Audio Spectrum CHOP

From TouchDesigner 099 Wiki


The Audio Spectrum CHOP calculates and displays the frequency spectrum of the input channels.

In the default Visualization Mode the CHOP is set to display the spectrum in a more understandable way by emphasizing the higher frequency levels and the lower frequency ranges.

In another Mode, the Time to Magnitude and Phase mode, the audio can be converted to the frequency spectrum domain, manipulated and then converted back to get a filtered audio signal. When converting a signal to its spectrum, two channels are created from the one containing the audio signal. One channel contains the magnitude of the frequency components, and the other contains the phase. The channels are named, for example chan1_m and chan1_p where _m and _p are the suffixes for the magnitude and phase channels.

Tip: You can reduce cook time if you decrease the FFT Size from its default 8192 samples. The fastest form of this CHOP is by setting the Output Length parameter to "Output Length Manually". For example set the output buffer size to 2048 samples and the FFT Size to 2048. Each time it cooks, the CHOP is looking at the latest 2028 samples (at 44.1 KHz that amounts to the 50 msec, or 3 frames), which is plenty. Note the default form of the CHOP gives you 22,000 samples: 1 Hz to 22,050 Hz in steps of 1 Hz (when set to Frequency vs Logarithmic scaling), designed for clear interpretation: sample 1000 is the level of audio at 1000 Hz.

See Audio Filter CHOP, Audio Para EQ CHOP, Audio Band EQ CHOP, Audio Oscillator CHOP set to White Noise.

PythonIcon.png audiospectrumCHOP_Class


Mode -

  • Visualization - Show the spectrum in a way that more useful, with (by default) high frequencies boosted in level, and lower frequencies boosted in horizontal range.
  • Time to Magnitude and Phase - Calculate the frequency spectrum assuming input is a signal.
  • Magnitude and Phase to Time - Reconstructs a signal assuming input is a frequency spectrum similar to that coming from and Audio Spectum CHOP set to the above option.

FFT Size - Converting to frequency needs a power-of-2 number of samples, like 512, 1024, 2048. (FFT means Fast Fourier Transform.) The more samples, the more accurate the spectrum but the more it doesn't represent the most recent sound. Whatever the size, the CHOP uses the most recent samples. Knowing that audio at 44100 samples per second with a timeline frame rate of 60 frames per second gives 735 samples per frame, if the CHOP cooks 1 frame later and the FFT size is 1024, then it will re-use 1024-735 = 289 samples, which is good as there's a little overlap. However if it cooks 2 frames later, it will miss using 446 frames since it will have advanced 735*2=1470 samples and it will only use 1024 of them.

Frequency <-> Logarithmic Scaling - Logarithmic (=1) is more tangible for human hearing. Each octave is represented with the same number of samples, so low frequencies are more readable. Frequency (=0) shows one sample for a fixed number of Hz, which is what the raw FFT gives, but most of the upper samples are uninteresting. Your ears hear ranges of octaves better. IMPORTANT NOTE: If Mode is set to Visualization and this parameter is set to 0, the output data is interpreted more simply: 1 sample per Hz. Set the CHOP viewer Units to Samples (via RMB on CHOP graph) and the level you see at sample 5000 is the level at 5 KHz.

High Frequency Boost - When 0, the levels are not changed. When greater than 1, the levels are boosted, mostly at the high frequencies. High Frequency Boost can be over-driven past 1.

Output Length - The method how output length will be determined.

Set Output Length - Number of Samples desired in output. The fewer the samples, the less the frequency resolution.


A reasonable scenario (in terms of CPU usage) is an FFT size of 2048. You get good definition with the spectrum, and also enough padding to (almost) deal with two dropped frames.

An Info CHOP can be attached and the channel hz_per_sample can be viewed. Applicable only of the Frequency axis is set to Linear. With it set to output to normal FFT, to determine the frequency that a given sample represents, use the formula:

In order to convert back to a signal, both channels are required. The suffixes should be the same as those used in the previous Audio Spectrum CHOP.