Why is My Noise Pink? Noise Contours Explained

Why is My Noise Pink? Noise Contours Explained

In audio, Noise is perhaps the most ubiquitous test signal for making acoustic measurements. For different use-cases and scenarios, there are different types of noise (known as noise contours). Smaart's default contour is pink noise - but what exactly IS pink noise? How can sound have a color? This article will serve as a guide to different contours of noise.

We'll start with white noise.

White Noise

White noise is a signal made of uncorrelated samples. Most white noise generators use uniformly distributed random numbers because they are easy to generate. As such, it is considered Random noise. 

Note: White Noise is not available in Smaart, but can be downloaded or generated elsewhere, if desired for use with Smaart.

White noise contains equal energy across all frequencies, causing its frequency response to appear flat... but only when viewed on a linear scale. This is because it has equal power in each bandwidth, meaning the 10 Hz bandwidth between 20 Hz and 30 Hz contains the same amount of sound power as the 10 Hz bandwidth between 10,000 Hz and 10,010 Hz. 

Since human hearing is logarithmic, white noise sounds much brighter than what one would expect from a "flat" spectrum. As we approach the higher end of the frequency spectrum, each octave spans a greater number of frequencies, so white noise sounds trebly and even harsh when played at high levels. This becomes clear when viewing white noise on an octave or 1/3-octave logarithmic scale, like you'd see on most RTAs. 

To appear flat on a logarithmic plot (sounding flat to our ears), a signal would instead need to instead have equal energy per octave, which is how we get pink noise.

Applications: 

White noise can be used as a reference tone for Transfer Function measurements, but since most measurement plots view sound on a logarithmic scale, it's not ideal. It is often better to use pink noise instead.

White noise can be used to mask sounds such as conversations, traffic, or other background noises that may be distracting or disruptive. This is especially useful in open office environments or other shared spaces where noise levels can be high.

White noise's name is analogous to white light, which contains all frequencies of the visible spectrum at equal levels. Just as light changes its color when its frequency distribution changes, noise can also change "color" by shaping its frequency content. Besides white, the best known "colors" of noise are pink and red (or brown).

Pink Noise

As we explored above, white noise is a signal with equal energy per frequency. Pink noise, on the other hand, has equal energy per octave, corresponding more closely with how we hear. Pink noise decreases at a rate of 3 dB per octave, so the energy of the noise decreases as the frequency increases. This makes pink noise "sound" more balanced and natural to the human ear.

In acoustics, pink noise is used to measure the frequency response of a room or a system. It is typically played through a speaker system while the sound level is measured at various points in the room using a measurement microphone. The resulting data is then used to analyze the frequency response of the space/system and identify any phase issues that may need to be addressed.

Pink noise is also commonly used to measure reverberation time, a measure of how long it takes for sound to decay in a space. Reverberation time is an important factor in the acoustics of a room, as it can affect speech intelligibility and the overall clarity of sound. A measurement of reverberation time can help determine how much sound absorption is needed to achieve optimal intelligibility.

Smaart can generate two basic types of pink noise: "Random" and "Pseudorandom." Random noise is created by streaming the output of a pseudorandom number generator (PRNG) with an extremely long cycle time through a digital filter network. It is randomly seeded every time the generator starts. Given its cycle length (~12.5 hrs at a 48k sample rate), it will effectively never give you the same noise sequence twice.

Pseudorandom noise signals in Smaart repeat in intervals that are powers-of-two samples in length, up to 2^20 (1024K) samples. When using pseudorandom noise, you should always select a cycle length that is at least as long as the longest FFT size that you will be using. For real-time frequency-domain measurements in general, 512K is generally fine. When using sample rates above 48000, however, 1024K is a better choice.

For impulse response measurements, it is common to use pseudorandom noise with a cycle period-matched to the FFT time constant, so that the measurement can be recorded without a data window. The "Drop IR Data Window" option automatically sets the sequence length to match the FFT size for dual-channel measurements in IR mode.

Starting with version 9.5, Smaart has a few different options for noise contours beyond pink.

Red Noise

Red Noise, sometimes referred to as brown noise, exhibits a -6dB/oct decrease in energy - or a power spectrum proportional to 1/frequency^2. A steeper high-end slope than pink noise's -3dB/octave slope, this noise source has a significant amount of low frequency energy and isn't reliable for finding delay times with transfer function measurements.

Pink-Red

Pink-Red noise is a hybridized noise contour created by Rational Acoustics which transitions from pink to red at 1 kHz. 

This way, the high frequency energy transitions to a Red contour higher in frequency, making it more useable than Red noise for full-range systems. while not exciting as much high frequency content, which can be fatiguing to both loudspeakers and our own ears.

Speech-Weighted noise has a spectral shape based on the STI test signal spectrum (without notching) as defined in IEC 60268-16:2020. It is used specifically for taking impulse response measurements, from which STI values will be calculated. As such, its weighting is meant to reflect the frequency response range of the human voice.

SMPTE Noise

While pink noise is commonly used to test sound systems, there haven't been specific standards to define it. In turn, the qualities of pink noise can differ between generation sources, hence the SMPTE-standardized pink noise test signal. SMPTE is designed to be consistent with widely accepted reference noise signals used in movie soundtrack production. 

Developed by SMPTE (the Society of Motion Picture and Television Engineers), a standards-developing organization, this signal type is a standardized form of Pink Noise that's generated from a specific Python script, designed to produce a .wav file of the signal. SMPTE is designed to be consistent with widely accepted reference noise signals that have been used in movie soundtrack production, adhering to the SMPTE ST 2095-1 draft (which can be found here).

Since this is a standardized test signal, there are no user-definable settings for it within Smaart.

SMPTE Noise has the following characteristics:

The AES75 Standard and M-Noise

The AES75: AES standard outlines a procedure for measuring the maximum linear sound levels of a loudspeaker system (or driver) using a test signal called M-Noise. M-Noise, short for "music noise," is a mathematically-derived test signal meant to approximate not only the spectrum response, but the crest factor of a musical program. 

The AES75 standard states that:

"In order to measure maximum linear sound levels meaningfully and repeatably, a signal is required whose RMS and peak levels as functions of frequency are representative of program material."

M-Noise, like other noise-based test signals (pink/white/red/etc.), has a diminishing RMS level with increasing frequency. RMS, short for Root-Mean-Square, is a measure of a signal's average level over a period of time. M-Noise, however, also features a relatively constant Peak level as a function of frequency (Peak level being a signal's highest level at each frequency). Because of this, its crest factor (crest factor = Peak - RMS) increases with frequency. During the development of M-Noise, an analysis on a large variety of music determined this to be an integral defining characteristic for programming material. There are two versions of the M-Noise test signal: one for use at a sample rate of 48k and one for use at a sample rate of 96k. 

These test signals can be downloaded off of Meyer's website on this page.

The procedure specified in the AES75 standard requires the incremental increase of the M-Noise playback level until a stop condition is met: either an unacceptable change in the transfer function's magnitude or an unacceptable change in the coherence of the transfer function. This will determine a loudspeaker's maximum linear sound levels.