Audio Transcript Pipe Out (CRT)

This node has exactly one job: take the bundled pipe object that Audio Transcript (CRT) produces and split it back out into the individual pieces you actually want to use - the transcript text, the isolated vocals, a translation, whatever you asked that node to compute. If you've used a "pipe" node in any other ComfyUI pack (bundling model+clip+vae into one wire, say), this is the same idea applied to a transcription job's outputs.

What it's for

Audio Transcript (CRT) can produce up to seven different pieces of data depending on which toggles you enabled - the raw transcript, translated text, isolated vocal and instrumental tracks, an "omnivoice" audio output, plus the original source audio and a status string. Rather than exposing seven output sockets on that node whether you need them or not, CRT-Nodes bundles them into one pipe and makes you attach this node to actually get at them. It's a clean pattern once you know it - one wire out of the transcript node, one wire into this one, and then every downstream node just picks off whichever socket it needs.

The inputs and outputs that matter

One input:

• pipe - required, typed CRT_AUDIO_TRANSCRIPT_PIPE. This only accepts a connection from Audio Transcript (CRT)'s output; it's a pack-specific type, not a generic passthrough.

Seven outputs:

• source_audio - the original AUDIO that went into the transcript node.
• text - the transcript, STRING.
• vocals and instruments - both AUDIO, populated if isolate_voice was enabled upstream; expect these to be empty or meaningless if it wasn't.
• translated_text - STRING, populated if enable_translation was on.
• omnivoice_audio - AUDIO, populated if enable_omnivoice was on.
• status - STRING. Worth wiring to a text-preview node while you're debugging, since it's the most direct way to see what actually happened inside the transcript node without digging through console logs.

Because several of these outputs only get populated conditionally on the upstream node's toggles, don't assume every socket has meaningful data - check status first if something looks empty.

Installing it

This node ships as part of CRT-Nodes' CRT/Audio category alongside Audio Transcript (CRT) - you get both together, since one is useless without the other.

ComfyUI Manager - search CRT-Nodes, install, restart.

Manual:

cd ComfyUI/custom_nodes
git clone https://github.com/PGCRT/CRT-Nodes.git
pip install -r requirements.txt

Restart ComfyUI. This particular node has no dependencies of its own beyond the pack loading - the heavy lifting (Whisper, MelBand RoFormer) happens upstream in Audio Transcript (CRT); see that node's page for the full dependency list, since if that node fails to load, this one has nothing to connect to.

Common issues

• Node won't connect to anything - pipe is a custom type. If you're trying to wire it from something other than Audio Transcript (CRT), it simply won't accept the link; that's by design, not a bug.
• vocals / instruments / translated_text / omnivoice_audio are empty - check the corresponding toggle (isolate_voice, enable_translation, enable_omnivoice) was actually enabled on the upstream Audio Transcript (CRT) node. This node just exposes what was computed; it can't produce data the transcript node never generated.
• Unsure why a run failed or produced nothing - check status first. It's the one output specifically meant to tell you what happened, and it's easy to overlook if you're only wired up to text.
• This node "does nothing" on its own - that's expected; it has no processing logic of its own. All the real work - and all the real troubleshooting - happens at Audio Transcript (CRT).