Measurements of vibrato

Josh Shaevitz

New Member
I am thinking of starting a small project to measure and document how vibrato has changed over the recent past in brass band cornet and flugelhorn playing and also quantify the properties currently used by a number of the top players. This started from a conversation Tom Hutchinson and I had during a lesson when he was visiting our band in NJ and my efforts to try to emulate what my ear can hear but my body can't yet create. Depending on how successful this is I imagine putting out a scientific paper and perhaps a pop-sci article for the ITG or brass band magazines.

Two questions:
1. Would this be interesting to any of you (I already know it's interesting to me)? While I recognize that people sometimes like to separate the art from the science, the physicist in me wants to know what really is going on in the sounds I hear.

2. Do you know of any past works that have done something similar? I found a good body of literature on vibrato in singing (that I hope to compare to) and a little bit on stringed instruments, but nothing on quantitative measurements in brass. Maybe I'm looking in the wrong place? I feel like there must be doctoral theses that approached this but haven't found them.

Thanks, Josh


Well-Known Member
It's an interesting question, but I certainly fall into the category you mention preferring to "separate the art from the science" on this particular question - philosophically, it should be possible to explain near-enough anything scientifically (with a complicated enough model).

Sorry if this sounds negative, but it sounds like you may be looking for a scientific answer to an artistic question?
Either way, no matter how much science/analysis you throw at this question, the key to unlocking and unleashing what works for you is still going to be exposure, experience and control (practice), surely?
No amount of scientific understanding of changes in wavelength or amplitude over time is going to make up for the fact that you have to be able to identify it with your own ears and produce what you want to produce (and learn which contexts require changes in approach and what they are).

The top-top players we're talking about here have a very high degree of control over their vibrato's and they use how much they want to use - there's a highly subjective element to how much these players want to hear in their sounds (which is probably partly a result of their particular backgrounds - are they brass-banders through and through? did they begin on cornet, flugel, trumpet or something else entirely? who did they like listening to as more inexperienced and impressionable players?), what a particular piece or genre calls for (eg: playing arrangements of operatic solos, going really strong on the vib is not only acceptable but appropriate), and in some cases even what the MD wants at a particular moment.

Even dealing with individuals who have been at the top for long periods, you've still got to somehow determine what their vibrato is, either independent of context (what it fundamentally "IS" - how?!) or with context controlled for (presumably the exact same piece played years apart, could be slim pickings? are you going to get enough data points to be statistically viable?)...
If instead we're just looking at analysis of individual players "right now" you've still got to somehow boil down to what their vibrato "is" in order to be able to ascribe values to it (or at very least, somehow find equivalent contexts for different players by which to be able to fairly compare).
This is the difficulty (IMHO, not a scientist) with trying to be scientific about this, if philosophically everything can be explained scientifically, then perhaps "Art" is one of the things we call situations that are just too complicated to model? (Part of which is that the model would somehow have to account for subjectivity?)

To answer your two questions directly:

"1. Would this be interesting to any of you (I already know it's interesting to me)?"

Honestly, probably not - unless there's some real-world applicability to it that I'm not able to gain through just listening .

2. Not that I'm aware of.

Euphonium Lite

Active Member
There is also the question on HOW vibrato is produced on brass compared to other forms of music making - for example strings is done primarily through the fingers. Brass can be done by oscillating the jaw, or shaking the instrument or controlling the air supply from the diaphragm - some are more "correct" than others (for example the latter 2 can lead to issues over control of the note, or by loss of control of the embouchure.

Also it depends on who you listen to and from where they come - brass players from the Yorkshire area of the UK for example probably have more vibrato than some other regions and that almost certainly will replicate across the world. I wouldnt say there is a technical "norm" or even something you can realistically measure - for most brass players its a case of finding a sound you like and replicating that, and vibrato is only one part - dropping the jaw to form a bigger mouth cavity, control over breathing, and choice of equipment (small v large bore, size and shape of mouthpiece) all play a part too. I wouldnt get fixated on one particular area unless you are certain you have everything else in place

Josh Shaevitz

New Member
Tom-King and Euphonium Lite, thank you for your responses. I think perhaps my intentions were not clear, so let me try to make a coherent argument for the two things I am interested in (which are indeed separate). The first motivation has to do with designing a training tool/method that will help players get closer to their desired sound. The second has to do with documentation/exploration of changes in vibrato over time and comparing this to trends seen in other areas of music (more of an academic exercise). The basic tools I have developed so far take sound files of solo cornet (there can be background instruments as long as they aren't too loud at this point) and extract a series of note waves using deep learning AI and classical signals analysis to automatically quantify for each note: pitch, pitch variation, vibrato frequency, vibrato amplitude envelope, bends... These can be visualized, compared, clustered etc.

1. Training:
As Tom-King points out, top players have both an excellent control of the pitch, loudness, brightness, etc. of their instrument and a sound concept that they wish to employ at any given moment in a given context. Through years of training (and natural ability) these players have gained these skills so that their emotions/ideas can be readily translated into sound out of the bell. However, I believe that for a much larger group of players, approaching this level of mastery is very difficult. The problem with the 'throw them in the pool' method of training is that if the student starts out far away from the desired goal, their process of listening, playing, listening, repeat may be very inefficient and may never reach their desired goal. Many have a fine vibrato, but few can create the nuance and subtle variability that I think really separates the top players from the masses. Further, I find for myself that I can have trouble separating what I am trying to do with my muscles from the sound coming out of the horn. I can hear all the nuance when I record myself and play it back, but this is harder in real time as I am playing in the practice room. Using a real time spectrum analyzer (such as the Analysis function in the mobile app 'TE Tuner') I can see the oscillations that are on a pitch when trying to create vibrato. By trying to match wiggles on the screen to patterns from a recording in addition to using my ear (i.e. using two modalities of feedback), I have had an easier time learning the art. In the psychology literature, it is well known that using multiple sensory modalities speeds/enhances training. For me, the combined audio/visual feedback works better than just listening to get close to the desired result. I agree that this kind of a tool is unnecessary as training progresses. But for many, I think this would be a valuable aid on the path to mastery of fine-scale instrument control and vibrato. This would also allow students to directly see the effect of the different methods of vibrato production that Euphonium Lite mentions.

2. Music History:
It seems clear that patterns and usage of vibrato have changed significantly over the last 50-100 years in brass bands (this is quite evident if you read the various arguments on this and other forums). Similar changes can be seen in orchestral instrument playing and vocal singing (both classical and popular) and a reasonably large literature exists (particularly for singing) that discusses this. For orchestral trumpet playing, there is even some work on regional differences as mentioned by Euphonium Lite. I am interested in looking at cornet vibrato and using machine learning to quantify changes in time (and perhaps region) and compare them to changes we see in other instruments such as the voice. There has been discussion of this in the voice literature, trying to link these changes to greater societal changes. Could something similar (worldwide or UK specific) also be at play? This is certainly an exercise in social science but given the computational tools I think it would be rather straight forward.

4th Cornet

Well-Known Member
I'm with Tom on this. If a study of vibrato is an interesting academic study for the OP then fill your boots. Expecting to then use it to inform and train players, like so many retrospective theories will kill the art and enjoyment for many.


Supporting Member
I guess my question would be, how can you best measure vibrato as both width (bending of pitch) and speed of vibrato are what seem to define vibrato.
Pitch relates to wavelength, so somehow writing a programing script to look at wavelength in a recording is potentially doable (assuming background noise can be filtered). Speed of vibrato seems to be a bit more of a challenge as you would have to look at the rate of change of those wavelength changes...but still doable. Might be good to connect with a CSE major. I realize Rowan University is not only PBB's rivals (ABB), but at least 1-hr away...but they've been doing some great stuff recently in the marketing aspect of Engineering with their partnership with the KEEN foundation.

Josh Shaevitz

New Member
DublinBass, thank you for the note. Here is an example of some of what the analysis extracts (pitch vs time) for the first ~20 seconds of Tom Hutchinson playing 'How Great Though Art' (sample here, great album if you don't already have it...). By segmenting out the different notes (red lines), I have so far shown that Tom uses a constant vibrato frequency on this song (i.e. the frequency at which he is modulating the pitch of the note--the 'wiggles' in the plot. This frequency varies by song and style, but not by much) and that there are three categories of notes found using a machine-learning clustering analysis: no vibrato (flat frequency, e.g. at t=~6s), an envelope vibrato (where the amplitude increases and then decreases smoothly, e.g. at t=~4s), and a decreasing vibrato where the note starts with strong vibrato and then the amplitude of the frequency modulation decreases to zero during the note (e.g. t=~13s).

I don't know anyone at Rowan besides the ABB folks, but I am on the faculty at Princeton and run an experimental physics group where we often do very similar kinds of analyses as part of our research. I will keep posting some results as I get them but with our semester starting this will be a bit of a side project likely until summer.


Well-Known Member
Interesting stuff, thanks for posting it Josh. I would be very interested to see your results - and perhaps to offer some spare-time physics modelling effort if it was of use. Do you know the PPPL people? I work on the JET tokamak as a physics modeller - mostly tending ancient Fortran code :) . I believe we'll be seeing quite a few of your NSTX people over here in the coming years to work on MAST?


Supporting Member
Josh... fascinating stuff!! I think some of the more interesting things that have also come up (based on these graphs), is 1) this is without amplitude (volume), correct? and 2) seeing where the vibrato comes within a phrase 1) some are very little hit vibrato and back off, 2) some are heavy vib an back off, etc...
(BTW...I think I friended you on FB to follow up more)

Product tMP members are discussing