Model for Predicting Loudness and their Application to the Fitting of Hearing Aids

 

Brian C. J. Moore - bcjm@cam.ac.uk

University of Cambridge

Department of Experimental Psychology,

Downing Street,

Cambridge CB2 3EB, UK

 

Popular version of paper 2pPP3

Presented Tuesday afternoon, April 20th, 2010

159th ASA meeting, Baltimore, MD

 

 

Loudness is a subjective attribute of sounds, corresponding to their perceived magnitude. It is strongly influenced by the intensity of sounds; the more intense a sound is, the greater its perceived loudness. The physical intensity of sounds is often specified in terms of units called decibels Sound Pressure Level: dB SPL. A sound with a level of 10 dB SPL would be audible but very soft for a person with normal hearing, a sound with a level of 60-70 dB SPL would usually be judged as comfortably loud, and a sound with a level of 90-100 dB would be judged as very loud. Sounds become painfully loud (and potentially very damaging to the ear) at levels of 110-120 dB SPL.

 

Loudness is also influenced by factors other than sound level in dB SPL. For sinewaves or pure tones (such as the sound produced by a tuning fork) with a fixed sound level in dB SPL, the loudness is greater for medium frequencies (500 to 5000 Hz) than for very low or very high frequencies. Loudness is also influenced by the complexity of a sound: the range of frequencies contained within a complex sound like the sound of a car or air-conditioner. For a fixed sound level, a sound is usually louder when it contains a wide range of frequencies than when it contains a narrow range of frequencies.

 

Loudness is measured in units called sones. One sone is defined as the loudness of a 1000-Hz sinewave with a sound level of 40 dB SPL when the sinewave is presented from a frontal direction and is heard with both ears. A sound with a loudness of two sones is judged to be twice as loud as a sound with a loudness of 1 sone, while a sound with a loudness of 0.25 sones is judged to be one-quarter as loud as a sound with a loudness of 1 sone.

 

It is long been recognized that it would be useful to be able to predict the loudness of sounds, as perceived by a typical human listener, from the physical characteristics of sounds, such as their intensity and frequency content. To do this, researchers have developed models of some of the known processes that occur in the auditory system, and have used these models to generate predictions of loudness. Pioneering work in this area was conducted by Harvey Fletcher, working at the Bell Telephone Laboratories, S. Smithy Stevens, working at Harvard, and Eberhard Zwicker, working first at Stuttgart and then at the Technical University in Munich, Germany. Later, in collaboration with Brian Glasberg and Tom Baer, I refined and modified some of the concepts used in these models so as to derive more accurate predictions of loudness. The model that we developed at Cambridge formed the basis for the latest ANSI standard for calculating loudness and also forms the basis for part of the forthcoming ISO standard. The model has been successfully applied to prediction of the loudness of noises produced by airplanes, cars, air-conditioning and ventilation systems, musical instruments and mobile telephones. The model can also be used to predict the threshold for detecting a sound, i.e. the lowest level at which the sound is just audible.

 

The perception of loudness can be strongly altered by hearing loss. In countries like the UK and the USA, hearing loss affects more than 10% of the adult population, and it is especially prevalent among the elderly. The commonest cause of hearing loss is a problem with the functioning of the inner ear, specifically the part of the inner ear which deals with the analysis of sound, called the cochlea. At Cambridge, we have developed a modified version of the loudness model to predict loudness as perceived by human listeners with cochlear hearing loss. The model can make predictions of loudness for a specific ear of a specific individual, based on the audiogram, which is the amount of hearing loss for that ear, measured as a function of frequency using sinewave signals.

 

Among other things, the model predicts an effect called loudness recruitment that is commonly experienced by people with hearing loss of cochlear origin. The effect can be described as follows. Weak sounds are not heard at all. However, if a sound is slowly increased in level, once the level exceeds the persons elevated threshold for detection, the loudness grows more rapidly than normal. At a fairly high level, typically about 90-100 dB SPL, the loudness perceived by the person with an impaired ear is the same as the loudness perceived by a person with normal hearing. In other words, the loudness perceived by the person with an impaired ear catches up with normal loudness at high sound levels. This fact has important consequences for the design and fitting of hearing aids, since the aids must be designed so as to avoid amplification of high-level sounds; otherwise they would have a much higher than normal loudness and would be very unpleasant.

 

Most hearing aids are designed to amplify weak sounds, so as to make them audible, while not amplifying strong sounds. To do this, they use various forms of automatic gain control (AGC), sometimes called amplitude compression. The amount of amplification (also called gain) is made to decrease as the incoming sound level increases. In practice, the sound is filtered into a number of frequency bands or channels and the gain is varied independently in each channel, much like the sliders on a graphic equalizer vary the gain over a limited frequency range. This is necessary because, for almost all individuals with hearing loss, the loss is greater at some frequencies than at others; for elderly people the hearing loss is usually greater for high frequencies than for low frequencies. Hence, more gain is needed for weak high-frequency sounds than for weak low-frequency sound.

 

Fitting a hearing aid so as to compensate for the hearing loss of the ear on which it will be worn involves three stages. In the first stage, initial fitting, a prescription formula is used to prescribe the gains that should be applied in each channel of the hearing aid for various input levels, based on the audiogram of the ear. These gains are programmed into the hearing aid, using the software provided by the manufacturer. The second stage (which is sometimes omitted by dispensers trying to save time and/or money) is verification that the programmed gains have actually been achieved, by measuring the sound level close to the eardrum, using a miniature microphone. Nearly always, some adjustment of the gains is needed at this stage. The third stage involves fine tuning to suit individual preferences and to allow for individual differences in loudness perception.

 

We have used the Cambridge loudness model to develop methods for the initial fitting of hearing aids, in other words to develop prescription formulae. The loudness model is used to predict the loudness of sounds that have been amplified by the hearing aid. The fitting methods are based on the idea that, for sounds with medium and high levels, the hearing aid should not lead to a greater overall loudness than perceived by a person with normal hearing listening to the same sounds without amplification. The methods have been evaluated in a series of clinical trials, and have been shown to give good results, requiring only a small amount of fine tuning after the initial fitting.

 

One of the fitting methods has recently been extended to deal with a new development in hearing aids. Until a few years ago, most hearing aids only provided useful gain for frequencies up to about 5000-6000 Hz. However, several manufacturers have introduced hearing aids that provide gain up to higher frequencies. We have shown that, for people fitted bilaterally (a hearing aid in each ear), extending the frequency range over which gain is applied leads to an improved ability to understand speech under conditions where the desired speech comes from a different position in space to the competing talkers, which is a common situation in everyday life. Because current prescription formulae for hearing aids only give gain recommendations for frequencies up to 6000 Hz, we used our loudness model to develop a new prescription formula which gives gain recommendations for frequencies up to 10000 Hz. Preliminary evaluations have shown that use of this new formula does lead to satisfactory loudness and sound quality. The work continues.

 

Acknowledgments

 

The work described here was done in collaboration with Brian Glasberg, Michael Stone, Tom Baer, and Christian Fllgrabe, and was supported by the MRC(UK), the RNID (UK), Deafness Research UK, Starkey, and Earlens Corp. I thank Ervin Hafter, Lee Hafter, and Erin Riley for helpful comments on an earlier version of this article.