Audio Test Signal Generator

When your audio workflow breaks, you don't want to debug it with a podcast file that's quiet in the wrong places. You want a signal you can predict exactly: a 440 Hz sine, a white noise burst, a silence gap of precisely three seconds. Audio Test Signal Generator (SET_AudioTestSignalGenerator) produces those on demand - no files, no downloads, just deterministic audio for testing, calibrating, or feeding a pipeline while you develop it.

The main input: waveform_type

Ten signal types, and understanding the menu is the whole game:

• Periodic: sine, square, sawtooth, triangle - the classic oscillator shapes, all at frequency Hz.
• Sweep: sweep - a chirp that glides linearly from frequency to frequency_end. Great for testing how a pipeline behaves across a frequency range in one clip.
• Noise: white_noise, pink_noise, brownian_noise - the three colors. White is flat across the spectrum, pink is the "natural" one that sounds like rain, brown is the deep rumble. A reproducible seed input makes noise generation deterministic, which is rare and genuinely useful for comparing runs.
• Special: impulse (one full-amplitude sample - the audio equivalent of a flashbulb, useful for finding latency) and silence (all zeros, which is exactly as useful as it sounds for pacing and padding).

The rest of the panel

• frequency (default 440) / frequency_end (default 880) - Hz, for periodic and sweep types.
• amplitude (0–1, default 0.5) - peak level; ignored for silence.
• dc_offset - a constant added to the whole signal, shifting it up or down. Niche, but it's how you'd deliberately build a DC-biased signal to see if your downstream node handles it.
• phase - degrees (0–360), for aligning periodic signals against each other.
• duration - a string, accepting "5.5" seconds or HH:MM:SS.ss format, same flexible parsing as Audio Cut.
• sample_rate (default 44100), channels (1–8), batch_size (1–64) - shape of the output. Generate a batch of identical test tones and feed the whole stack through your batch-based pipeline.

The output is audio_out: the generated signal in standard ComfyUI AUDIO form.

What it's actually for

Debugging, mostly. Insert a sine before a suspected-broken node and you can hear/see exactly what that node does to a known input. Use silence to inject a precisely-timed gap into a concatenation. Use impulse to measure pipeline latency. And pair it with Audio Musical Note - note-to-frequency converter - to generate actual notes: the pack's generate_and_blend.json example builds a chord from four generated notes blended together.

Install & gotchas

From ComfyUI Manager, search "Audio Batch", or:

cd ComfyUI/custom_nodes/
git clone https://github.com/set-soft/ComfyUI-AudioBatch
pip install seconohe

Restart ComfyUI. No models to download - the whole point is nothing external.

Gotchas are few but real. A square wave at full amplitude is loud; keep amplitude modest (0.5 is a sane default) or you'll clip downstream. frequency above your sample rate's Nyquist limit (half the rate) produces nonsense - the node's max of 22050 assumes 44.1 kHz. And the sweep is linear, so it spends equal time per Hz band, not equal time per octave - fine for testing, just don't expect a log-sweep's frequency response curve.