Learning Temporal Dependencies in Data Using a DBN-BLSTM · HackerPapers discussions

Learning Temporal Dependencies in Data Using a DBN-BLSTM — mentioned by kastnerkyle

Sun, 09 Aug 2015 01:51:45 GMT

This is really cool stuff - the network structure reminds me a lot of Graves' MDRNN[1] and Grid LSTM[2], as well as some work I helped with (ReNet [3]) I wonder if the structure over frequency/time is too "regular" - in general for sound the frequency correlation and the time correlation are on wildly different scales. Also if you are looking to go farther you might reconsider adding NADE or RBM [4] on top, or latent variables in the hiddens[5][6] to add more stochasticity. There was some alternate work by Kratarth Goel extending RNN-RBM to LSTM and DBN, it might give you some ideas to look…

Paper and discussion history. A mention is not an endorsement.

Learning Temporal Dependencies in Data Using a DBN-BLSTM — mentioned by kastnerkyle

Fri, 22 May 2015 00:04:07 GMT

The work of Nicolas Boulanger-Lewandowski was extensively focused on this topic, see his work [1]. He wrote a Theano deep learning tutorial on this topic [2], and several people (Kratarth Goel) [3][4] have advanced the work to use LSTM and deep belief networks. For a brief while RNN-NADE made an appearance as well, though I do not know of an open source implementation There are also a few of us who are working on more advanced versions of this model for speech synthesis, versus operating on the MIDI sequence. Stay tuned in the near future! I can say from experience that some of the samples…

Paper and discussion history. A mention is not an endorsement.