JSON Chain Audio Array Pair (Soze)

A ComfyUI AUDIO batch can hold more than one clip, and when you need all of them in a payload - a set of voice samples, a sequence of TTS takes, clips to concatenate on the API side - the single JSON Chain Audio Pair only takes the first frame. JSON Chain Audio Array Pair (Soze) iterates the whole batch: one MP3-encoded entry per frame, bundled into a JSON array under your name.

It's the audio counterpart of the Image Array Pair, and it reuses the same encoding core, so the decisions are familiar: data_uri mode for self-contained payloads, filepath mode when you want tiny strings and don't mind files on disk. ComfyUI's AUDIO shape is [B, C, S] - batch × channels × samples - and this node walks the B dimension in order.

How it works

Each frame is encoded to MP3 (via torchaudio, which needs the ffmpeg backend - same dependency caveat as every node in the audio family) and appended to the array. In data_uri mode you get an array of data:audio/mpeg;base64,... strings. In filepath mode each frame is written out with a guaranteed-unique _NNNNN per-frame suffix so nothing overwrites, and you get an array of absolute paths. An empty batch yields an empty array rather than a crash.

Inputs and outputs

• name - required string widget, the JSON key.
• audio - required AUDIO input (forced). Every frame is encoded.
• mode - data_uri or filepath.
• save_path - filepath mode; each frame gets its own _NNNNN suffix.
• chain_in - optional JSON_PAIRS.

Output: chain_out.

Installing it

Part of the Soze pack:

cd ComfyUI/custom_nodes
git clone https://github.com/SozeInc/ComfyUI_Soze.git
pip install -r ComfyUI_Soze/requirements.txt

Restart ComfyUI, or search ComfyUI_Soze in ComfyUI Manager.

Gotchas

The audio-family failures are all ffmpeg-shaped: if MP3 encoding fails, check torchaudio's backend before suspecting the node. And remember the output is an array, so if your downstream expects a single audio value you've grabbed the wrong variant. The decoder round-trip is supported: JSON Audio Decoder accepts this array format (plus newline lists and single values) and stacks everything back into one AUDIO batch, resampling mismatched sample rates to the first frame.