The one issue I can think of is that the CLA isn't set up to predict multiple fields right now. You could either make that modification (some people have talked about it on here) or use separate models for each field which might be prohibitively expensive computationally (since you would probably want to include all fields in each model).
On Wed, Aug 21, 2013 at 8:57 AM, Scott Purdy <[email protected]> wrote: > Seems like a reasonable plan but I don't know what to expect. I would be > really interested to see the results. > > > On Tue, Aug 20, 2013 at 8:56 PM, Ralph Dratman <[email protected]>wrote: > >> Thank you, Scott. I watched the Hackathon video about the midi hack. >> Since that midi testing behaved as expected, I want to go further and input >> actual audio. At each time step, the training input stream would consist of >> a vector of integers representing the energy seen in each of a set of audio >> filter frequency bins covering the voice band of the audio spectrum. >> >> I want to train NuPIC on such a vector stream, derived from voice sound >> files, and get Nupic to predict a stream of those energy vectors. The >> result would be output as an audio stream. >> >> I have plans to extend the above, depending on the results. >> >> Does that plan seem reasonable to you, using NuPIC in its present form? >> >> Ralph >> >> >> On Tue, Aug 20, 2013 at 1:20 AM, Scott Purdy <[email protected]> wrote: >> >>> There is currently no audio-specific encoder in NuPIC and I am not aware >>> of one existing. For our hackathon there was a group that converted a song >>> into midi eighth notes and fed the integer values into the scalar encoder. >>> This simplified the audio enough for the CLA to learn fairly easily but >>> didn't differentiate between held notes (quarter or half notes) verses >>> repeated eighth notes and was restricted to a single exact note per step. >>> On Aug 19, 2013 9:33 PM, "Ralph Dratman" <[email protected]> >>> wrote: >>> >>>> Is there an existing encoder for audio, aimed (eventuallyat an >>>> eventual live speech input stream? EverybodI want other live ambient sound >>>> to be included as well, >>>> >>>> _______________________________________________ >>>> 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 >> >> >
_______________________________________________ nupic mailing list [email protected] http://lists.numenta.org/mailman/listinfo/nupic_lists.numenta.org
