Bpm Analyzer

ChrisHooker · 6 Apr 2012

Welcome to SynthMaker and the forum, Khela.

There are two versions of a "Tempo" primitive already in SM's toolbox. (One with stream output, one with green.) These output the BMP of the host program.

If that doesn't help you, I would recommend making a combination of that "Tap Tempo" schematic you already found with a compressor module you can also find in the toolbox -- Instead of using the Tap Tempo's button to tap, use the threshold/detection circuit of the compressor to detect peaks, using the "Bool to True" primitive to send a "tap" (trigger). You will have inconsistencies due to multiple instruments causing extra beats, skipped beats/pauses, fluttering of signal around the threshold... but you can try applying EQ filters before the detector to help stabalize the signal, and some kind of averaging/logic scheme to help with skipped/doubled beats.

There have been discussions on the forum about this, do a search and see if you get any other ideas from what you can find in those threads. Try searching for BPM, Beat Detector, etc.

Hope that helps ya.

tektoog · 6 Apr 2012

Hey, Have a look there It might help...

CoreStyler · 7 Apr 2012

The green privite is useful for displaying tempo from the host. You could read bpm from host program and maybe act with a triggers changes into the schematics. (but be aware of making any kind of precise triggered sequencing, SM is buggy with green timing) The stream is useful also for making triggered sequencing (in code), and you can read the tempo by analysing the signal 🙂

oddson · 7 Apr 2012

I think he's looking for something that will listen to a wave and output a BPM.

Not a simple task as it the case with any pattern recognition.

If I'm correct then the likely best steps are:

low-pass//envelope follower//peak locator//error correction//averaging or stabilizing output.

1) Low-pass: focus on the low-end of the spectrum where the beat is most-likely found

2) Envelope follower: get rid of the audio data and focus on the volume of the signal

3) Peak Locator: identify peaks (threshold surpassed) and track their position in time

4) Error Correction: Deal with peaks values that are off beat and possibly with missed beats

5) Averaging/Stabilizing output: Deal with minor rubato fluctuations that don't really constitute a tempo change as well as variations from the specifics of each beat's attack.

1) and 2) have modules in the stock toolbox. 3) is a bit of tricky coding - not a beginner thing but I'm sure someone could whip it together in under an hour. 4) & 5) are very tricky - likely here you'd come out of code and into green data processing See this thread: tempo detection from audio stream

khela · 12 Apr 2012

after a week of attempts, I have decided to surrender 🙁

tester · 12 Apr 2012

Split your problem into sub-topics (like described above for example), and interact with the forum learning how to solve particular problems. Like lowpass. Learn how to make it, search the forum how to make it, discover options of lowpass filters. And make one. If your lowpass will work on any input sound, but will work correctly - then focus on next topic. Of course chronology is not so important if you have data to cover other parts. That how it works 😉 So - don't give up!

oddson · 12 Apr 2012

khela wrote: after a week of attempts, I have decided to surrender 🙁

I was waiting to hear I was on track with what you were trying to accomplish.

I'm not surprised you're not able to do it on your own... but maybe with some help.

Start by confirming my assessment of what you're trying to do is correct or not.

martinvicanek · 13 Apr 2012

Here is a sketch of a BPM analyzer based on FFT of a time-compressed audio envelope. It does work more or less with a suitable input (Kool and the Gang: She's Fresh), however it also demonstrates how tricky this problem is - much harder than pitch extraction! I was surprised to see that most popular music is far from being periodic on the BPM scale even if it has a strong audible beat so a child could easily clap his hands to the rhythm.

stw · 14 Apr 2012

martinvicanek wrote: Here is a sketch of a BPM analyzer based on FFT of a time-compressed audio envelope. It does work more or less with a suitable input (Kool and the Gang: She's Fresh), however it also demonstrates how tricky this problem is - much harder than pitch extraction! I was surprised to see that most popular music is far from being periodic on the BPM scale even if it has a strong audible beat so a child could easily clap his hands to the rhythm.

😍 is there any source concerning that approach or did you just throw in some of your nifty ideas into the code box?

CoreStyler · 14 Apr 2012

martinvicanek wrote: Here is a sketch of a BPM analyzer based on FFT of a time-compressed audio envelope. It does work more or less with a suitable input (Kool and the Gang: She's Fresh), however it also demonstrates how tricky this problem is - much harder than pitch extraction! I was surprised to see that most popular music is far from being periodic on the BPM scale even if it has a strong audible beat so a child could easily clap his hands to the rhythm.

Works a little when around 120-150. But if you make 2000 or 20 it does not.

martinvicanek · 14 Apr 2012

stw wrote: is there any source concerning that approach or did you just throw in some of your nifty ideas into the code box?

It is rather naive to analyze the (squared) amplitude directly, it was just something to start with. Yes, there are sources which suggest better beat indication functions based on transients or onsets. Here is a good review: http://www.elec.qmul.ac.uk/people/juan/ ... P-2005.pdf. Then, to use SM's FFT I needed somehow to transform the BPM domain to the audio range, hence the time compressor. (Hehe, sometimes I wish time could be compressed - or dilated, for that matter - in real life, too! 🙂 ). Anyway, FFT is just one option to look for rhythmic patterns. Other approaches are based on comb filters or autocorrelation. For some reason I thought that FFT would be more tolerant wrt missing or extra beats. Here is an interesting discussion on BPM detection http://stackoverflow.com/questions/6570 ... rogramming and here is a promising algo http://www.clear.rice.edu/elec301/Proje ... talgo.html.

CoreStyler wrote: Works a little when around 120-150. But if you make 2000 or 20 it does not.

Hehe, 2000 BPM would probably challenge even the toughest techno freak 😀 😀 Yes, you are right, there are lower and upper limits based on the choice of design parameters. Pop music usually is in the range between 50 and 200 BPM, although there are exceptions (any speed metal fans here?).