"Sound" prediction!

:idea: What about predicting roulette by using the sound from the ball! :twisted:

Is it sound or mad?

I think I read somewhere that someone claims to listen to that sound in order improve his predictions intuitively. While that seems to require super human abilities, it is clear that the sound does change as the ball slows down and that, theoretically, it should provide information about the speed of the ball and ultimately about when it is about to fall down towards the wheel, which in turn tells us alot about where the ball is most likely to end up.

One could use a microphone and a pocket computer which digitally analyses the sound in real time. By sampling the sound and note the number coming up, one can use a regression to estimate the constants which constitutes the mathematical connection between sound data and time until the ball/wheel impact event. Add information about where the ball is relative to the wheel (say, a single click when the ball passes by the zero) at some point in time, and a prognosis of the kind “13 is the most likely number” would be possible! Hopefully with a precision such that the probability is comfortably greater than 1/35…

Irrelevant sounds such as voices should be easy to filter out, and their effect would be purely random anyway, not systematically affecting the prognosis. The sound approach would allow us to work with alot of data in the form of the digitalized soundwaves. The data is also a perfect picture of the reality, without involving human reaction times. Of course this approach would be extremely dependant on the details of the table and ball. New sampling and regression must be made for every new table.

One could tap the finger to the shirt close to where the mikrophone is, in order to mark the exact time when the ball passes zero (yes, here one must use human senses and reaction times unfortunately). The computer could recognize that for what it is meant to be. No switch and hidden clicking necessary. Easier to use than pointing a laser or what ever happened at Ritz in London…

Oh my Lord! Everything has already been done. AND published on the internet!


The link by no ways shows good conclusive works, but it proves that someone had the idea before and quite some energy to make som tests too. And among the thousands who have found that link?

Thing is, computing powers seems to 10-fold every 5 years, at the same dollar cost. This changes the landscape of opportunities. Those naive graphs in the link can hardly be said to represent the sound of the ball. They are quite pedagogical, but they do not represent the welth of sound data which COULD be processed.

Hello Rollo,

I did try to find opportunities with sound of the ball.
From noise that ball makes to increase in amplitude or sound strength when ball is passing closer point to recording device. Same as it may be valuable to Laurance to define right revolution needed with tilted wheel prediction, it may generate advantage to system such as my leveled wheel E2, to have constant starting time and increase in accuracy.

I did not spend much time with that subject but I did not find anything that can be used and give me any more advantage. Apparently when ball leaves edge of the wheel sound changes. For me that is to late for prediction and applying it together with any of my systems is not beneficial. Another option is to measure time period for each rotation by sound intensity change. Problem with that is that with intensity of sound changes very gradually within single rotation.
So triggering point that some kind of electronics may use can be easy mistaken by slightest noise or transients generated from the ball or from environment around the wheel. Well that part can be more investigated but it is not priority for me.

Scatter, I did detailed research on Bob Gordon’s wheel.
I do have all his DVD’s. I come across some interesting facts.
One day I will publish all research but that will take time because it does need some graphs, pictures and lot of explanations.
Interesting thing is that scatter is much better for us when wheel is traveling 3 sec per rotation compared to wheel that travels 4.5-5 sec. It does not mean that our prediction will be more profitable. That is balanced with how accurate we are with prediction.

My explanation for that is when the ball is in impact with number divider with faster wheel impact will be stronger, that takes more energy from ball causing it to stop in shorter time. Slow rotor does not takes so much of energy with first hit therefore ball continue traveling across pockets with very narrow angle almost in level with wheel therefore dividers have chance to hit the ball only at very low points of the ball, which does not take much of ball energy so the ball continue going until loses speed so it can drop deeper in to pocket.

First impact with rotor is the most important, and it would be better for us if ball is smaller in size or if distance in between pockets is greater.
At that moment the ball is coming towards rotor with angle of about 45 deg so it has chance to go deeper in pocket giving higher chance to divider to hit it closer to centre and to absorb more energy.

Another option is to measure time period for each rotation by sound intensity change. Problem with that is that with intensity of sound changes very gradually within single rotation.
Yeah, that´d be the way to go. Difficult certainly. Depends on how much info the sound of the ball reveals about its speed and position. I can easily imagine huge problems there. But with computing power exploding as it does, just a huge amount of experimentations might do wonders soon...
My explanation for that is when the ball is in impact with number divider with faster wheel impact will be stronger, that takes more energy from ball causing it to stop in shorter time.
Interesting! It sure is a funny phenomena. Important thing is that you have identified and quantified it. Be it then the will of God or whatever, I don´t care...

I just want the numbers for my regressions!

Hi guys,

I beleive the sound route is viable with some digital signal processing. Its one of the timing improvement concepts I’ve added to a fresh topic in the Members area.

Picking the amplitude (intensity) of the ball signal out of the background noise is difficult (still not impossible) as the ball noise is quite brodband and then there is background noise.

Ask yourself this - close your eyes and listen to the ball rotation. Can you tell what the ball is doing ? The answer is yes. Its the pitch that changes as the ball approaches and recedes from your vantage point.
That tells you the information is in the audio signal you just need to know what to look for and how to recover a clean signal to extract timing from.

We are acustomed to thinking in terms of signal amplitude (loudness) as its dimensions are tangible to us. The phase of a waveform or its rate of frequency change are not so tangible, but they are there and we can hear them, so they are extractible.

Just a few comments to the above:

Barnetts analyze of Scotts acoustic method has little if anything at all to do with what Scott is doing. It has nothing to do with a sound decreasing or descending but a sound that changes from whaa to woo.

Yeah well, listen to the wheel and you know what i mean. The dealers hears it and calls no more bets so its nothing fancy.

Using devices for detecting with sound when the ball was closest to the device (fixed point) is already mentioned and described with graphs in 1987 in Basieux, Pierre: Roulette Die Zähmung des Zufalls. It was intended for use with his own roulette computer at the time.

Obviously we wants smaller balls compared to the slot sizes. The casino supplyers has with intention lowered the fret height so that the center of the ball was higher than the height of the frets, which means it rolls longer.


"The specific gravity and trueness of diameter, of a given ball, plays a large effect on the outcome of a spin at the roulette wheel. The specific gravity of various ball substances, and their performance of co-efficient restitution (bounce) is shown hereunder:

Nylon…1,14 … (light)…Excellent
Acetal…1.42… (light)…Excellent
Ivory…1.75… (Medium)…Good
Composition (billiard ball)…1.80… (medium)…Good
Teflon…2.20… (Heavy)…Poor

Note that these lines were intended for Casino maintenance and Surveillance so Excellent bounce is in players eyes “terrible” Light, medium and heavy means the weight off the balls.

Anyone can hear the destinct difference between a dropping Teflon ball and a Nylon ball. Not to mention the huge difference in scatter patterns for those 2.

Recently I played few times next to system player.
He believes casino controls the ball against his system.
In one occasion ball jumped far from table, he picked it up and put in his pocket.
After a while manager come and asked him for ball insisting that only one person that could take was him. Because he was recorded on camera picking up something from the floor.
He denied. At home he cut the ball in half expected to find something.
But he found nothing.

I would always think that light ball bounces more.

Everything could be possible therefore and picking up sound and using in prediction.
It would be convenient because it would not need much of wheel watching.
But ball clocking if done right way can be very accurate.
First I thought it must generate huge errors based of change on human response time.
And it does but with few clicks and error correction it can be reduced almost to nothing.
I tested my way with deliberately injected time errors, simulating errors that player may create.
In most conditions error is reduced to only few milliseconds. So total error of wheel clocking + ball clocking does not exceed more then 3 numbers change in prediction.
With half wheel clocking it may be 5 numbers.


Regarding an accurate method of determining ball speeds…

Let’s say we make a video recording of an actual spin and analyze the ball speeds using a video program such as Avidemux. This would give us a super accurate model of how the ball speeds decelerate with time.

Couldn’t we then make several ball timings of an actual roulette spin in real time and ‘extrapolate’ or fit those (inaccurate) ball speeds to our accurate model obtained from our video study, thus yielding accurate ball speeds to use in our prediction?


Video spins in slow mode are only to analyze what is going on.
We can do many tests such as:
What is the difference if ball is 1.00 or 1.01 sec, how that effects remaining time, Will wheel speed effect ball traveling time, how ball deceleration changes, etc. Video spin also is good to test system.
Good system mo matter where you clock the ball should indicate same number. By shifting position we actually testing what system is doing with different ball speeds.
Kelly is right when he says at some spots 2 different speeds can have same result. Or linearity can be affected by slight deformation or tilt on the wheel. That is why tilted wheel is easier to predict and does not require so much accuracy in ball speed detection.

Yes 2 speeds can have the same result or 2 similar speeds can have 2 different results.

In the past week i have seen 2 different wheels where the last round lasted ~3.2 sec on the one and ~2.5 on the other. Not to mention the different scatter lenght.

If we at some point got into the same decelleration scheme at those 2 wheels, it would definately be screwed up in the final ball revoloution.

I doubt anyone can clock different wheels in the same way, computer or not. Each wheel is a closed circuit with a variabel: The ball material.

Yes, each wheel even looking same is different.
A little surprise if you will believe me.
I know for each ball speed on particular wheel how long will take next revolution.
Example 1
Rotation 1.000 sec next rotation 1.160 sec
Rotation 1.020 sec next rotation 1.190 sec (the difference changes)

Example 2
Rotation 1.000 sec next rotation 1.160 sec
Rotation 1.020 sec next rotation 1.200 sec

Example 1 and 2 after 1.000 coming same 1.160
But remaining time can be very different. Deceleration can not be used to define it.
Maximum difference that I found on same looking wheel and ball is 5 sec.
So same ball speed on one wheel can travel for 7.5 sec until hits rotor but on another wheel it will be 12.5 sec.

Oh i believe you, even though im not used to timing in those measures it resembles my experience.

Without going too deep into playing level wheels with visual ballistic i can roughly schedule the MO

  1. Rotor speed, estimated at least down to 0.05 sec.

  2. Entry point (roughly in the 4th or 5th last round.

  3. Marching pattern.

  4. Exit and Prediction in the 2nd or 3rd last round.

  5. Under Monaco conditions, throw 2-3 chips in on top in the final ball rev.

Best results is entry at 5th and exit in 2nd or final rev.

We can move this setup backwards but on a level wheel the edge will drop at least 5% for every revoloution we move backwards.

I know for each ball speed on particular wheel how long will take next revolution. Example 1 Rotation 1.000 sec next rotation 1.160 sec Rotation 1.020 sec next rotation 1.190 sec (the difference changes)

Example 2
Rotation 1.000 sec next rotation 1.160 sec
Rotation 1.020 sec next rotation 1.200 sec

Example 1 and 2 after 1.000 coming same 1.160


Are Example 1 and Example 2 timings taken from different reference points on the wheel? If we are to be able to predict the ball drop point based on ball timings, don’t we require consistency in speeds of consecutive ball revolutions?


Anam, we can test system for accuracy in detecting ball speed simply by shifting clocking point in steps.

When Kelly says 5% I think he is generous if he talks about real world.

On tilted wheel if system predicts zero the ball will hit zero on the rotor probably 90% of time. But on leveled wheel even if time is accurately taken and calculated that may not be the case. There are 2 reasons for that. One is that wheel is never leveled properly and clear of any damages on balls way. Another reason is that is hard to define which particular speeds will hit which diamond and on which spot. To know where is breaking point in between 2 diamonds. Earlier in prediction the time to work with is smaller therefore gap for error is smaller as well. If we try to clock ball of 0.5 sec per rotation next rotation would be probably 0.550 so we would have only 50ms difference. That is not enough to have reasonable prediction specially if we do not have 100% accuracy in ball clocking. Where Kelly is predicting he would have something as 200-300ms difference and if he has constant rotor or good feeling for speed change he would not have problem to predict correctly specially on tilted wheel where with reasonable tilt something as 70% of that would belong to same rotation. He can play selected spins and have wide gap for error\ so his chosen spins will be accurately predicted.
What I do not understand is Mark Howe’s DVD Where he first clocks the ball then he clocks wheel. Even if he has some advantage, by doing it that way he loses a lot in accuracy. The bal clocked at that speed range requires much more then to define is it right revolution.
Recently I received email from him where he proudly claiming big improvements. Some newspapers advertised his computer and he has more then 600 orders so maybe he fixed it. I hope that he does not screw industry.

Thanks Forester, now I get it.