Talk:Sound pattern matching using Fast Fourier Transform in Windows Phone
Galazzo - Improvements
Hi All,as promised I made important improvements an matching algorithm. I'm talking about the MFCC approach. Tests made gave great and better results and once again I was impressed of Lumia 800 performances as adding MFCC's floating point computation to FFT the device responce was great no significant delay to the global process.
00:21, 1 October 2012 (EEST)
Hamishwillee - Awesome!
MFCC looks like a great improvement for handling tone, volume or environmental noise differences between sample and the recorded sound. Great job. FYI I posted a note on facebook.As an aside, when you refer to numbers, its best not to punctuate them, and if you do, to use the US/English punctuation. So for example ten thousand is punctuated as 10,000 in English - but I'd use 10000 unless you've got a lot of zeros. For decimals we use the point, and you don't have much choice but to do so - e.g. 10.1 for ten point 1.
04:19, 3 October 2012 (EEST)
Toddmain - Trouble implementing thisThis is really great, but I'm having trouble implementing it. Is there anyway to add a small sample project that detects for a few words (like "up", "down", "go straight", etc)? This would be nice to see something a little more concrete. Also, the app referenced "Shooter Assistant" doesn't seem to be available any more. Where can I download that to try it out?
talk) 23:20, 21 August 2013 (EEST)(
can you provide more detail on your troubles ? Without I'm not able to help you. Consider that in the 90% of cases people wrong as they use a too small microphone buffer. I used a so small buffer ( 100ms ) as for me is important to catch whatever sound in the smaller time window, but for you I think is better to have a bigger one.
My suggestion is to take the longest word you want to use and use the shorter time ( power of two ) able to wrap that word speaching.
Shooter Assistant was a prototype and now an App not for free to be published so I can't publish the code of course, but I'm available to help you.Sebastiano
talk) 01:51, 10 September 2013 (EEST)(
Toddmain - Toddmain - Details on implementing
Hi Sebastiano, I have tried to create a C# project using the above, but obviously I'm doing something wrong. Is it possible to create a very simple project with just one word to recognize (like "jump") in order to test and extend it? Again, the main issue I'm having here is that I can't seem to find a way to implement your algorithms above in a sample project. For example: 1) the DSP.cs file you referenced does not match what you use above, 2) Do I use the microphone_BufferReady in the "Spectrum()" section, or the one in the "Compute()" section or the one in the "How to use MFCC" section? These are just a few of the issues I'm having in implementing this.
In particular, I'm looking for local sound recognition (not speech recognition). In Windows Phone 8, that would be the equivalent of the "confidence level". I'm wondering if this solution above fits that need as the samples you provided are all just one word. Can they be sentences, like "I want to jump on the bed" that are recorded and then compared against using your algorithms. I'm looking to create a "pronunciation improvement" program that will tell a user A) if they got it right, and B) how close are they to the pre-recorded pronunciation (set of sounds). This is for speech therapy and I need WP7 platform as well, so WP8 speech capabilities just won't be sufficient.Any help you could provide would be helpful. If you'd like, you can reach me via email at "to dd p ma in @ g m ail .co m" (put those together for the email address). Thanks Sebastiano.
talk) 20:49, 25 October 2013 (EEST)(