SupaVoice Media Audio Trim

Building a voice-clone training set is mostly a cutting problem: you've got an hour of video and you want eight-second clips of clean speech. SupaVoice Media Audio Trim is the pack's answer to that. Point it at an audio or video file path, give it a start and a duration, and it returns a ComfyUI AUDIO plus the trimmed WAV saved on disk - ready for the transcriber or the denoiser.

The nicest part isn't in the README at all: the pack ships a bit of frontend JS that turns this node into a drag-and-drop target with a little timeline widget. Drag a file onto the node and it uploads to ComfyUI's input folder, then you can drag the start/end handles on the widget to set your trim instead of typing timestamps. Small thing, but it's the difference between this and doing it in an audio editor by hand.

How it works

It shells out to ffmpeg with -ss / -t to extract the audio track, forces it to mono or the sample rate you ask for, and writes a WAV into the pack's media-trim folder before reloading it as a plain AUDIO dict. sample_rate defaults to 16000 - right where Whisper likes it.

Inputs: media_path (the file), start_seconds, duration_seconds (default 8), sample_rate, and output_prefix for the saved filename. Outputs: audio and wav_path. There's no end_seconds here - duration is length, not end point - so duration_seconds = 0 isn't a thing to rely on.

Installing it

cd ComfyUI/custom_nodes
git clone https://github.com/mexxmillion/ComfyUI-SupaVoice-Transcriber

Restart ComfyUI. No model downloads. Two gotchas worth knowing: the default media_path is the author's own Windows test file (E:\git\SupaVoice\input\new_test.mp4), so replace it before you queue anything, and because the node calls bare ffmpeg from your PATH, make sure ffmpeg is actually available to ComfyUI's Python process - Windows embedded installs usually are, but a bare Linux install sometimes isn't.

Output files land under E:\SupaVoiceData\comfy_media_trim by default. Set the SUPAVOICE_DATA_ROOT env var to something real for your machine - that hardcoded Windows path becomes a literal weird directory on Linux otherwise. For the full picture, if you also need the frames from that video, the sibling SupaVoice Video Loader does audio plus video in one go.