Syntrillium Backup Forums :: View topic

sync · Posted - Thu Feb 06, 2003 5:44 pm

I have several recordings of voices. For each one I need to determine what is the average pitch of the speaker. I also need to know what is the range of pitch. I've looked at both the Spectral view and Frequency Analysis but I can't figure out how to use them to get what I need.

lonah · Posted - Fri Feb 07, 2003 3:06 am

Complicated....

With the spectrum, you can see which harmonics are used for your vocie sample. (please excuse my *frenchy* english). Theorically, the lowest harmonics (fondamental) gives you the pitch ( in Hz ). Only problem is that it won't always be right (depends of your sound...).

Second problem is "average pitch"... do not see what you mean. Average frenquecy? If you have a singing voice, ok, I can see, but if it is a speech, beeuuuhhh (french typical answerX()

In fact do you just want a numeric data (aver. frequency used... and then, I propose you to use MatLab), or do you want an other thing?

Desesperating

Lonah.

ozpeter · Location: Australia

The Cool Edit Pro Frequency Analysis window should give you what you want - check out its features in the product's help file - you will probably need to highlight the area of the waveform of interest then press 'scan' to get an overall picture of the features present. But there's no avoiding becoming familiar with its operation and how to interpret what you see.

Alternatively, if you find and download Paul Kellett's Wavetools package, (free I believe) it has a frequency analysis component which may also assist you - it identifies the peak frequency at any given moment and it can show graphically the maximum level reached for each frequency over time.

- Ozpeter

SteveG · Location: United Kingdom

Analysing speech has traditionally been a difficult thing to do, and most of the successful attempts to do this use rather sophisticated analysis tools to do it. If you want some sort of idea of what you're up against, there are several websites that will give you some sort of an idea - and there are some specific tools here that you might want to try. A Google search will reveal plenty more information on the subject as well. I was going to post a reference to Tony Robinson's 'speech analysis' page, which is the first thing that should come up on a Google search using the term 'speech analysis', but for some strange reason the forum gets seriously upset about the link to it.

Unfortunately, the CEP analyser isn't really intended for this task, as its temporal storage ability is limited to four snapshots, unless you want to drive yourself demented by breaking your files up into very small chunks. This would work, because it's quite possible to output the data straight into Excel, and reconstruct the information there. But the amount of effort involved is considerable, and I would have thought that an analyser that was intended to do the job would have been a far better bet.

sync · Posted - Fri Feb 07, 2003 9:11 am

If I highlight a word or a syllable and then do a Frequency Analysis, would the highest peak be the fundamental frequency?

SteveG · Location: United Kingdom

Not necessarily, but it might be... it really isn't that simple!

sync · Posted - Fri Feb 07, 2003 9:40 am

SteveG · Location: United Kingdom

You could look at the tools site that I mentioned...

And guitars tend to produce nice, clear harmonically structured notes that decay sensibly, without changing their sound significantly. It's not too hard to track sounds like this - even if they are produced by a human voice - this is how Antares autotune works. But the speaking voice has a mixture of voiced and unvoiced sounds, with a complex, changing harmonic structure, and this is where analysis gets difficult...

sync · Posted - Fri Feb 07, 2003 2:38 pm