-1

I would like to know if pitch detection is a solved theme for an instrument like a bass (bass tones).

Is it, or there is always a percentage of error rate on the detection?

Meliodas
  • 17
  • 5
  • wait, a bass doesn't even have discrete tones, how can you have an error rate? – Marcus Müller Jun 20 '20 at 10:43
  • 1
    And it seems your definition of "solved estimation problem" is "makes no errors". Bad news: according to that definition, there's no solved noisy estimation problems out there, at all, and there won't ever be. – Marcus Müller Jun 20 '20 at 10:45
  • 2
    Defining pitch, much less than solving for it, is even difficult as overtones aren't always harmonic. Check this out: https://www.dsprelated.com/thread/7902/the-spectral-complexity-of-a-single-musical-note – Cedron Dawg Jun 20 '20 at 11:29
  • @marcusmueller I was referring that, in a pitch estimator for a bass where are searching for the fundamental frequency, you get these results. So, in a set of results, There are got a error rate. That was the point. Then is inevitable to get an estimator with a 0% error rate. Thanks. – Meliodas Jun 20 '20 at 16:44
  • @Cedron DawgThe article's title is "Automatic Music Transcription", I think this is a different pitch detection where is implicate polyphonic music. I'm talking about a single instrument, a bass one. When you use a method like Harmonic Product Spectrum, Autocorrelation, etc; you can search for the fundamental frequency without that kind of problems. Thanks. – Meliodas Jun 20 '20 at 17:14
  • 2
    If you want the frequency of the fundamental tone, I would suggest this approach: https://www.dsprelated.com/showarticle/1284.php On a pure tone, it is exact. This is different than pitch. Pitch is what it sounds like. You can have a waveform with the fundamental missing, yet the pitch is still at the fundamental. – Cedron Dawg Jun 20 '20 at 17:42
  • 1
    Pitch is a perceptual metric, which means it’s not directly quantifiable. Pitch detection is a process by which features of a waveform can be correlated to our perception, but it cannot be equated, and as such can not be directly solved. There are pitch detection algorithms, some of which work better than others, but any error rate would require a presumption of what is correct. In reality, the best metric is probably whether or not people think it works. – Dan Szabo Jun 20 '20 at 17:59
  • @Meliodas your description of an estimator makes no sense. Even if there was a "single" fundamental frequency, (hint: there's no such thing. The existence of timbre says that a tone evolves over time. Really, this discussion comes up every few months here.) you don't get an error rate, but an estimator variance: since a frequency is a real value, the probability of estimating the same value as reality predicts is literally and provably 0. You'd need to define some bounds for "what is still a correct estimate", and there everything goes downhill – Marcus Müller Jun 20 '20 at 18:24
  • because suddenly your estimator quality is mainly defined by what you still say is an OK estimate. So, really, error rate makes no sense for a continuous number to be estimated. But let's not mince words and instead act as if you said "estimator variance": you can never push estimator variance to zero. See things like the Cramér-Rao lower bound, or rate-distortion theory. So, your problem is still undefined. – Marcus Müller Jun 20 '20 at 18:28
  • "you can search the fundamental without that kind of problems": oh, so do that then ;) – Marcus Müller Jun 20 '20 at 18:28
  • By the way, your title question "is it solved" has already been answered in an answer to your questions. – Marcus Müller Jun 20 '20 at 18:30
  • @marcusmueller In the post Knut told me that the fundamental pitch tracking is a long standing challenge, and i had understood that it was a challenge that had been standing a long time, and therefore, there is a lot of information. My bad, sorry. I guess I'm going to delete the question. – Meliodas Jun 20 '20 at 18:51

1 Answers1

1

Pitch is a human psychoacoustic perception phenomena, so an absolute ground truth might not be well defined enough to “solve”, except in some statistical sense, especially for a string bass, whose waveforms can evolve over time in both frequency and inharmonic as well as harmonic spectral composition.

A pure perfect periodicity for many natural sounds does not exist, which is likely part of what makes them interesting enough to want to hear them.

hotpaw2
  • 35,346
  • 9
  • 47
  • 90