Comfy.ICU
ExploreDocsPricing
Support

ComfyUI Extension: ComfyUI Signal Processing

Authored by c0ffymachyne

Created

Updated

3 stars

NODES:SignalProcessingPaulStretch, SignalProcessingPadSynth, SignalProcessingFilter, SignalProcessingLoadAudio. This repo contains 4 nodes for ComfyUI that allows for audio filtering

Custom Nodes (0)

    README

    ComfyUI Signal Processing

    THIS IS WORK IN PROGRESS REPOSITORY

    This repo contains signal processing nodes for ComfyUI allowing for audio manipulation.

    Licensing And Attribution

    • LICENSE-GPL-V3 The source code in this repository is licensed under "GNU General Public License Version 3".
    • LICENSE-APACHE-2 Some components are built from parts of code licensed under "Apache License Version 2".
    • LICENSE-CCA-ANY Some components are built from parts of code licensed under "Creative Commons Zero v1.0 Universal"

    Latests Updates

    • Clipping removal - removed all the clipping functions to ensure audio is not clipped throught processing chain
    • Double-precision floating-point - changed internal processing to use double precision floating point.
    • Keeping loudness - all processing nodes use lufs loudness normalization to keep pre-processing gain.
    • GPU support - ported some of the nodes to gpu to gain speed
    • Removed some nodes - if you want them back contact me through this repository. I am planning to add new nodes with better quality

    Mastering Nodes

    Baxandall EQ/Baxandall 3 Band EQ

    The Baxandall EQ is a smooth, wide-band tone control circuit widely used in audio systems, offering gentle boost or cut for bass and treble frequencies. Its simple design and musical response make it ideal for achieving natural tonal adjustments. Implementation is using the standard shelf filter equations from the Audio EQ Cookbook by Robert Bristow-Johnson

    Parameters:

    • audio self explanatory...
    • bass_gain_db Bass gain in decibels
    • mid_gain_db Mid gain in decibels
    • treble_gain_db Treble gain in decibels
    • low_freq : The corner frequency for the low shelf (e.g. ~100 Hz).
    • mid_freq : The center frequency for the mid peaking filter (e.g. ~1 kHz).
    • high_freq : The corner frequency for the high shelf (e.g. ~10 kHz).
    • mid_q : Quality factor for the mid peaking band. Adjusting Q controls the bandwidth of the mid peak. A typical Q might be 0.7 for a broad bell.

    Enhance Harmonics

    Harmonic enhancer boosts selected harmonics to enrich the sound

    Parameters:

    • harmonincs comma separated numbers of harmonics to boost
    • mode whether to automatically detect base frequency based on an audio or use manual setting
    • base_frequency base frequency to use in manual mode
    • gain_db How much to boost harmonics
    • gain_db Width of the filters

    Normalizer

    Normalizer is an amalgamate of multiple normalization approaches, including Loudness Units Full Scale (LUFS) with standard default set to -14db.

    Parametes

    • audio_input: audio input
    • mode: "lufs","rms","peak","auto"
    • target_rms: The desired RMS value for the audio signal. Default is 0.1, which corresponds to a moderate average signal level.
    • target_lufs_db: The desired loudness level in LUFS. Default is -14.0, which is a common loudness target for streaming platforms like Spotify.
    • target_peak: The desired peak amplitude for the audio signal. Default is 0.9, meaning the loudest sample will be scaled to 90% of the maximum possible amplitude.
    • target_auto: The desired amplitude level for the audio signal. The algorithm scales the audio to match this level. Default is 0.7 (normalized scale from 0 to 1).
    • target_auto_alpha: The smoothing factor for the gain adjustment. A smaller value of alpha makes the gain adjustment slower and smoother (avoiding sudden jumps). A larger value makes the gain adjustment faster but potentially introduces abrupt changes.

    Loudness

    The get_loudness function calculates the integrated loudness of an audio signal in LUFS (Loudness Units relative to Full Scale). This is a perceptual measure of loudness, taking into account the human ear's sensitivity to different frequencies and the entire audio signal's duration.

    SignalProcessingStereoWidening:

    Open Source Stere Widening Plugin. The implementation is a direct copy of parts of the source code corresponding to this paper developed by Orchisama Das. The code is distributed under **CC0 1.0 Universal** license. Original Source Code

    Parameters:

    • audio: input audio
    • mode: "decorrelation" and "simple" - "decorrelation" is based on "Open Source Stere Widening Plugin" as described above
    • gain: post width gain
    • width: width of the stereo effect

    Effects Nodes

    Convolution Reverb

    Convolution reverb simulates realistic acoustic spaces by applying the impulse response of a physical environment to an audio signal. It captures the natural reverberation characteristics, providing authentic spatial depth and ambience.

    How do I use it ?

    I recommend downloading impulse response files from this location Voxengo-IR and Greg Hopkins EMT 140 Plate Reverb Impulse Response. They sound absolutely fantastic and have great licensing. In order for the files to show up for selection in the convolution reverb please download the files and organize them like this :

    • comfyui_signalprocessing/audio/ir/Voxengo/ <- copy wave files into this directory
    • comfyui_signalprocessing/audio/ir/EMT-140-Plate/ <- copy wav files into this directory

    Parameters:

    • impulse_response: impulse response file selected
    • audio_input: audio to apply reverb to
    • wet_dry: mix amount of the effect

    SignalProcessingPaulStretch

    PaulStretch excels at extreme time-stretching with high-quality results, preserving the pitch and tonal characteristics of the original audio. This node contains a port of algorithm developed by Nasca Octavian Paul.
    Original Source Code

    Parameters:

    • audio
      The input audio signal to be stretched

    • stretch_factor
      Determines the amount of stretching applied to the audio.
      Range: 0 (no stretch) to 100 (maximum stretch).
      Example: A stretch_factor = 10 stretches the audio to 10 times its original length.

    • window_size_seconds
      Specifies the window length for the stretching algorithm, in seconds. Larger values produce smoother and more ambient results by averaging the time-domain samples over a longer period.
      Example: window_size_seconds = 1.0 provides smooth stretching for most applications, while smaller values retain more transient detail.

    SignalProcessingPadSynth :

    This node is a synthesiser "PadSynth" based on a PADSynth algorithm This node contains a port of algorithm developed by Nasca Octavian Paul Original Source Code

    Parameters:

    • samplerate: samplerate
    • fundamental_freq: fundamental frequency for the sounds generation
    • bandwidth_cents: bandwidth centers
    • number_harmonics: number of harmonics
    • amplitude_per_harmonic: amplitude per harmonic as a json, takes as list of amplitudes [0,1,4,...], it's count must match number of harmonics
    • audios: audio output one channel per note - use "SignalProcessingMixdown" node after to get single audio with monot channel copies to L and R

    SignalProcessingPadSynthChoir

    This node is a synthesiser "PadSynth" emulating choirs Original Source Code

    Parameters:

    • samplerate: samplerate
    • base_freq: base frequency
    • step_size: step size
    • num_notes: number of notes
    • bandwidth_cents: bandwidth cents
    • number_harmonics: number of harmonics to produce

    SignalProcessingFilter :

    Classic filters

    Parameters:

    • audio: input audio
    • cutoff: filter cutoff
    • filter_type: filter type - "lowpass", "highpass", "bandpass", "bandstop"
    • q_factor: width of the filter

    SignalProcessingMixdown

    mixdown outputs from PadSynths with volume control per note

    Parameters:

    • audios: audios input
    • audio: audio output

    Testing/Visualization Nodes

    This section contains nodes enabling basic analysis and development of other nodes

    SignalProcessingSpectrogram

    Renders Mel Spectrum Into An Image

    Parameters:

    • audio: audio input
    • image: image output

    SignalProcessingWaveform

    Renders Wave Shape Into An Image

    Parameters:

    • audio: audio input
    • image: image output

    SignalProcessingLoadAudio :

    This node lets you stretch audio to about 100x it's original speed whil mainting pitch. it's great for making pad sounds:

    Parameters:

    • audio_file: input audio file
    • gain: when to start audio from