Re: Finding audio frequency
Re: Finding audio frequency
- Subject: Re: Finding audio frequency
- From: Andrew Capon <email@hidden>
- Date: Tue, 26 Jan 2010 16:44:05 +0000
A good overview of the differences between time domain, frequency domain and using stats is http://www2.cs.uregina.ca/~gerhard/courses/Audio/PitchDetection.pdf
The most successful frequency tracking for guitar I have seen is the Axon range of converters, they use stats based neural networks. DFT solutions are usually too slow for realtime work.
Also are you looking at polyphonic pitch detection for the guitar as usually more than one string is playing at a time unless you have individual audio for each string which is how commercial guitar to midi converters work?
Andy
On 26 Jan 2010, at 16:35, John Proctor wrote:
> I would have thought that the average number of times that the string oscillates per second would be equivalent to the average highest frequency present in the signal.
>
> John.
>
> On 26 Jan 2010, at 16:30, Paul Bruneau wrote:
>
>> On Jan 26, 2010, at 10:29 AM, Ian Kemmish wrote:
>>
>>> My analysis/resynthesis module needs accurate estimates of frequency. What I do is take several FFTs at different times. Find some peaks in each FFT. Sort the peaks roughly into bins (one for each harmonic, basically). Throw out any obvious outliers in each bin. Calculate the media frequency for each bin. Then look at the median frequencies you have left, and see if they form an obvious harmonic series. If they do, the harmonic is the least common difference between them.
>>>
>>> In the real world, remember that you have to deal with situations where the fundamental is not necessarily the loudest harmonic. And for plucked string instruments, the frequency of oscillation is falling ever so slightly for as long as the string sounds (the time-averaged tension of the string is reducing as the amplitude decreases). And for stiff strings the partials aren't *quite* harmonically related (which is also why looking for zero crossings in the time domain isn't the best idea - the shape of the waveform changes gradually).
>>>
>>> Hope this gives you some ideas to try.
>>
>> I want to thank everybody for all the pointers. Thanks for the links to the FFT info--I had heard of this before and I'm glad to know it's something I should learn more about.
>>
>> It's not a guitar tuner (those are readily available :), it's more of a guitar "reader".
>>
>> To Ian, I hope I don't have to get as complicated as above for this. In my mind (maybe in the real world), the string is vibrating at a fundamental rate (that I admit will change over time). This is all I care about. So with some internet-provided audio as an example, I expect to see something like this (I hope this list will accept a small picture):
>>
>> <pastedGraphic.png>
>>
>>
>> I know that there are harmonics in this, but all I want to know is how many times per second (on average over my sample as was mentioned by Ian) the string oscillates.
>>
>> Thanks again, I'm reading all your links :) _______________________________________________
>> Do not post admin requests to the list. They will be ignored.
>> Coreaudio-api mailing list (email@hidden)
>> Help/Unsubscribe/Update your Subscription:
>>
>> This email sent to email@hidden
>
> John Proctor - Developer
> Camel Audio - www.camelaudio.com
> Twitter - http://twitter.com/jprocto
>
>
>
> _______________________________________________
> Do not post admin requests to the list. They will be ignored.
> Coreaudio-api mailing list (email@hidden)
> Help/Unsubscribe/Update your Subscription:
>
> This email sent to email@hidden
_______________________________________________
Do not post admin requests to the list. They will be ignored.
Coreaudio-api mailing list (email@hidden)
Help/Unsubscribe/Update your Subscription:
This email sent to email@hidden