Is it possible to compare two waves files?

[Alexandre Garcia]

Hello Guys,

I need a help 

I'd like to know if it is possible to reach a task with your components, let me explain:

 

I have two waves files and I'd like to compare them.

 

The two waves files are:

 

1) the voice a native american speaker that record the word "House"

2) the voice a non native american speaker that also record the word "House"

 

I would like to compare the two wave file and get in percent the correct accents or precise

 

Is it possible to do it? If so, what should I use to do that?

 

Thanks a lot in advanced

Alexandre / Brazil

[Vulcanos]

Hi,
I am working on comparing two sample, however not exactly what you need. I compare my voice word "one" with another record of my voice sound "One". I think it is possible to compare small waves like few words.

There is editors demo called Filters test. I have reworked it I bit to display signal of two files:

You can create your own comparing algorithm.

My comparing program will be finished at saturday so if you send me two samples of the same word, I can compare them. It's challenging. Note: there must not be any disturbing sounds on background like clicking of the mouse, otherwise the samples will not match.

[Vulcanos]

I couldnt attack zip file with program so I uploaded it here
https://uloz.to/!NgEZvyZxrSnI/compare-dft-zip
to download click on
Stáhnout pomalu
and then fill in CAPTCHA code and click on Stáhnout.

On Sunday, 19 July 2015 16:31:15 UTC+2, Alexandre Garcia wrote:

[Alexandre Garcia]

Hello 

Wow, thanks a lot for sending an email. Although it has been such a long time, I haven't had any solution for this so far.

I am sending you three kinds of word record wav. You can compare them.

Thanks a lot

Regards 
Alex

--
--
You are recieving this message because you are subscribed to "NewAC users" Google group.
To post message to this group send it to
newac...@googlegroups.com
To unsubscribe from the group send a mail to: newac-users...@googlegroups.com
Additional variants are available at the groups page http://groups.google.com/group/newac-users?hl=en
---
You received this message because you are subscribed to a topic in the Google Groups "NewAC users" group.
To unsubscribe from this topic, visit https://groups.google.com/d/topic/newac-users/gkFJTfEbzWs/unsubscribe.
To unsubscribe from this group and all its topics, send an email to newac-users...@googlegroups.com.
For more options, visit https://groups.google.com/d/optout.

[Vulcanos]

Hello Alexandre,
thanks for your samples. During weekend I worked on algorithms related to the comparation. It is not finished yet and I will need more time. There are more ways how to compare the signals. In this moment I can compare your files and I can say, they are comparable. However, when we compare the files using DFT there is the rule, the shorter the sound, the more precision. When I compare Throw files, one says [fru] and the second says [frau + noise of the microfon], so the sounds are not quite similar. When I check the curves so file with noise has smaller values of the frequencies in the 2/6 of the DFT range. But seems like 5/6 of the signal is similar. Similar with the "Thanks". If I would evaluate it as "similar" is dipends on some level of precision which would be needed to be set. But I don't have the evaluating algorithm finished.

It is also possible to compare using LogPower instead LogMagnitude. That is different curve. I think here is more visible the audio value difference. The curves of Throw are just a bit similar, while the Thanks file are very different.

Note that the first Thanks contans long "s" sound, which is very expresive. The second file has just short "s", which makes the sounds so different.

The excelent file, is quite different. When you record voice to compare you need clear background, but I heare that the second file contains click of the mouse. The first file also is very carefull about expressing syllables. When he says ex-ce-lent I can see 3 waves. When you say ex-ce-lent, the "ce" is cca 1/2 smaller then the next wave. That is problem. If you would have algorithm to compare syllables you need to separate the syllabels. You need to set the level of valume and the middle wave may not be recognised because it is not loud enough.

I would need to work on valume detection, however I did not find out how to do it. There is demo called audio cutter using array of values, but I found that buf[i] contains some strange values (array of byte value 128), which is like there is no sound in the file.

My conclusion is that you can compare the file using DFT, but you need to care about saying the syllables carefully and check the graphical depiction of the valume of the wave.

[Vulcanos]

I wanted to say:

It is also possible to compare using LogPower instead LogMagnitude. That is different curve. I think here is more visible the audio volume difference.

v := (FA.LogPower[0, i] + 1)/-6*350;
r := Round(v);

[Vulcanos]

Sorry I was wrong:

v := (FA.LogPower[0, i] + 1)/-6*350;

Produces bigger curve, but LogPower and LogMagnitude seems to be the same because

v := (FA.LogMagnitude[0, i] + 1)/-6*350;
and
v := (FA.LogPower[0, i] + 1)/-11*350;
produces the same curve.