voice-to-me

Voice Reply

Organize the answer into a concise and natural spoken script, generate an MP3 in the current session workspace, and return it via QQ.

Dependencies

python3 -m pip install "edge-tts>=7.0,<8"

The system must also have ffmpeginstalled, used for default noise reduction and appending trailing silence.

Generate Audio

1. Keep the read-aloud text natural, and remove Markdown syntax, URLs that should not be read aloud, file tags, and model or session metadata.
2. The output must be located in the current working directory so that the QQ gateway can receive it. The default path is .qq-voice/reply-<timestamp>.mp3
3. Get the directory path of this Skill from the list of available Skills, then run:

python3 "<skill-dir>/scripts/voice_to_me.py" \
  --text "要朗读的内容" \
  --output "$PWD/.qq-voice/reply-$(date +%Y%m%d-%H%M%S).mp3"

Default persona configuration:xiaoyi（zh-CN-XiaoyiNeural）、+7Hz pitch and --rate 1.13（+13% speaking rate). Only use --voice xiaoxiao

Parameter	Default Value	Description
--voice	xiaoyi	Voice; use when the user explicitly prefers a mature voice xiaoxiao
--rate	1.13	Speaking rate, range 0.5–2.0
--tail-silence	0.5	Seconds of silence at the end of the voice, range 0–5
--denoise	On	Light high-pass, de-essing, and low-pass to suppress low-frequency rumble and high-frequency hiss.
--no-denoise	Turn off noise reduction	Use only when the user requests preserving the original voice.

For longer text or text containing Shell special characters, save the read-aloud content to the session workspace, then use --text-file <path> instead of --text

By default, light noise reduction is enabled, and trailing silence is appended via MP3 stream copy to avoid re-encoding for pauses. Noise reduction consumes a small amount of extra CPU and re-encodes the audio; it may not necessarily eliminate the high-frequency texture of the synthesized voice itself, so use --no-denoise

Return via QQ

After successful generation, add the following tag on a separate new line in the final reply:

<qqvoice>/absolute/path/to/reply.mp3</qqvoice>

• Use the absolute path printed by the script. Do not place this tag in a Markdown code block.
• When replying with voice, no other text reply is needed.