[sdiy] Measuring Roughness, Sharpness, Tonality, Fluctuation Strength, Stationary Loudness, and Time-Varying Loudness, was: VCF caps in modern synths

Sun Mar 28 17:11:15 CEST 2021

BTW, if anyone wants to know how to measure Roughness. It's a fairly
new concept in psychoacoustics which seems to have first been
published about in 1985 in German and in 1997 in English (see
references in [1] which cite [2] and [3]). A more detailed description
of an algorithm is described in [4]. Roughness is part of a bunch of
other novel psychoacoustical measurements which seem in specific to be
great tools for anyone working with sound or music, and would probably
be very popular, if every last deadbeat wasn't trying to monetize
knowledge. There's a bunch of them described in documentation for the
National Instruments Sound and Vibration Measurement Suite, a LabVIEW
suite which implements those measurements [5] which I copy in below.
Finally, Mattias Rickardsson brings up a nice overview of roughness
measurement in [6].

[1] Martin Pflueger, Robert Höldrich, Franz Brandl, Werner Biermayer:
Psychoacoustic measurement of roughness of vehicle interior noise,
1999
https://www.researchgate.net/publication/243523940_Psychoacoustic_measurement_of_roughness_of_vehicle_interior_noise
[2] W. Aures: Ein Berechnungsverfahren der Rauhigkeit, Acustica Vol.
58, 1985, pp268-281, Paywalled here for $30:
https://www.ingentaconnect.com/contentone/dav/aaua/1985/00000058/00000005/art00005
[3] P. Daniel, R. Weber: Psychoacoustical Roughness: Implementation of
an optimized Model. - Acoustica Vol. 83, 1997, pp 113-123, Paywalled
here for $30: https://www.ingentaconnect.com/contentone/dav/aaua/1997/00000083/00000001/art00020
[4] R. Hoeldrich: An Optimized Model for the  Objective Assessment of
Roughness Sensations in Vehicle Noise, free at:
https://iem.kug.ac.at/fileadmin/media/iem/altdaten/projekte/publications/paper/pub_avlpaper/pub_avlpaper.pdf
[5] National Instruments: Sound Quality Algorithms (2020)
https://www.ni.com/de-at/support/documentation/supplemental/08/sound-quality-algorithms.html
[6] Sound Modulation Metrics: Fluctuation Strength and Roughness
(2020) https://community.sw.siemens.com/s/article/sound-modulation-metrics-fluctuation-strength-and-roughness

8<-----------------------------------------------------------

Sound Quality Algorithms
Updated Feb 4, 2020

Overview

Sound quality, a set of algorithms developed from the science of
psychoacoustics, is used to define the relationship between the
physical quantities of sound and the subjective impression as heard by
the human ear. These algorithms examine physical parameters such as
sound pressure level, frequency, and modulation depth and correlate
them to human hearing perception. Sound quality algorithms have
revolutionized acoustical measurements in the automotive, white goods,
and other industries by transitioning the focus from simple noise
emission reduction to a more sophisticated process of increasing sound
desirability. Sound quality consists of the following algorithms: ISO
532B stationary loudness, time-varying loudness, Aures roughness,
Aures sharpness, Aures tonality, and fluctuation strength. The NI
Sound and Vibration Measurement Suite provides LabVIEW VIs with which
you can implement these algorithms.

Contents

* Introduction
* Stationary Loudness
* Time-Varying Loudness
* Roughness
* Sharpness
* Tonality
* Fluctuation Strength
* Conclusion

Introduction

For decades, the automotive and aerodynamics industries have used
acoustical measurements to understand how sounds from turbines,
motors, and other physical actions affect humans. Until recently, the
analysis performed on those measurements has been quite simple,
consisting of sound pressure level analysis, octave analysis, FFT
analysis, and the application of basic weighting filters. These
algorithms, while good at revealing the decibel level or frequency
content of a signal, do not uncover a number of important phenomena
that determine the desirability of the signal. To move beyond simple
noise level analysis and perform practical environmental noise
measurements, sound quality algorithms have been developed to explain
how sounds are perceived by the human ear.

Sound quality algorithms are the product of research across a spectrum
of sciences including acoustics, physics, communication engineering,
mechanical engineering, musicology, marketing, physiology, and
psychology. These algorithms combine the psychoacoustical, physical,
and cognitive aspects of sound in order to provide new performance
metrics to design engineers. An example application of these
algorithms in the automotive NVH industry is to design an engine with
a more pleasing sound or a door handle with a more soothing click. The
sound quality algorithms also are applicable in the production of
consumer electronics. These algorithms allow an engineer to design a
better-sounding product, which psychologically increases the chances
of consumer adoption.

Sound quality consists of the following algorithms: ISO 532B
stationary loudness, time-varying loudness, Aures roughness, Aures
sharpness, Aures tonality, and fluctuation strength.

Stationary Loudness

Loudness is a term referring to the human perception of sound volume.
The definition of loudness states that 1 sone, the unit of loudness,
corresponds to a 1 kHz tone at 40dB. The loudness scale quantifies the
loudness linearly to the human ear in which a doubling of the
sone-value maps directly to a doubling in loudness.

Stationary loudness is an algorithm for steady-state noises, or
signals that do not vary with time. This algorithm measures the 1/3
octave spectrum, combines the fractional-octave bands into critical
bands, and then applies spectral masking. This algorithm returns the
result as the specific loudness versus critical band rate and then
integrates the specific loudness to measure the total loudness and
loudness level. This algorithm also is known as ISO 532B loudness or
Zwicker loudness and is in compliance with ISO 532B, DIN 45631, and
ISO/R 131.

Time-Varying Loudness

Time-varying loudness is an algorithm for calculating the loudness of
non-steady-state noises, or signals that vary with time. This
algorithm measures the 1/3 octave spectrum using exponential averaging
with a 2 ms time constant, combines the fractional-octave bands into
critical bands, and applies temporal and spectral masking. This
algorithm then returns the result as the specific loudness versus
critical band rate, integrates the specific loudness, and applies
temporal post-masking filters to measure the time-varying loudness.
This algorithm calculates the time-varying loudness in compliance with
DIN 45631/A.

Roughness

Roughness is another algorithm used to determine the subjective
judgment of sound quality. Roughness correlates to how noticeable or
annoying a sound is as heard by the human ear. More specifically,
roughness is a hearing sensation related to loudness modulations at
frequencies too high to discern separately, such as modulation
frequencies greater than 30 Hz.

The roughness algorithm measures the energy in 24 barks, computes and
filters the envelope of the signal in each band, measures the
amplitude modulation of each envelope, and then weights the level in
each band using both the modulation index of that band and a
frequency-dependent weighting function. The algorithm returns the
result as the roughness spectrum versus critical band rate and then
integrates the roughness spectrum to measure the roughness.

Sharpness

Sharpness is a hearing sensation related to frequency and independent
of loudness. Sharpness corresponds to the sensation of a sharp,
painful, high-frequency sound and is the comparison of the amount of
high frequency energy to the total energy. The sharpness algorithm
computes sharpness from the sound pressure signal waveform, the
1/3-octave band spectrum calculated over the frequency range 25 Hz to
12.5 kHz, or the specific loudness.

This algorithm normalizes the specific loudness spectrum by the total
loudness and weights the spectrum according to frequency. The
algorithm returns the frequency-weighted result as the specific
sharpness versus critical band rate and then integrates the specific
sharpness to measure the sharpness. Higher frequency components in the
signal generally result in higher sharpness measurements.

Tonality

Tonality is used to determine whether a sound consists mainly of tonal
components or broadband noise. The algorithm measures the relative
strength of the tones in a signal compared to the overall signal. For
each time block, this algorithm first varies the frequency resolution
according to the frequency selectivity of human hearing, searches the
spectrum for likely tones, and then compares the loudness of the tones
to the loudness of the sound.

Fluctuation Strength

Fluctuation strength is a hearing sensation related to loudness
modulations at low frequencies that are discernable individually.
Fluctuation strength uses a similar method to "roughness versus time"
analysis except that it focuses specifically on signal variations with
very low modulation frequencies.

Fluctuation strength measures the energy in 47 overlapping barks,
computes and filters the envelope of the signal in each band, measures
the amplitude modulation of each envelope, and weights the level in
each band using a frequency-dependent weighting function. The
algorithm returns the result as the fluctuation strength spectrum
versus critical band rate and then integrates the fluctuation strength
spectrum to measure the fluctuation strength. The algorithm examines
modulations between 0 to 30 Hz, with special emphasis on those near 4
Hz.

Conclusion

The Sound and Vibration Measurement Suite provides VIs you can use to
implement sound quality algorithms. The Sound and Vibration
Measurement Suite also provides in-depth documentation about each of
these algorithms. You can download the Sound and Vibration Measurement
Suite for free as a 7-day trial. You also must install LabVIEW in
order to use the sound quality algorithms in the Sound and Vibration
Measurement Suite.

Download and evaluate the Sound and Vibration Measurement Suite

On Sun, Mar 28, 2021 at 3:52 PM cheater cheater
<cheater00social at gmail.com> wrote:
>
> > On Mon, Jan 18, 2021 at 11:49 AM ColinMuirDorward <colindorward at gmail.com> wrote:
> > In my 4pole 3320 BPF, the polypropylenes sound slightly better. The obvious difference is that at lower frequency settings, they go into self-oscillation more readily than the ceramics. To my biased ear, the ceramics make some small distortion perhaps, or for some reason add a slight edge to the sound. I did do a blind test using a 2pole 2164 SVF and the difference there is more noticeable than in any other filter I've tested. Here, the polys are a clear win. I want to use words like "liquid, juicy, lush" for the polys vs "bite, edge, grit" for the ceramics. However those are too strong words to describe a very subtle difference. I am waiting to find the right modern mono synth that I can do a capacitor "upgrade" on and test if it's possible to perceive a change in the sound in a factory-made unit. I have an uno synth on the way (was on sale for $135cad?!), perhaps I'll be lucky enough to find space inside that tiny case for four WIMAs.
>
> The sound quality you're talking is "roughness" and is measured in Aspers.
>
> Tiny capacitors have a very large voltage gradient across a very short
> casing. They're probably self-modulating, i.e higher voltage means
> less capacitance. Essentially self-filter-FM.
>
> Look at this article:
> https://passive-components.eu/dc-and-ac-bias-dependence-of-mlcc-capacitors-including-temperature-dependence/
>
> Look at Figure 4. They took ten samples of "1uF 0603 16V X5R parts
> from Vendor-F" and invariably they lose 40% capacitance at +15V /
> -15V. So unless you're running your synth at low voltages, you're
> running into exactly that. I believe there's a filter out there that
> leverages this, but I forget what it was. There's some filter for sure
> where the audio was low voltage AC (maybe 100mVpp or less) and the
> cutoff was high voltage DC (0-x volts). Hell if I know what it was...
> some EQ maybe? IDK
>
> Figure 5 shows capacitors range from -40% to even -80% capacitance at
> +/-15V depending on make and model.
>
> C0G as I understand specifies capacitance retention across temperature
> ranges. I don't know that they're specified across voltage ranges. A
> quick look at the first C0G datasheet I could find (from AVX) shows
> that they don't even specify voltage dependence.
>
> So yeah, in this case, physically larger components are just better if
> you want to stay at +/-15V. Or alternatively go to something like mV
> range and have a much more linear, tiny, capacitor - but at the cost
> of EMI adding a bunch of noise. Your choice. Fully-balanced synths
> anyone? I mean you could probably fit two copies of the same circuit
> in 0201 in the footprint that you'd otherwise be using for silver mica
> or red brick wimas.
>
> One more option is to add self-filter-fm cancellation to your filter,
> and have it as part of the tuning or calibration procedure. Probably
> for the best anyways.
>
> Well, that should explain the issue hopefully. Cheers