Voice Pitch Analyser for Android?

[AmySomething]

Hey, sorry to make the first post of mine a question... but I've been trying to train my voice for quite a while now, and I've only just started listening to recordings and analysing my voice (it sounds more ridiculous than I thought!). I just tried analysing the pitch with two different applications I found on the Android market and I think they must both be wrong, because they were saying the pitch I've been practising on the last few months was 280Hz, and that the highest I could go to (falsetto, I got curious!) 700Hz! Obviously that's quite confusing, so I started whistling and it gave me numbers like 900Hz, the lowest I could whistle said 690Hz...

So... I don't really know how accurate these can be. So, my question is, does anyone know any good software for Android that'll tell me the pitch of my voice? Also, what's the best pitch I should try reaching?

Thanks!

[Seras]

Those pitches sound kinda plausible, though I just tested a whistle now and I could not whistle below 1000hz.
However if accurate you should probably be speaking at a lower pitch than 280hz.

Either way, pitch is just a small part. With a good tutorial I got my pitch sorted in a number of days. Do not focus on it too much. How you sound is much more important than a number on your phone.

[Jumpingcats]

I heard there was an app that tells you if you're in the female voice range but I don't know the name of it, or if it was for iphone or android. But I heard you speak into it and it tells you if your in a male range or female range.

[Misato]

It's hard to write a good voice analyzer.  I did it as part of my Trans related Human and Computer Interaction research project. 


Male


Female

Only does windows.  Blue is male range, purple is andro and pink is female.  I built it as a heat map so you could look at your voice over time as it isn't only the frequency but the range in tones that matter.

Turning sound waves into hertz is the tricky part.  Gotta go through the Fast Fourier Transform and it really helps to understand how the algorithm works.  Even then I was paranoid that I was doing something wrong.  I ended up calibrating mine against a bunch of videos from YouTube of men and women talking.  I also have it so it generates MIDI tones and it was able to identify them correctly.

I only mention it to say, be skeptical of Apps that do this to make sure that the App isn't really intended to be a toy.  It's also tough to do QA on an App like this so I'd advise that a good goal is making you happy, not getting some app to say Male or Female.  That and good luck!

[Elspeth]

It's a bit misleading to look at the base pitch, since for any recording of speech, you will see harmonics far above that level. I confess I'm usually a bit puzzled by this concentration on pitch, when it's one of the least relevant factors when it comes to developing a feminine voice in terms of resonance, style and so on.  I just downloaded (for the first time) one of the first apps I could find (not for Android, but maybe it's been ported there too?) -- not sure I'm helping matters by linking, but here's a listing of various software aimed at speech analysis in general.

The one I downloaded is WASP (Waveforms Annotations Spectrograms and Pitch).

Display sample shows the higher harmonics that make the base pitch somewhat misleading.



Here's another listing of free speech analysis applications, though mainly for Windows again.

[Misato]

Background noise was a problem too.  Another reason for the heat map, you could see what wasn't you. Case in point, the bright spot on the low end of the female image was something outside.

Then there is the matter of usability.  I saw waveform ones but was unsure of how to make sense them.  So, I thought I'd try to find a new visualization.

I get the resonance importance though.  Still, there's something to be said about taking it one thing at a time. 

[Elspeth]

Background noise was a problem too.  Another reason for the heat map, you could see what wasn't you. Case in point, the bright spot on the low end of the female image was something outside.

Then there is the matter of usability.  I saw waveform ones but was unsure of how to make sense them.  So, I thought I'd try to find a new visualization.

I get the resonance importance though.  Still, there's something to be said about taking it one thing at a time. 

Everything you said... I did a lot of video and audio processing advice for YouTube users for a few years, and learned a fair bit of interesting stuff about these things (only a little bit of it was relevant, though, when it comes to feminine voice training).

I understand that the programming challenge is interesting in and of itself, and actually, having spent so much time looking at transcoding problems on YouTube, looking at those waveforms (and especially the wider range spectrographs) for me is kind of interesting... I now want to compare the image and sound (which WASP allows you to do fairly simply, if you have some basic notion of what you're doing, at least) between my current female voice and my old voice to see what I can make of the differences... it might actually help me a bit, but only because I've spent so much time looking at issues in recording, processing defects and so on, going back as far as one of my very first jobs in radio news and advert production.

I tend to feel it's not an especially useful area to concentrate on, though, if you're not already very well informed about recording technology and such... a lot of the recordings many will get, are often coming from very inferior equipment that itself is going to add further levels of complication and potentially useless confusion, when so much of getting the right voice is really going to be accomplished by practice, exercises, and attention to the differences between what we record in our first attempts and comparing those to the female voices we want to come close to replicating.

[Anna++]

It's hard to write a good voice analyzer.  I did it as part of my Trans related Human and Computer Interaction research project. 


Male


Female

Only does windows.  Blue is male range, purple is andro and pink is female.  I built it as a heat map so you could look at your voice over time as it isn't only the frequency but the range in tones that matter.

Turning sound waves into hertz is the tricky part.  Gotta go through the Fast Fourier Transform and it really helps to understand how the algorithm works.  Even then I was paranoid that I was doing something wrong.  I ended up calibrating mine against a bunch of videos from YouTube of men and women talking.  I also have it so it generates MIDI tones and it was able to identify them correctly.

I only mention it to say, be skeptical of Apps that do this to make sure that the App isn't really intended to be a toy.  It's also tough to do QA on an App like this so I'd advise that a good goal is making you happy, not getting some app to say Male or Female.  That and good luck!

This looks really cool!  You should open-source it so interested people can port to other operating systems

[Elspeth]

I gave in. Here's the Soundcloud recording these comparisons are based on.

Female voice.



The voice I'm leaving behind.




Overlaid.



No great surprise here, but it's interesting to note how much more dense the overtones become, starting at the lower frequency.  Kind of self-evident to anyone who knows about the nature of harmonic overtones, but still, the narrowband spectrograph does show a very dramatic difference (and yes, I still need a lot more regular practice... the recording is actually worse that what I did about a month ago, because I have not been practicing nearly as much as I should be).

[Misato]

This looks really cool!  You should open-source it so interested people can port to other operating systems

I've thought about making binaries available as I don't do open source.  I work hard to understand what I'm doing, I ain't about to deprive anyone else the same opportunity.  I hold too much code is witten with crtl-c and ctrl-v these days without the copier understanding what that code actually does.

Plus, I'd like to get the porting practice myself.  I do way too much annoying we development these days.

[Elspeth]

I've been following up my analysis of self-recordings by using WASP to take a look at samples from some of my voice role models: Meryl Streep, Jody Foster, Mary-Louise Parker and Emma Thompson, for starters. 

One thing I've noticed is that while those wider gaps in overtones are present for them as well, in many cases their range (fundamental frequency, as charted by WASP) actually drops quite a bit lower than mine did in some spots...as low as 120-150Hz in some cases, and often not going as high as I've been tending to. The average range is often not much higher than 200Hz. In one recording of Emma Thompson, in fact, she starts around 180 Hz or lower and drops as low as 100 Hz.

Chart:



(Because there's also added music in the background of this example, the spectrogram is a lot "noisier" than it was with my samples, but you can still see the telltale bands that match the spoken words... wish I had an easier way to find just the voice recordings for these... and I expect i will find some in due course.  But the point of this exercise is really to confirm for myself that analysis of frequency will only go so far, and much of it is going in a useless direction when it comes to vocal training... though taking some samples to mimic and compare is in itself probably not a bad idea.  I tend to think, though, that working on modifying my voice to remove the "male" clues and cues is going to be a somewhat advanced exercise. This is giving me some ideas of how to approach the challenge, and how to listen to and for those differences.

So much of getting this right has to do with finding ways to avoid activating those lower vibrations that are usually described as "chest voice" but which you can usually tell you are doing by placing fingers lightly on the voice box and watching out for the perceptible vibration that is common when speaking in a more "male sounding" voice.

[Misato]

One thing I've noticed is that while those wider gaps in overtones are present for them as well, in many cases their range (fundamental frequency, as charted by WASP) actually drops quite a bit lower than mine did in some spots...as low as 120-150Hz in some cases, and often not going as high as I've been tending to. The average range is often not much higher than 200Hz. In one recording of Emma Thompson, in fact, she starts around 180 Hz or lower and drops as low as 100 Hz.

If I follow, I think I noticed something similar when developing my app, and was one reason I didn't trust it totally.  I was finding statements about the normal female frequency being higher than I was getting and my frequency range was more varied as well and going into frequencies I did not expect.

That chest voice is a bugger of a problem though.

This is kinda fun!

[Elspeth]

I know that my level of information on speech dynamics is not nearly expert... where I do feel comfortable (from past physics, math and musical study) is understanding that any tone has harmonic overtones (they're almost impossible to prevent, and we'd sound very strange without them). There might be an issue, in terms of detecting a "fundamental frequency" in that one might have a very low amplitude waveform at some lower frequency, and yet have the overall sound affected by greater richness (and amplitude/loudness) in some of the first few harmonics? Don't quote me on that, as I'm only speculating.

To the degree that I'm trying to get some personal benefit from looking at the waveforms and spectrography, it's by looking at those overall, rather complicated combo of signals in comparison to each other, to try to get a better sense of why my attempts to mimic a more credible feminine voice is or is not working. The fact that my recording shows a higher frequency than most of the samples I've looked at from my role models does tell me that what I really need to work on is learning how to avoid bringing out those "male pattern" resonances, which is not much related to pitch alone.