kokoro-tts

Kokoro TTS Local Speech Synthesis

Use the Kokoro TTS engine to generate high-quality speech locally, with zero API cost and fully offline.

Use Cases

• The user says "read this text aloud" or "generate an audio version of this article"
• Need to convert documents, emails, and news into audio
• Preferred over sag (ElevenLabs, requires an API Key and payment)

Prerequisites

pip install kokoro-onnx soundfile

First use requires downloading the model file (about 300MB).

Execution Method

Basic Usage

from kokoro_onnx import Kokoro

kokoro = Kokoro("kokoro-v1.0.onnx", "voices-v1.0.bin")

samples, sample_rate = kokoro.create(
    "你好，这是一段测试语音。",
    voice="af_heart",
    speed=1.0,
    lang="z",  # z=中文, e=英文
)

import soundfile as sf
sf.write("output.wav", samples, sample_rate)

Available Voices

Voice ID	Gender	Language	Style
af_heart	Female	Chinese/English	Warm and natural
af_bella	Female	Chinese/English	Clear and professional
am_adam	Male	Chinese/English	Calm
am_michael	Male	Chinese/English	Energetic

Long Text Processing

Long text is automatically segmented by sentence for synthesis to avoid memory overflow:

import re

sentences = re.split(r'[。！？\.\!\?]', long_text)
all_samples = []
for s in sentences:
    if s.strip():
        samples, sr = kokoro.create(s.strip() + "。", voice="af_heart", lang="z")
        all_samples.append(samples)

import numpy as np
combined = np.concatenate(all_samples)
sf.write("output.wav", combined, sr)

Output Specifications

• The generated audio is saved as a WAV file and the path is returned
• Inform the user of the audio duration and file size
• Mixed Chinese-English text automatically handles language switching