Hi Harel, Zach, This has been brought up before, but there were simply no man/hours available to actually do the job (I am extremely busy with my PhD and other things). There is nothing in MARF itself that prevents you from accomplishing the task. MARF itself is a collection and a testing platform for various algorithms to be used in the pattern recognition pipeline in order to select the best suited for the task and use it afterward.
However, to do what you want to do you'd have to put a lot of load on the application itself which uses MARF. The application would contain the business logic necessary to do the search and call on MARF when needed. Like SpeakerIdentApp is a sample MARF application to do a task of speaker biometrics, you have to build an equivalent for pattern search of the desired fingerprint. That means of course you'd have to have the reference music available somewhere. Then go through the stream in chunks and compute either the smallest distance or highest cosine similarity score after certain threshold of each chunk to your fingerprint, when found there is a strong liklihood of the match. Another way, is to convert both the song and fingerprint into the frequency domain from the time domain, and see of the fingerprints' frequencies appear in the song -- in this latter case you don't need to do stream pattern analysis, but can just accumulate the spectrum of the entire song e.g. using FFT. The more details are needed, the larger the FFT window should be (e.g. 1024 or more instead of default 512). You may also need to increase the quality of the sampling frequency in the WAVLoader (the default is 8 kHz, which is an analog phone line quality) to something significantly higher. -s On Fri, 3 Jun 2011, Zachary Kosarik wrote: > Date: Fri, 3 Jun 2011 16:55:02 -0400 > > Hey, > I have never personally tested anything like that, however MARF uses bit > pattern recognition, which when audio is broken down that is all it is. So > in theory if you had a database of songs and they're bit patterns you could > easily use marf to get the patterns and then match them. If you are trying > to do this live i.e. record on mic and then compare the results are going to > vary greatly depending on background noise and scilence gaps. But to answer > your question yes I believe it can be done, but I'm not entirely sure how to > go about it. > > Hope I helped a little, > Zach > > On Sat, May 28, 2011 at 8:37 AM, •HΛ®əŁ° < > [email protected]> wrote: > > > > > > > Hi, > > Is it possible to use MARF as an 'Audio Search Algorithm' a bit like Shazam > > & SoundHound for creating a searchable audio fingerprint for a song? > > > > Best regards, > > > > •HΛ®əŁ° -- Serguei A. Mokhov, PhD Candidate | /~\ The ASCII Computer Science and Software Engineering & | \ / Ribbon Campaign Concordia Institute for Information Systems Engineering | X Against HTML Concordia University, Montreal, Quebec, Canada | / \ Email! ------------------------------------------------------------------------------ EditLive Enterprise is the world's most technically advanced content authoring tool. Experience the power of Track Changes, Inline Image Editing and ensure content is compliant with Accessibility Checking. http://p.sf.net/sfu/ephox-dev2dev _______________________________________________ marf-devel mailing list [email protected] https://lists.sourceforge.net/lists/listinfo/marf-devel
