,
text=True,
check=True,
)
subprocess.run([“aplay”, output_file], check=True)
Use a queue or background worker if your home automation app handles several events, so speech generation does not block other tasks. For a voice assistant, the same approach lets your response code turn a text reply into spoken output.
How Does TTS Fit Into an Offline Voice Assistant?
Piper handles spoken output; a separate speech-recognition component turns microphone audio into text. You can pair Piper with Whisper or another local speech recognition tool, then route the recognized text through your assistant’s logic before speaking the reply.
Keep each step clear: capture audio, transcribe it, create a response, then synthesize speech. Running Whisper and Piper together increases CPU and memory use, so test the full pipeline on your chosen Pi and voice model. If response time is slow, shorten spoken replies or reduce the number of services running at once.
Keep Your Pi’s Speech Setup Lightweight
Keep Piper’s workload small by using only the voice files and services your project needs. On a Raspberry Pi 4 or 5, a focused setup makes it easier to manage storage, memory, and audio playback.
Remove unused voice models after you settle on a speaker, and save generated WAV files only when your project needs them. Keep the operating system and Piper environment updated, then test the exact voice and audio device your project will use. For a combined voice assistant, measure performance with speech recognition active too, since each model shares the Pi’s CPU and memory.
