Very Interesting Comment By James May on Tone Dexter

[caballero59]

"For all intents and purposes, the condition of your strings when you train your ToneDexter doesn't matter. The end result will be virtually the same. This is because what it is learning is the characteristics of the instrument as hear by the mic in position.

The string condition is common to both the pickup signal and the the mic signal and therefore washes out of the equation. It's the same reason that you can play with a pick or with your fingers during training and get the same result." - James May


James, I hope you see this and can chime in. I'm thinking this comment may be a key to my misunderstanding of the foundational principle of TD works. If the condition of strings and pick vs fingers "washes out", then won't the quality of your box wash out also since the guitar box is common to both inputs? I'm guessing what he's saying here is that when you divide the FFT's to perform the convolution, that products common to both of them divide out. So the IR becomes a mapping between differences? So what tone are we actually ending up with and what is it's origin?

[The Kid!]

That confuses me because my guitars sound different with new strings vs. old. If it is picking up the wave map through a mic, won’t the mic hear the difference too?

[AlfredFelix]

If I understand it correctly, please re-see the word "if", his point is that the Tone Dexter is picking up the difference between the mic'd signal and the input pickup signal. Therefore, it will make those same adjustments everytime it hears the guitar. Hence, if you play with your fingers, it is adjusting (made up numbers) 13 different frequencies by specific amounts. Let's say that is how you trained it. However, then, if you play with a pick, it will make those same 13 adjustments by the same amounts. So, it would translate evenly, even though you only trained it one way. Does that sound right/make sense? That's at least what I took the quote to mean. I am very capable of misunderstanding though. Lol.

[Peter Z]

That makes sense and sounds right, Alfred. I see it in the same way.
If you have old strings, the pickup signal has less highs - and the mic signal has less hight. So the difference will be (roughly) the same.

[guitaniac]

What I find most important here is the anecdotal evidence that string brightness doesn't make a difference. The Audio Sprockets folks have surely observed much more anecdotal evidence in this regard than any of us.


On the other hand, I just got a dose of anecdotal evidence that string selection can definitely have a pro or con impact on amplified sound. On Thursday night I ran sound for a player with an Aura system equipped guitar. I'd run sound for him before without much problem, but on this occasion the boomy lows and low mids were quite severe. It turned out that he'd just switched to silk and steel strings which caused problems in two ways #1) they were weaker in the high frequencies, so it was harder to get an even tone. #2) they were quieter than his regular strings, so I needed to apply more gain (which invited feedback) to get a usable level.

[Cuki79]

I am not James but I can give my 2 cents


As shown by the picture above. Making IR is about comparing the signal coming from the pickup (top) with the one from the mic (bottom). You can see that in the "signal chain" the common parts are fingers, pick, strings and bridge.

The difference comes from the pickup sensing the bridge while the mic senses the top, box, air, room and mic response. We will focus on this difference and not on the common part.

The principle of IR convolution pedal for acoustic (Fishman Aura, Tonedexter, Yamaha SRT...) is to treat the pickup signal and apply the responses of a "virtual" top, box, air, room and mic. It is shown in the picture below. The tone coloration applied by the pedal should be independent from the common parts.



Why? well if look carefully at the last picture:

* If one uses old strings, the treated signal will sound like old strings. If one uses new strings, it will sound like new strings. Because the strings are still in the signal chain.

* If the fingers are mine... it will sound like me. If the fingers are Clapton's... It will sound much better

If the IR process was sensitive to the strings.... it would be also sensitive to the player who trained the IR. It would not only model top, box, air and mic.

It would model everything.

It would not be called an IR pedal: It would be a talent pedal!!!

Imagine, you could just ask your best guitarist friend to train the pedal... and then you'd sound like him just by putting the pedal on (Something like auto-tune?).

More seriously, the goal is to capture the response of the top/box/air/mic... So the process is made to be independant of all other factors.

[caballero59]

The key to this is that the bridge is common to both processing paths but diverges at the point of pickup vs top.

[gfirob]

This is anecdotal but I think James May told me on the telephone in a conversation about the Tonedexter after I bought it and asked for some clarification, that new strings were better than old strings since it would improve the information going into the Tonedexter during training in the first place. The caveat in this is: 1) I don't really understand how the Tonedexter works, 2) I may have misunderstood him in the first place, 3) I may have had this conversation with somebody else entirely.

It makes sense to me though, because the better the microphone and the better the position, the better the wavemap. That is, the better that training signal is, the more the device has to work with.

Also, I really enjoy Cuki79's graphic explanations, not only because they are interesting, but because they are such a great example of over-thinking. It suggests to me that if you could really look under the hood of a Tonedexter, your head would explode...

[gfirob]

Also, Cuki, at some very small level, I think there must be an element of time in this since it will take the sound longer to go from the top of the guitar to the microphone than from the string to the pickup. I am not talking about a lot if time, but it must be part of what we are calling "air".

[caballero59]

Cuki's contributions on this subject have been very valuable for me. His latest graphic cleared up my question of where the signal paths diverge.

I'm thinking the term IR is a misnomer with respect to how the ToneDexter operates. A true IR would be the characteristic response of the guitar box to a generic and infinitely short input signal. The wavemap from ToneDexter is a mapping between the pickup and mic responses. Just trying to get my head straight on all this.

[jonfields45]

Originally Posted by caballlero59

Cuki's contributions on this subject have been very valuable for me. His latest graphic cleared up my question of where the signal paths diverge.

I'm thinking the term IR is a misnomer with respect to how the ToneDexter operates. A true IR would be the characteristic response of the guitar box to a generic and infinitely short input signal. The wavemap from ToneDexter is a mapping between the pickup and mic responses. Just trying to get my head straight on all this.

Any Finite Impulse Response Filter (FIR) can be excited by an Impulse, which will cause it to output its Impulse Response (IR), which will be each filter coefficient (the filter's programming) in succession. The Impulse Response for an electric guitar amp modeler is really just the amp's impulse response. For TondeDexter, the filter active during playback is a FIR filter with an IR, but deriving the programming for that filter (being refered to as its IR in this discussion) is the more complex problem described by Cuki79 (1979 -- the year I got my engineering degree...).

The earlier comments about time and air, are not really in the ToneDexter realm. A response long enough for a human to notice is a much longer period of time and in the domain of reverbs. The time being consumed by ToneDexter's FIR delay is about reinforcing and suppressing frequency content to get you from the pickup's tone to the mic's tone. Any longer tail to the signal has its origin in the guitar and ToneDexter is only filtering it.

Finally, pickup systems that introduce purposeful nonlinearity (distortion) like the Baggs Session VTC (or an electric guitar into a tube screamer) are generating frequency content that was never in the original guitar pickup's output. There is no way for ToneDexter's filter to compare the mic and pickup, when the pickup has artificial harmonic content from distortion, to do anything but try to suppress that distortion, which will be a function of what notes are being played and completely useless for programming a fixed playback filter. That distortion, is in-theory, why you bought the Session in the first place.

[Cuki79]

John Fields is right it is a Finite Impulse Response (FIR) filter

If you want to know how Tonedexter works, just read the patent. It's very well written.
https://patents.google.com/patent/US...2c583%2c088+B1

This is anecdotal but I think James May told me on the telephone in a conversation about the Tonedexter after I bought it and asked for some clarification, that new strings were better than old strings since it would improve the information going into the Tonedexter during training in the first place.

I shall explain why James said that (in my opinion).

The impulse response fully characterizes the response of a system. Take the Fourier transform and you'd get the frequency response.

As a result, a proper built FIR filter that would describe top, box, air and mic response should not depend on the strings.

However nobody is able to extract a perfect IR from a guitar. In the scientific litterature, there have been experiments with hammer, robots... and of course human players.

If you calculate the frequency content of an impulse you would find that the shorter the impulse the wider is the frequency content. So if you want to record the impulse response of a system you should make the shortest excitation possible in order to get the full spectrum response.

That's why, people record room reverb IR by clapping hands in the room and recording the response (echo). The hand clap plays the role of the impulse.

Another possibility it to use a speaker and perform a sine sweep. In this case you just sweep over all the frequencies and then rebuild the IR.

In the case of the guitar, none of those technics work or is easily feasible. You have to start with human playing. However human playing does not cover the whole frequency response of your top/box/air/mic... Remember that the highest fundamontal note on a guitar is close to 1KHz.

It means that for the higher end of the spectrum, you are working with harmonics and sometimes high order harmonics that have very few energy. So you get a really bad signal to noise ratio. So Tonedexter like my algorythm are all about extracting valuable information and dump noise. Of course, bright new strings will sound sharper and will have stronger harmonic contents... But at the end of the day, with Tonedexter having quite a strong low pass filter... I think it is quite negligible (my 2 cents).

To come back to Tonedexter's recipe. Here is how I understand it. They make hundreds FIR filters from your playing. Each time they evaluate the "value" of the nth FIR filter by

comparing it to a template of what a known good filter looks like

if the nth FIR filter passes the test it is added to an accumulator. Once the process is completed. The accumulator gives the final FIR filter.

Comparing to a template enable them to avoid "ringings" due to bad signal to noise ratio.

The fact that Tonedexter's method use a template as reference is the main reason why people have not been successfully making IR for the Lyric. My process does not work the same way for example, it makes no assumption how the final filter should look like.

If you are interested to learn things on IR extraction, you can read some of my early posts on the subject. Those were early stage of my IR process, I don't really work like that anymore but I was learning by trial & error... (keep in mind that Tonedexter's patent was not published then... and the pedal was not released yet either)

https://www.acousticguitarforum.com/...d.php?t=447334

https://www.acousticguitarforum.com/...d.php?t=448436

https://www.acousticguitarforum.com/...d.php?t=483802

[James May]

To the OP: unless the strings are really quite old and very dull, it won't make much difference to the trained result. New, brighter strings won't really give better results than moderately dull strings. This is because there is still enough high frequency information coming off the dull strings to satisfy the learning algorithm. Even your pick attacking the string has considerable high frequency information in just the attack.

Thanks to Cuki for the picture and explanations, well done. The one point I would add is this: not only is the system learning the body response and air and mic, but it is also learning how to undo any bad frequency response the pickup may have. In other words - correct for it.

Thanks to Jon Fields for pointing out that yes this system does depend on things being linear. The Session VTC is one example that has purposeful compression and that does indeed get in the way of training.

Regarding the amount of time that the WaveMap captures, I'll go ahead and reveal something that has not been precisely revealed before. Our impulse response is about 100ms long. This means the IR can affect frequencies as low as 10Hz if it needed to. It also means any short echos that occur in the chamber of your instrument are captured, and it quite effectively captures the personality of your instrument. To put that in perspective, a very small amount of the room's character can be heard within 100ms, but not much. 100ms is much too short to convey a reverb or room simulation, for example.

[Mr. Jelly]

Originally Posted by caballlero59

This is because what it is learning is the characteristics of the instrument as hear by the mic in position.

Anyone have an educated explanation of what he is trying to convey by characteristics? Please no guesses. I can guess as well as anyone.

[jonfields45]

Originally Posted by Mr. Jelly

Anyone have an educated explanation of what he is trying to convey by characteristics? Please no guesses. I can guess as well as anyone.

BSEE 1979, MEE 1980, not guessing...

The difference between the frequency content of the microphone and the guitar's pickup. ToneDexter in playback mode is simply a very fancy equalizer. The magic of ToneDexter is a simple, do-it-yourself, procedure to program that equalizer as modern digital electronics makes implementing that equalizer with a digital filter trivial.

Earlier in this discussion, ToneDextor's designer is quoted asserting that this difference was not materially impacted by the freshness of strings. That statement would require experience with the system to state and there is nothing obvious in the "engineering" of the system to know that is true a priori.