Text-to-Speech With Piper

Text-to-speech has a genuinely free, local, standard option already, SpeechSynthesis, the output half of the Web Speech API, and it’s worth trying first for anything simple: it needs no server and generally works offline once a browser has its voices loaded. Piper exists for when you need more than that, a specific voice, better prosody, or speech generated server-side rather than only in a visitor’s browser. This lesson gets Piper running locally and calls it from WordPress. It also flags something that actually matters for anyone shipping this in a real product: the original project was archived, and its actively maintained successor changed license.

Step 1: Piper’s real, current status, verified directly

That license change is worth sitting with for a moment before writing any code. MIT let you bundle Piper into a closed-source commercial product with no obligation to share your own source. GPL-3.0 carries real reciprocal terms, if you distribute a product that includes GPL-3.0-licensed code, you generally need to make that product’s source available under a compatible license too. Whether that’s a problem depends entirely on how you ship: a WordPress plugin already released under the GPL, which most are, has no new conflict here. A closed-source SaaS product bundling Piper for distribution is a different situation, and worth a real conversation with whoever handles licensing for that project, not a decision to make silently in a code review.

Step 2: Install the current, maintained package

# Context: on the machine that will run text-to-speech, can be the same host as Lesson 3's whisper-server
python3 -m venv piper-env
source piper-env/bin/activate
pip install piper-tts fastapi uvicorn

Then download a voice model, Piper’s voices are separate .onnx files distributed alongside the project:

python3 -m piper.download_voices en_US-lessac-medium

Step 3: Confirm it works from the command line first

echo "This is a test of local text to speech." | piper \
  --model en_US-lessac-medium \
  --output_file test-output.wav

Play test-output.wav. Hearing an actual synthesized voice, not an error, confirms the model downloaded correctly and the package is runnable before wiring any WordPress code around it.

Step 4: Wrap it as a small local HTTP service

# File: piper-server/main.py
from fastapi import FastAPI
from fastapi.responses import Response
from piper import PiperVoice
import io
import wave

app = FastAPI()
voice = PiperVoice.load( "en_US-lessac-medium.onnx" )

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

# Context: starting the local text-to-speech service on port 5002
uvicorn main:app --host 127.0.0.1 --port 5002

Step 5: Call it from WordPress

// File: my-plugin/class-local-tts.php
function my_plugin_speak_text( string $text ): string|WP_Error {
	$response = wp_remote_post(
		'http://127.0.0.1:5002/speak',
		array(
			'timeout' => 30,
			'headers' => array( 'Content-Type' => 'application/json' ),
			'body'    => wp_json_encode( array( 'text' => $text ) ),
		)
	);

	if ( is_wp_error( $response ) ) {
		return $response;
	}

	if ( 200 !== wp_remote_retrieve_response_code( $response ) ) {
		return new WP_Error( 'tts_failed', 'Local text-to-speech request failed.' );
	}

	$upload_dir = wp_upload_dir();
	$filename   = 'tts-' . wp_generate_password( 8, false ) . '.wav';
	$file_path  = trailingslashit( $upload_dir['path'] ) . $filename;

	file_put_contents( $file_path, wp_remote_retrieve_body( $response ) ); // phpcs:ignore WordPress.WP.AlternativeFunctions.file_system_operations_file_put_contents

	return trailingslashit( $upload_dir['url'] ) . $filename;
}

This writes the returned WAV audio into the uploads directory and hands back a real, playable URL, the shape Lesson 5’s widget expects when it plays a spoken reply back to a visitor.

Test it: generate and play a real reply

wp eval '
$url = my_plugin_speak_text( "This is a test of the local text to speech pipeline." );
echo is_wp_error( $url ) ? $url->get_error_message() : $url;
'

Open the returned URL directly in a browser tab and confirm you actually hear the sentence spoken back, not silence or a corrupted file.

Recap

SpeechSynthesis, the browser’s own built-in text-to-speech, is worth trying first since it needs nothing installed. Piper is the real, local, neural alternative for when a specific voice or server-side generation matters, run as a small local HTTP service and called with the same wp_remote_post() pattern this course has used throughout. The one detail that genuinely changes how you ship this: the original MIT-licensed rhasspy/piper repository is archived, the maintained project now lives at OHF-Voice/piper1-gpl under GPL-3.0, and that license’s reciprocal terms are worth a real decision, not an assumption, before bundling it into a distributed product.

Resources & further reading

• OHF-Voice/piper1-gpl repository, GitHub, the actively maintained successor
• rhasspy/piper repository (archived), GitHub
• SpeechSynthesis interface, MDN Web Docs, developer.mozilla.org