telegram-offline-voice

telegram-offline-voice 🎙️

Local generation, no Token required — Uses Microsoft Edge-TTS to generate high-quality Chinese speech, fully offline processing, no need to apply for any API Key.

Features

🔒 Fully local: No need for OpenAI / Google / Azure or other cloud service Tokens
🎯 Zero cost: Edge-TTS is free to use, no call limits
🗣️ High-quality voices: Uses Microsoft Xiaoxiao (zh-CN-XiaoxiaoNeural) by default
📱 Native Telegram support: Output format conforms to the voice bubble standard

Install dependencies

# Edge-TTS
pip install edge-tts

FFmpeg (Debian/Ubuntu)

apt install ffmpeg


## 使用方法

```bash
# 1. 生成原始音频
edge-tts --voice zh-CN-XiaoxiaoNeural --rate=+5% --text "你好，这是测试" --write-media raw.mp3

# 2. 转换为 Telegram 语音格式
ffmpeg -y -i raw.mp3 -c:a libopus -b:a 48k -ac 1 -ar 48000 -application voip voice.ogg

Technical specifications

Text format cleaning

Text must be cleaned before generating speech to avoid reading out markup symbols:

Needs to be removed	Example
Markdown markup	`加粗`、`代码`、`# 标题`
URL links	`https://example.com`
Special symbols	`---`、`***`、`>>>`

Cleaning example:

# 简单清洗（去除常见 Markdown）
TEXT=$(echo "$RAW_TEXT" | sed 's/\*\*//g; s/`//g; s/^#\+ //g')

Telegram voice bubble format

Telegram requires voice messages to use OGG Opus format, otherwise they will be displayed as files instead of voice bubbles:

ffmpeg -i input.mp3 -c:a libopus -b:a 48k -ac 1 -ar 48000 -application voip output.ogg

Parameter description:

Parameter	Value	Description
`-c:a`	`libopus`	Encoder required by Telegram
`-b:a`	`48k`	Audio bitrate
`-ac`	`1`	Mono (voice standard)
`-ar`	`48000`	Sampling rate 48kHz
`-application`	`voip`	Optimized for speech

Available voices

# 查看所有中文声线
edge-tts --list-voices | grep zh-CN

Recommended:

zh-CN-XiaoxiaoNeural — Female voice, natural and friendly (default)
zh-CN-YunxiNeural — Male voice, steady and professional
zh-CN-XiaoyiNeural — Female voice, lively and young

Speech rate adjustment

# 加速 5%
edge-tts --rate=+5% --text "..." --write-media out.mp3

# 减速 10%
edge-tts --rate=-10% --text "..." --write-media out.mp3

Acknowledgements

Tuned and maintained by @sanwecn .