pyopenjtalk

0.4.1 verified Mon Apr 27 auth: no python

A Python wrapper for OpenJTalk, a Japanese text-to-speech engine. Current version 0.4.1, requires Python >=3.8 and cmake. Release cadence is irregular; last release was 2025-03-01.

pip install pyopenjtalk

Common errors

error ModuleNotFoundError: No module named 'pyopenjtalk' ↓

cause Package not installed or installed incorrectly.

fix

Run 'pip install pyopenjtalk'. If on Windows, may need to install Visual C++ Build Tools.

error ImportError: libopenjtalk.so: cannot open shared object file ↓

cause Missing system dependencies or broken installation.

fix

Reinstall: 'pip uninstall pyopenjtalk && pip install pyopenjtalk'. On Linux, ensure build essentials are installed.

error ValueError: numpy.ndarray size changed, may indicate binary incompatibility ↓

cause numpy version mismatch between build and runtime.

fix

Upgrade numpy: 'pip install --upgrade numpy' to at least 1.20.

Warnings

breaking In v0.3.0, run_frontend behavior changed due to introduction of 'marine' accent model. Output format may differ from v0.2.x. ↓

fix Check the full documentation for new feature structure; verify your parsing code against output.

gotcha Building from source requires cmake. If pip fails to install, install cmake first (e.g., 'apt install cmake' or 'brew install cmake'). ↓

fix Ensure cmake is installed and in PATH before 'pip install pyopenjtalk'.

deprecated Legacy API functions like 'pyopenjtalk.run_tts' are removed as of v0.3.4. Use 'pyopenjtalk.synthesize' instead. ↓

fix Replace 'run_tts' with 'synthesize'.

Imports

run_frontend

wrong

from pyopenjtalk.frontend import run_frontend

correct

from pyopenjtalk import run_frontend

run_frontend is a top-level function, not part of a submodule.

synthesize

wrong

import pyopenjtalk.synthesize

correct

from pyopenjtalk import synthesize

synthesize is a function, not a module.

Quickstart

Basic usage: synthesize speech or extract linguistic features from Japanese text.

import pyopenjtalk

# Synthesize speech from text (returns wav as numpy array)
text = "こんにちは"
wav, sr = pyopenjtalk.synthesize(text)

# Or get linguistic features
features = pyopenjtalk.run_frontend(text)
print(features)