The Guitar Experiment

The results of the guitar study at Lancaster University have been published in The Journal of The Acoustical Society of America. This page will provide a short introduction to the study and present its main results. If you want to see all the results and read their discussion in detail you can download a copy of the article by following the link below:

Carcagno, S., Bucknall, R., Woodhouse, J., Fritz, C., and Plack, C. J. (2018). Effect of back wood choice on the perceived quality of steel-string acoustic guitars. J. Acoust. Soc. Am., 144(6), 3533-3547. doi:10.1121/1.5084735

You can also download a copy of the raw experimental data on the OSF project webpage.

  1. Introduction
  2. Methods
  3. Results
  4. Conclusions
  5. Photo Gallery

Introduction

The steel-string acoustic guitar is one the world’s most popular musical instrument. Unfortunately, some of the most prized woods for guitar backs (e.g. Brazilian rosewood) are from unsustainable sources, a problem highlighted by organisations such as Greenpeace and the Global Trees Campaign. Guitar manufacturers are also well aware of this issue and have taken a range of initiatives to address it, such as using different and more readily available woods as a substitutes for the traditionally used tropical hard woods, and researching into new materials for guitar construction, such as high-pressure laminates. The challenge is to find woods that are sustainable and cheap without compromising on the sound quality and playability of the instrument. Although acoustic guitars are routinely compared in guitar magazines, these comparisons are susceptible to many different kinds of bias, including factors such as perceived brand quality, price, and visual appearance.

Using an interdisciplinary approach, the current study evaluated the impact of different back woods on the acoustic characteristics of a guitar, as well as the perceptual effects of back wood choice on the playability and sound quality of acoustic guitars assessed by expert players using a controlled, “blinded,” experimental procedure.

Methods

The Guitars

Six handmade Fylde Guitars guitars built on commission from our lab. All the six guitars were based on the same model (Falstaff model), and were built to the same dimensional and material specifications, except for the back and side plates. The woods used for the back and side plates of the six guitars were: Brazilian Rosewood, Honduras Mahogany, Indian Rosewood, Maple, Sapele, and Claro Walnut. These six woods cover a wide range of prestige among players, availability, and price. A picture of the six guitars still unfinished is shown below.

6guitars2

Laboratory Tests

The acoustic characteristics of the guitars were assessed by measuring the their input admittance at the bridge. This measurement provides a “fingerprint” of the vibrational characteristics of a guitar that will be reflected in the sound it produces.

We ran several perceptual tests. The most important ones were a blinded rating test, and a blinded ABX discrimination test. Guitarists were tested in a dimly lit room while wearing welder’s goggles during these blinded tests, so that they could see their fingers on the fingerboard to play each guitar accurately but could not identify the wood the guitar was made of.

For the blinded rating test guitarists played each guitar for a few minutes and rated it on a 1-5 scale for its overall sound quality and playability (as well as a number of other attributes).

For the blinded ABX discrimination test guitarists played first one of a pair of guitars (guitar A), then the other guitar (guitar B), and then they were given again one of the two guitars to play, and had to decide if it was guitar A, or B.

Online Tests

Two perceptual tests similar to the ones that we ran in the lab are available online. The results of these online tests have not yet been analysed or published, and you can still take part in them (whether you’re a musician or not). This will give you a chance to listen to the different guitars and test your ability to tell them apart. These tests are based on recordings of the six guitars made by a professional guitar player. You can click on this link to take part in the online tests.

Results

Bridge Admittance Measurements

The figure below shows the input admittance of the six experimental guitars. The bridge admittance of a guitar of a different make and model, a Yamaha FG-403MS, is also shown for comparison purposes (note that you can zoom into selected areas of the figure and remove/add traces by clicking on the respective entry in the legend).

All the guitars show three strong low-frequency peaks. Previous research into guitar acoustics indicates that these peaks are responsible for the dominant aspects of the sound radiated by a guitar. These three low-frequency peaks are very similar between the six experimental guitars, while the trace for the Yamaha guitar appears markedly different from the other six. The first two of these peaks can be influenced by the back plate, while the third peak is mostly influenced by the front plate, and should have only a very weak influence from the back. Overall this points to residual differences in the top plates as the main cause of the subtle differences in these low-frequency peaks among the set of six experimental guitars. The reasoning behind this conclusion is that 1) the third peak, which should essentially be independent of the back plate, shows variations at least as big as those of the other two peaks 2) the three peak frequencies show strong correlations.

Blinded Ratings

The figure below shows the average ratings for “overall sound” quality and playability given in the blinded rating test by 52 guitar players, error bars are ±1 standard error of the mean (s.e.m.):

It is evident that the average rating differences in the sample of players that we tested are minimal (the rating scale ranged from 1 to 5). A Bayesian analysis of the ratings did not show any credible differences in ratings between the guitars for either “overall sound” quality or playability. The results of this analysis are summarized in the figure below, which plots the estimates of the rating differences and their 95% credibility intervals (CIs). The fact that credible differences could not be detected does not in itself prove that such differences do not exist, but the fact that the 95% CIs are quite narrowly centred around zero clearly shows that even if such differences existed, they would be pretty small.

ABX Discrimination Test

The fact that players do not seem to show preferences for the “overall sound” or playability among the six guitars does not say anything on whether they are able to tell the guitars apart. For example they may well be able to distinguish them, but like each one for its own characteristic tonal quality. To find out whether guitar players could indeed easily discriminate between the guitars we ran an ABX test. For this test guitarists played first one of a pair of guitars (guitar A), then the other guitar (guitar B), and then they were given again one of the two guitars to play, and had to decide if it was guitar A, or B. You can read the methodological details of the test in the paper. The results of this test are shown in the figure below, which plots the d’ discriminability index. To read the figure the important facts to know are that a d’ of zero corresponds to chance performance (complete inability to tell the guitars apart), as a rule of thumb a d’ of one represents poor discrimination performance, a d’ of two correspond to a mediocre performance level, a d’ of three to a good performance level (d’ can range all the way from zero to infinity, but d’ values at or above five can already be considered pretty close to near perfect performance). It is clear that players had a hard time telling the guitars apart, although the average d’ values (±1 s.e.m.) shown in the figure below vary somewhat as a function of player professional status and the specific guitar pair tested the discriminability of the guitars was overall poor.

The figure below shows the d’ estimates and their 95% CIs obtained with a Bayesian analysis of the data. When considering the scores across all guitars and player categories the 95% CI ranged from 0.21 to 0.8 (not shown in the figure), indicating that overall, discrimination performance was poor. There were no credible differences between player categories or guitar pairs (also not shown in the figure below, but see the supplementary material of the paper).

ABX test with synthesized guitars

A subset of seven players performed another ABX test which used sounds synthesized on the basis of the measured bridge admittances of the guitars, rather than a playing test using the actual guitars as in the previous ABX experiment. This test allowed us to include a control condition with a guitar of a different make and model, a Yamaha FG-403MS, to check that players could easily do this task when the guitars to be discriminated differed in more than just the back wood. There were also other reasons to run this additional test, which has both advantages and disadvantages compared to a playing test with the actual guitars, you can read the details in the paper if you’re interested. The average d’ values obtained in this test (±1 s.e.m.) are shown in the figure below:

The figure below shows the d’ estimates and their 95% CIs obtained with a Bayesian analysis of the data. There are three important results to highlight here: 1) discrimination performance between two guitars of a different make and model (Ind. Rsw. vs Yamaha) is good, with a d’ around three, this shows that players can quite easily do this task when the guitars differ in more than just the back wood. 2) discrimination performance for the other guitar pairs that had been previously tested in the playing test was poor, with d’ values around one. 3) discrimination performance for the guitar pair with the most divergent bridge admittance functions (Ind. Rsw. vs Sapele) was better than for the other guitars, but nonetheless mediocre, with a d’ around two. Regarding this last result, it should be noted that the analysis of the bridge admittance functions strongly points to residual top plate differences, rather than to back plate differences, as the main cause of the differences in bridge admittance between the guitars. So even this mediocre performance level in discriminating the Indian rosewood from the sapele guitar is very unlikely to be due to differences in the back plate woods.

Conclusions

Overall our results indicate the species of wood used for the back and sides of a steel-string acoustic guitar has at best a marginal impact on its acoustic properties and perceived sound, and that cheaper and sustainable woods can be used as substitutes of expensive and endangered woods without loss of sound quality. This page presented only a quick summary of the main results of the study. The paper presents the results of additional tests and more in depth discussions of the results shown here. The Leonardo Guitar Research Project, a project that is completely independent of our own study, has so far reached similar conclusions.

Photo Gallery