icsi-speech-tools : Message: Re: Why are there positive posteriori probabilities when I use MLP feature in LV

icsi-speech-tools

Yahoo! Groups Tips

Did you know...

Message search is now enhanced, find messages faster. Take it for a spin.

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.

Messages

Messages Help

Advanced

Why are there positive posteriori probabilities when I use MLP featu

Message List

Reply | Forward

Message #159 of 162 |

Re: Why are there positive posteriori probabilities when I use MLP feature in LV

--- In icsi-speech-tools@yahoogroups.com, "qingqing.zhang"
<qingqing.zhang@...> wrote:
>
> 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
> larger according to your experience?

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) are typical. If it's really 40% accuracy (60%
error), there's almost certainly a bug.

Adam

Mon Nov 3, 2008 7:23 pm

Show Message Option

"Adam Janin" <janin@...>

ajanin

Offline

Send Email

Forward

Message #159 of 162 |

Expand Messages	Author	Sort by Date
When I use MLP as my HMM features, the POSITIVE posteriori probabilities always occurred. The POSITIVE posteriori probabilities will break the decoding rule...	qingqing.zhang	Oct 30, 2008 7:26 am
... 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...	Adam Janin ajanin	Oct 30, 2008 4:58 pm
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...	qingqing.zhang	Nov 3, 2008 3:26 am
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 ...	qingqing.zhang	Nov 3, 2008 7:37 am
... 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)...	Adam Janin ajanin	Nov 3, 2008 7:23 pm
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...	Adam Janin ajanin	Nov 3, 2008 7:21 pm

< Prev Topic | Next Topic >

Advanced

Yahoo! My Yahoo! Mail
	Sign In New User? Sign Up
		Tech - Groups - Help

Yahoo! Groups Tips

Did you know...

Best of Y! Groups

Messages