Audio to Markdown

Here's the thing nobody tells you about "audio to Markdown": converting an audio file to text means transcribing speech, and the way this node does it is by shipping your audio off to Google's speech-recognition web API over the internet. If you assumed it was a local, private, offline tool - it isn't. Audio2Markdown accepts .wav and .mp3 files and returns metadata plus a transcript as Markdown, and the "phoning home" part matters before you point it at anything sensitive.

How it works

Like every converter in ComfyUI-MarkItDown, this is a whitelisted subclass of the generic MarkItDown node, accepting only .wav and .mp3. markitdown then does two things. First, it reads whatever audio metadata it can get its hands on (duration, tags, and similar, if available). Second, it transcribes: the speech_recognition library records the audio and calls recognize_google - that's the Google Web Speech API, an unofficial, free endpoint that does not require a key but does require network access and does receive your audio. For MP3s there's an extra step: the file gets decoded through pydub, which needs ffmpeg available on your system, then transcribed the same way.

The inputs and outputs that matter

One input: source_file, a STRING with the absolute path to your audio file. One output: markdown, a STRING containing the metadata and a ### Audio Transcript: section - or [No speech detected] / an error note if transcription failed. That's the whole node. No language selection, no local-model toggle, no settings.

Why you'd reach for it

Voice memos, meeting recordings, interview clips - turning them into text you can paste into an LLM or search through is genuinely useful, and this is the pack's only audio path. Just know what you're signing up for before you use it as your go-to transcription tool.

Installing it

It's part of the ComfyUI-MarkItDown pack:

cd ComfyUI/custom_nodes
git clone https://github.com/ciga2011/ComfyUI-MarkItDown

Restart ComfyUI (or install via Manager, search "ComfyUI MarkItDown"). The Python side comes from markitdown's bundled dependencies. But the ffmpeg requirement for MP3s is a system dependency - on Windows that usually means a ffmpeg.exe on your PATH, on Linux sudo apt install ffmpeg, on macOS brew install ffmpeg. No ffmpeg, and MP3 conversion fails while WAV files can still work.

Gotchas

Three things worth knowing before you rely on this node. It's not offline and it's not private - your audio goes to Google's servers, so keep that in mind for sensitive recordings. MP3 needs ffmpeg installed at the system level, which is the most common silent failure here. And it's a general speech recognizer, so heavy accents, background noise, or non-English audio will produce rough transcripts. If you want real local transcription with Whisper, this isn't the node - this is the "I need a quick text dump" option.