Merge Voice Audios

Merge Voice Audios does for reference clips what Merge Voice Emotions does for emotion objects: it bundles up to ten AUDIO inputs (audio1 through audio10, first one required) into a single audios list you can feed to the Generate nodes. It's the quick route to multi-voice synthesis when you don't care about emotion control - just a stack of voices for a scripted dialogue.

The workflow is simple: load one reference clip per character (a LoadAudio node per voice), merge them here, and run the list into the reference_audios input on IndexTTS Generate or Generate Simple. Then script your conversation in the text box using [s1], [s2], ... markers - same auto-naming convention as the emotion side, with voices assigned by position in the merge order. Position one is s1, position two is s2, and so on.

When it beats the emotion path

When you want the raw model default behavior - each line spoken in the cloned voice's natural delivery, no emotional conditioning layered on. The emotion path (per-character emotion nodes + Merge Voice Emotions) gives you a cast with moods; this gives you a cast with voices only. It's fewer nodes, less to tune, and it's the right call for neutral narration or when the reference clips are already expressive enough on their own.

The gotcha

Reference quality rules everything here. IndexTTS-2 clones from whatever you feed it, and loud, dynamic clips clone dramatically better than quiet, flat ones - a lesson the community relearns constantly. Also remember that in the Generate node's priority order, emotions beats reference_audios beats reference_audio - so if you wire both an emotions list and this audios list into the same Generate node, the emotions win and your merged audios get ignored. Keep them as alternatives, not companions.

One input, one output, no settings. It's a wire with a zip-tie, but it's the piece that turns single-voice TTS into a scripted scene.