SupaVoice Audio to WAV

ComfyUI audio is a mess of dialects. The core AUDIO type is a dict with waveform and sample_rate, but Video Helper Suite hands you a lazy audio map, some packs pass objects with a .samples attribute, and the node downstream just wants the plain dict. SupaVoice Audio to WAV is the pacifier: it takes any of those, normalizes it into a plain ComfyUI AUDIO, and drops a real WAV file on disk while it's at it.

The README is blunt about the use case: "useful as an adapter before nodes that reject Video Helper Suite LazyAudioMap." If you've ever wired VHS output into a transcriber and watched it reject the input type, you know exactly why this node exists.

How it works

Under the hood it runs the pack's audio normalization helper, which understands ComfyUI AUDIO dicts, paths, and objects exposing waveform/sample_rate or samples - so it swallows VHS lazy audio without complaint. You can then force channel_mode to keep, mono, or stereo, optionally resample, and apply gain. The result is written as a 16-bit PCM WAV into the pack's convert folder and reloaded as a fresh AUDIO dict, so whatever comes out is guaranteed vanilla.

The inputs that matter:

• target_sample_rate - 0 means keep the original rate (that's the real default behavior; anything else resamples). Whisper prefers 16kHz, so 16000 is a common pick before a transcriber.
• channel_mode - keep by default. Whisper runs mono internally anyway, but some downstream nodes are picky.
• gain - 0.1 to 4.0, default 1.0. Turn up quiet clips before transcription; it clamps to ±1.0 so you don't get clipping mush.

Outputs: audio (the clean AUDIO dict - wire it into anything that expects core AUDIO), wav_path (the saved file, if you need it on disk), and sample_rate (the INT you can hand to a loader or sampler).

Installing it

cd ComfyUI/custom_nodes
git clone https://github.com/mexxmillion/ComfyUI-SupaVoice-Transcriber

Restart ComfyUI. No model downloads - this node is pure plumbing. Saved WAVs go under E:\SupaVoiceData\comfy_audio_convert unless you set SUPAVOICE_DATA_ROOT (which you should, since that hardcoded Windows default is a weird relative directory on Linux/macOS).

The honest take: it's not glamorous, and the only thing it does that core ComfyUI doesn't is handle the lazy VHS format and write the file in one step. But that's precisely the niche - when a transcription node rejects your audio and you can't figure out why, slapping this adapter in front of it is the fastest fix in the pack.