FireRedAudio AudioBatch 试听选择 · T8star-Aix

An AudioBatch in this pack is a list of finished WAVs plus a manifest. It's great for moving a whole session around, but most of the time you want one line - the good take, the one line the client flagged, the sample you're going to loop. T8_FireRedAudio_AudioBatchSelect is the small node that pulls a single item out of the batch and hands it to you as a normal, wireable AUDIO object. Utility node, no magic, and genuinely the glue for half the workflows in this pack.

What it does

You feed it an audio_batch (from BatchDubbing, BatchRetry, EvidenceClips, or any node that outputs T8_FIREREDAUDIO_AUDIO_BATCH) and choose how to find the line:

• position - the 1-based index among successful items (default, starts at 1).
• line_id - the script's stable line ID, e.g. L0037.
• speaker - first match for a character/voice.

It returns the selected clip as native AUDIO, plus a JSON of the item, the selected line_id, and a batch_summary (total lines, playable count, what you picked). The audio output has a native player/download entry in ComfyUI's results area, so you can audition without dragging anything into a preview node.

The selection is deterministic and the inputs are boring on purpose - line_id and speaker are optional strings, so you only fill in what your selection mode needs. That's the whole interface.

Where it slots in

The honest answer: everywhere you'd otherwise improvise a "pick one from a list" node. Three concrete uses:

1. Extract the winning take after TakeReviewBoard adopts a candidate - the adopted batch is a normal AudioBatch, and this node gives you the adopted WAV to save or compare.
2. Pull a single repair out of a repaired batch to spot-check it against the original before you commit the whole batch to delivery.
3. Feed one line into a follow-on node that only takes a single AUDIO - a loudness check, a manual pitch tweak, a compare against the reference.

It also plays nicely with SaveAudioBatch: that node exports the whole batch and ZIPs it, this one lets you audition individual lines first and know what you're exporting. Workflow order matters - select first, then save what survived your ears.

Install

Part of the comfyui-fireredaudio-T8 pack, installed once for all of it:

cd ComfyUI\custom_nodes
git clone https://github.com/T8mars/comfyui-fireredaudio-T8.git
cd comfyui-fireredaudio-T8
python scripts\setup_runtime.py

This node is pure selection - no model, no worker, no model download required. It only touches files already on disk.

Gotchas

The one thing that trips people: position counts successful items, not raw manifest entries. If your batch had 10 lines and 2 failed, position 1 is the first successful WAV, not line one of the script. Use line_id when you care about the script identity - it's immune to that ambiguity. Also, don't use this node to modify a batch; it's read-only. "Replace take 3 with a better one" is a BatchRetry or CandidateApply job, not this. Select is for extraction, and if that's all you need it for, it's exactly right.