Audio Classification

Audio Classification is the node that tells you what a sound is. Feed it an audio file and it returns ranked labels - "speech," "guitar," "dog bark," "engine" - with confidence scores. In a ComfyUI graph it's the audio cousin of image classification: a quick way to tag audio clips, sort a folder by content, or sanity-check what's actually in a sound file before you build a bigger pipeline around it.

It wraps the audio-classification pipeline from Hugging Face, defaulting to MIT/ast-finetuned-audioset-10-10-0.4593 - an Audio Spectrogram Transformer fine-tuned on AudioSet, Google's massive audio event dataset. AST chops the audio into a spectrogram and runs it through a vision-style transformer, which is why it can recognize hundreds of everyday sound categories. The node returns the top labels as JSON with scores.

The inputs and output

• audio_path - a string path to an audio file on disk. The recurring pattern in this pack's audio nodes: it's a file path, not an audio signal, so make sure something upstream saved your audio to a real file.
• model_name - the Hub model ID, typed as a string.
• top_k - how many labels to return, default 5, range 1–20.

The single output, results_json, is a STRING with the ranked labels and scores - [{"label": "Speech", "score": 0.83}, ...] and so on.

How to install it

Standard pack install - ComfyUI Manager (search "ComfyUI-Transformers") or:

cd ComfyUI/custom_nodes
git clone https://github.com/kadirnar/ComfyUI-Transformers

Restart ComfyUI. The pack's requirements.txt includes soundfile and scipy for audio I/O, and the AST checkpoint is a few hundred MB on first download.

Where people get burned

The path-string input is the usual trap - beginners try to wire an AUDIO output from another node directly in and hit a type mismatch. You need a file path. And note the default model's scope: AudioSet categories are everyday sounds, so a clip that's purely speech will often come back labeled "Speech" with music or other secondary labels - it's not a music-genre or speaker-ID classifier. If your goal is transcribing the words, the ASR node in this pack is the one you want; if you want to know what kind of sound it is, this is it.