[CLAM] request

abe abe.kazemzadeh at gmail.com
Tue Jun 12 14:21:29 PDT 2007

Hi Liviu,

If you need to calculate acoustic similarity, you can use dynamic time 
warping (DTW, http://en.wikipedia.org/wiki/Dynamic_time_warping ), which 
basically windows the speech from the target and test files, extracts 
features for each frame, and aligns them in the best possible way.  The 
algorithm defines a cost for insertion and deletion of frames and the 
similarity of the features, so the overall cost (or the cost normalized 
for the length of the file) provides a good measure for difference.  
There's other refinements to the algorithm.

Clam can do the feature extraction, but I'm not sure if there's the dtw 
algorithm (I'm in the process of learning clam so I'm not an expert), so 
that might be something you'll have to do yourself.  I've used this 
methodology for comparing accents (native speaker vs non-native speaker 
reading the same sentences).   Back then I used HTK for the feature 
extraction and a perl script for the dtw. 

Hope this helps and maybe stimulates more ideas for doing it with clam...

David García Garzón wrote:
> Hi, Liviu.
> Do you mean speech recognition? Speaker recognition? Just sound 
> classification? Which is the concrete use case? I am not sure of what you 
> mean but your statement seems too general to be achieved. Depending on your 
> purpose you might be in a research bleeding edge, specially if you go to the 
> semantic level. 
> CLAM currently has no sound classification system but it has many of the 
> building blocks such systems use. That's better than starting from scratch.
> A new document we added to the wiki [1] gives you an overview on the steps to 
> get introduced into CLAM:
> [1] http://clam.iua.upf.edu/wikis/clam/index.php/Approaching_CLAM
> If you need more help, just ask.
> On Tuesday 12 June 2007 16:44:20 Liviu Macoviciuc wrote:
>>  I am a newbie to CLAM and I don' t understand much.
>> However, I need to write a program that says if 2 audio files are distinct
>> For example, a file might contain a voice saying "I am John", and another
>> file the same voice  or another voice saying "I am Bill"
>> Can anybody help me to get started ?!
>> Best regards,
>> Liviu
> _______________________________________________
> CLAM mailing list
> CLAM at iua.upf.es
> http://www.iua.upf.es/mtg/clam
-------------- next part --------------
A non-text attachment was scrubbed...
Name: abe.kazemzadeh.vcf
Type: text/x-vcard
Size: 267 bytes
Desc: not available
URL: <http://lists.clam-project.org/pipermail/clam-users-clam-project.org/attachments/20070612/884c98aa/attachment-0003.vcf>

More information about the clam-users mailing list