I made a small routine to extract rpm from sound analysis and I’ve found that the main problem is the choice between the length of the interval and required accuracy. Without entering too much in the details when you have to analyse the engine sound you have obviously to separate the audio sequence in intervals and then find the rpm in each interval. But, the more you reduce the length of the interval (trying to follow the rpm variation), the more you are sacrificing your sensitivity on the rpm determination (meaning for example that you’re unable to distinguish 17500 from 17800 rpm). “Unfortunately” a F1 engine has a fast rate of rpm variation, so you have to choose between ability to follow the rpm variation, but with poor precision, or great accuracy but on a large time interval, meaning an average of rpm in that interval and a stepped signal. The latter is surely good for an analysis of the peak rpm (just consider a straight line near peak speed, rpm variation is relatively slow) but the tacho during the race requires to follow the rpm variation => bad accuracy. Add to this that any perturbation on the signal can generate peaks at different frequency, and that downshifts or TC can generate more confused signals. Once you have the graph of the rpm during the full lap it’s quite easy to eliminate meaningless data and also to improve precision post processing the data, but in real time it isn’t, hence the crazy figures we see sometimes.
About the indication of selected gear, I had during the race the suspect that the downshift was manually operated, it seemed to me there was an inconstant lag between the actual downshift and the change in gear indication, at least that was my impression. Upshift is quite easy to recognize, but downshift can be a nightmare.
About the marcush observation that it’s not a rpm race, I totally agree. You can just imagine the jokes here in Italy when BMW declared they broke the 19000 rpm barrier in qualifying and the day after Ralf car stopped at the 4th lap in the race...
But the point of a good analysis of engine sound isn’t only to know peak rpm, it allows lot more. For example, analysing Alonso start in Malaysia I was able to obtain, still considering the approximations, the speed vs time from about 140 km/h to about 280 km/h. That’s the polynomial fit (speed in km/h, time in seconds) :
v = -0,0137 t^4 + 0,3328 t^3 - 4,4305 t^2 + 39,32 t + 138, 0 < t < 8
A more accurate analysis (mine surely isn’t as good as it could be) and the knowledge the teams have of all the involved parameters (tyre grip and his variation with speed load, typical aero coefficients etc) would allow to know a lot about opponents cars/engines just from the engine sound.