persian-speech 1.1.2

Natural Persian Text-to-Speech. One command.

Persian Speech 🎙️

Natural Persian Text-to-Speech. Zero config, one command.

Built on Mana-Persian-Piper, trained on 114 hours of high-quality Persian speech.
Runs on servers, laptops, Raspberry Pi, and mobile devices. Entirely offline.

Why Persian Speech?

• Natural voice — Trained on the largest Persian speech dataset (Mana-TTS)
• Runs anywhere — Linux, Windows, macOS, Raspberry Pi, Android, iOS
• Edge & mobile ready — ONNX runtime, lightweight model, no GPU needed
• Offline first — No internet after initial model download
• Blazing fast — 25x faster than realtime on CPU
• One command — speech "سلام" and you're done

Install

pip install persian-speech

Model downloads automatically on first use. That's it.

Usage

Command line:

speech "سلام به همگی"
speech "سلام دنیا" hello.wav

Python:

from persian_speech_runtime import talk
talk("سلام به همگی", "output.wav")

Server:

python server.py
# http://localhost:8000/tts?text=سلام

import sys

sys.path.insert(0, "src")
from persian_speech_runtime import PiperSpeechEngine
from http.server import HTTPServer, BaseHTTPRequestHandler
from urllib.parse import urlparse, parse_qs

tts = PiperSpeechEngine()


class Handler(BaseHTTPRequestHandler):
    def do_GET(self):
        q = parse_qs(urlparse(self.path).query)
        text = q.get("text", ["سلام به گل های توی خونه"])[0]
        self.send_response(200)
        self.send_header("Content-Type", "audio/wav")
        self.end_headers()
        self.wfile.write(tts.stream(text))


print("http://localhost:8000")
HTTPServer(("0.0.0.0", 8000), Handler).serve_forever()

API

from persian_speech_runtime import PiperSpeechEngine, talk

talk("سلام", "out.wav")              # One-liner

tts = PiperSpeechEngine()
tts.save("سلام", "out.wav")          # Save WAV
tts.stream("سلام")                   # WAV bytes
tts.synthesize("سلام")               # NumPy array

License

MIT — Use it anywhere. Built on Mana-Persian-Piper (MIT) and Mana-TTS (CC0).