微软文本转语音

The name is a lie, in the good way: "微软文本转语音" reads like something that would need an Azure account, but Text2AutioEdgeTts doesn't call any API and needs no key. It's a thin wrapper around the edge-tts Python package, which rides Microsoft Edge's online neural TTS endpoint. You type text, you get a high-quality Chinese MP3. That's the entire deal, and it's genuinely great for anyone making Chinese narration.

How it works

You feed it text and pick a voice, and it saves an .mp3 to ComfyUI/output/autio/<prefix>_00001_.mp3, returning the full file path as its output. The heavy lifting happens in edge-tts, which talks to the same neural voices that power Edge's read-aloud - Xiaoxiao, Xiaoyi, Yunjian, and friends - so the quality bar is way above the old robotic TTS you might remember.

The inputs that matter:

• voice - a dropdown of 14 Chinese voices: the standard zh-CN Neural set (XiaoxiaoNeural, XiaoyiNeural, YunjianNeural, YunxiNeural, YunxiaNeural, YunyangNeural), two regional ones (zh-CN-liaoning-XiaobeiNeural, zh-CN-shaanxi-XiaoniNeural), and the zh-HK and zh-TW sets. XiaoxiaoNeural is the default and the one most people stick with - it's the clear best general-purpose female narrator.
• text - multiline string. Your script.
• filename_prefix - defaults to comfyUI, so files land as output/autio/comfyUI_00001_.mp3.

The single output is 音频地址, a STRING with the absolute path to the saved file. The node also registers a UI preview entry, so the MP3 shows up in the workflow's output panel.

Where it fits

This is the cheapest way to get decent Chinese narration into a video pipeline. Pair it with the pack's VideoAddAudio to mux the MP3 onto a clip, or do a re-dub loop: VideoExtractAudio to pull the original track, decide you hate it, generate a new narration, mux it back. No GPU, no model download, no API billing - it's a free ride on Microsoft's infrastructure.

Installing it

Part of the ComfyUI_Lam pack - install via ComfyUI Manager (search "ComfyUI_Lam") or clone:

cd ComfyUI/custom_nodes
git clone https://github.com/yanlang0123/ComfyUI_Lam

Restart, and the pack's install.bat (or your own pip install -r requirements.txt) pulls in edge-tts along with the pack's heavier deps. If you somehow end up with the pack installed but edge-tts missing, pip install edge-tts is the only requirement for this node.

Gotchas

• It's online-only. There's no local model - the audio comes from Microsoft's servers, so no internet, no audio. Don't put this in an offline pipeline.
• It's a free ride, not a contract. Microsoft throttles aggressive use and has changed endpoints before; the edge-tts project tracks that. Fine for personal batches, risky if you script thousands of files a day.
• The voices are Chinese-first. All 14 options are Mandarin, Cantonese, or Taiwanese voices. For English narration you'd want a different tool - this node is deliberately aimed at the Chinese-speaking audience the pack serves.

Troubleshooting

If the node returns a path but no file exists, check your internet connection first - a failed network call can leave an empty output. If the MP3 is truncated, you hit a timeout on a long script; split the text. Otherwise it's one of the most reliable nodes in this pack: no models, no keys, just text in, MP3 out.