Ok, the biggest thing to keep in mind is that NuPIC currently does not contain HTM. The white paper describes it as well as the CLA, but there is no hierarchy within NuPIC today. However, there is an underlying architecture that allows a hierarchy. There are still some pieces missing, like proper temporal pooling, that is required for hierarchical CLAs.
This is something we want to work on in 2014, and I hope to see some progress. You still might find some value passing your data through a single region of the CLA (currently implemented), but this will not find higher-level patterns within the input. --------- Matt Taylor OS Community Flag-Bearer Numenta On Fri, Jan 17, 2014 at 9:49 AM, Scheele, Manuel <[email protected]> wrote: > Hi Matthew, > > > > I would like to see how well HTMs performs in phoneme recognition compared > to deep recurrent neural networks (like these: > http://arxiv.org/abs/1303.5778). > > > > The idea is to use the TIMIT database training set to train the HTM network. > > During training I was thinking of inputting > > · Mel-frequency cepstral coefficient (MFCC) vectors obtained from > cepstral analysis of TIMIT training set audio files (floats) > > · Phoneme transcripts of audio files (text) > > > > Once training is finalized I was thinking of inputting > > · MFCC vectors obtained from cepstral analysis of TIMIT test set > audio files > > > > I would then like to see how well the HTM network recognises the phonemes. > > ( I think the problem is very similar to > http://dx.doi.org/10.1016/j.neucom.2011.10.005) > > > > Regards, > > Manuel > > > > > > From: nupic [mailto:[email protected]] On Behalf Of Matthew > Taylor > Sent: 17 January 2014 14:52 > To: NuPIC general mailing list. > Subject: Re: [nupic-discuss] Speech Recognition using NuPIC > > > > What exactly will you be attempting? What is the speech input, audio or > textual? Is there training data? If you can define your problem, we should > be able to help you more. > > > > Regards, > > > --------- > > Matt Taylor > > OS Community Flag-Bearer > > Numenta > > > > On Fri, Jan 17, 2014 at 3:58 AM, Scheele, Manuel > <[email protected]> wrote: > > Hi all, > > I am currently working on a speech recognition application based on NuPIC. I > have done my initial reading and have been following the mailing lists, but > now that I have started to think about the implementation I have encountered > a problem: I don't know where to start. I lack an understanding of the > workflow. How do I go about defining the model, setting up the network, etc. > I couldn't find anything on the wiki pages. Could someone please help me out > and give me a few hints. (I have looked through the examples and NLP > projects on the wiki pages. However, there is so much code everywhere that I > am not sure where to start looking). > > Thanks in advance. > Manuel > > > _______________________________________________ > nupic mailing list > [email protected] > http://lists.numenta.org/mailman/listinfo/nupic_lists.numenta.org > > > > > _______________________________________________ > nupic mailing list > [email protected] > http://lists.numenta.org/mailman/listinfo/nupic_lists.numenta.org > _______________________________________________ nupic mailing list [email protected] http://lists.numenta.org/mailman/listinfo/nupic_lists.numenta.org
