... Rejecting frames is no joke! We've recently found that it's a great way to speed up training if you reject them in such a way as to leave a subset of data...
Arlo Faria
arlo@...
Oct 25, 2007 1:27 am
134
... Eric> Or, does someone have some code to munge the weight files Eric> for me? ... Eric> aha. I still have old rap3 weights (from qnstrn) so I can Eric>...
... Arlo> Rejecting frames is no joke! We've recently found that it's Arlo> a great way to speed up training if you reject them in such Arlo> a way as to leave...
... Yeah, that's similar to the kind of thing that Jun Hou and Larry Rabiner were seeing in their TDNN trainings, if I interpret what you're saying to mean...
... Yes, that's the idea in principle. In practice, I actually somewhat over-sample the more frequent classes for performance reasons. ... I randomly sample...
Arlo Faria
arlo@...
Oct 25, 2007 2:08 am
138
Hi Eric, ... Arlo already proposed a good solution. I cooperate with Brno guys on the new shiny Bottle-Neck features :) (hopefully more in ICASSP'08 paper) and...
Hi all, IN SHORT: mlp_bunch_size > 256 does not train properly I have been using for years qnstrn/qnmultitrn with a bunch size of 256, which trains much faster...
... Petr> Hi all, IN SHORT: mlp_bunch_size > 256 does not train Petr> properly Petr> I have been using for years qnstrn/qnmultitrn with a bunch Petr> size of...
(This is a question & answer from the ICSI's new internal FAQ. Since it's on a Wiki which is not accessible without a password, I decided to post it here for...
(This is another question & answer copied from ICSI's new internal FAQ.) What is the difference between ftr1_window_len and window_extent? On Mar 30, 2006, at...
(This is another question & answer copied from ICSI's new internal FAQ.) What feature normalization should be used? It's generally recommended to always use...
(This is another question & answer copied from ICSI's new internal FAQ.) Where can I find examples of how to do Quicknet MLP-based ASR? There is an example as...
(This is another question & answer copied from ICSI's new internal FAQ.) How should the input be randomized when training? This article in the old speech FAQ...
Hi, Within a week or so, we are planning to make pfile_utils version 0.51 available at ftp://ftp.icsi.berkeley.edu/pub/real/davidj/. The planned changes...
Hi, This week, we are planning to make noway version 2.9.3 available at ftp://ftp.icsi.berkeley.edu/pub/real/davidj/. The only change compared to version 2.9.2...
When I executed ./labels2pfile, it printed "can't find package Dpwe_Utilfns while executing "package require Dpwe_Utilfns" (file "./labels2pfile" line 16)" Did...
When I run qnstrn with online CMS, it prints "./qnstrn(QN_InFtrStream_OnlNorm.ftr1_file): WARNING - slow seek to mid-seg (seg/fr 4802/81) with online norm" ...
... If you're computing normalization online and then try to seek to the middle of a feature stream, it has to back up to the start and read all the...
... Does your CV set start at segment 4802 by any chance? If so, that's what causes the message -- when the CV run starts, it seeks to the start of the CV run....
When I use MLP as my HMM features, the POSITIVE posteriori probabilities always occurred. The POSITIVE posteriori probabilities will break the decoding rule...
... I'm not sure I understand. If you're using softmax as the output layer of the MLP, then all the values should be between 0.0 and 1.0. If you're not using...
Dear Adam, I am so sorry that I did not explain the problem clearly. Actually, we found that MLP has the much larger dynamic range than PLP has (In the log...
There is another problem: When we use MLP feature on Mandarin CTS, the Frame Error Rate of training data was up to 60% after NN training. Does this a little ...
From your message, I assume you're using the MLP outputs as features to a GMM (so-called "tandem") rather than using the posteriors directly (so-called...
... I haven't run Mandarin CTS myself, so maybe somebody else on the list can answer in more detail, but for English, high 60s for frame ACCURACY (not error)...
Until now, PLP_MLP feature increased the performance accuracy of Mandarin CTS in our experiment. Now we want to try TRAPs into our system. If I use qnstrn to...
Dear Adam, When I use qnstrn to get TRAP feature, it prints "(QN_InFtrStream_RandWindow.ftr1_file): WARNING - Segment 2292 has only 90 frames, which is too...
Dear Adam, When I use qnstrn to get TRAP feature, it prints "(QN_InFtrStream_RandWindow.ftr1_file): WARNING - Segment 2292 has only 90 frames, which is too...
Having problems with message search?
Fill out this form to ensure your group is one of the first to be migrated to the new message search system.
