Speech-God Export

ComfyUI has a built-in Save Audio node, so why would a TTS pack ship its own exporter? Because TTS output has one property images don't: a single generation can contain many files. Render eight takes of a scene and Generate hands you one batched AUDIO with eight entries. The stock saver will write that as one weird clip or make you fiddle with batch indexing; Speech-God Export writes each entry as its own numbered file automatically - dialogue_00001.wav, dialogue_00002.wav, and so on. That's the whole reason this node exists, and it's a genuinely nice quality-of-life thing for anyone who's trimmed take files by hand.

Four inputs, all simple:

• audio - the AUDIO to write. Feed it Generate's output directly, usually through Post Process first.
• filename_prefix (default speech-god/dialogue) - the output folder and file stem. It goes under ComfyUI's output directory, so speech-god/dialogue lands in output/speech-god/dialogue_00001.wav. Put a character or scene name in here and you get tidy per-scene folders.
• format - wav, flac, or mp3. WAV is the default and the safe choice; flac if you want lossless at half the size; mp3 for quick sharing.
• sample_rate - 44100, 48000, or 22050. The pack renders internally at 44100, so that's the native output; 48000 is the video-project convention (most NLEs and video tools expect it), and 22050 is there if you're just making little clips or are tight on disk.

It's an output node, and it returns saved_files - a string listing every path it wrote, one per line. Hook that to a text node if you want the filenames captured, or just read it in the node's output area.

How it writes, and the one real gotcha

Under the hood it routes through ComfyUI's standard save-path machinery, resamples each batch entry to your chosen sample rate, and writes each as its own file. There's also a bit of defensive engineering worth knowing: on Windows, recent torchaudio routes saves through torchcodec, which needs FFmpeg DLLs that may be missing - the pack detects that and shims with soundfile automatically, so WAV and FLAC usually just work.

MP3 is the exception. If ffmpeg isn't on your system PATH, mp3 export will fail or fall back to shelling out to ffmpeg and silently do nothing if it's not there. The README's advice is blunt: if MP3 export fails, either install ffmpeg or just use wav/flac. For any serious editing workflow you'd export WAV anyway and encode at the end.

Installing

Same pack as the rest: ComfyUI Manager (search "Speech-God") or git clone https://github.com/realTNEU/ComfyUI-SpeechGod into custom_nodes, then pip install -r requirements.txt into the ComfyUI venv and restart. The exporter itself adds no models and no heavy deps - soundfile (already in requirements) is the only real dependency, and it installs with the pack. If the rest of Speech-God loads, Export just works.