I want to use semi supervised training using neural network on the basis of likelihood and posterior probabilities
If any one has this neural network could he pleased help me and we can do the research together and perhaps we can publish it in one good journal or conference
If you would like to refer to this comment somewhere else in this project, copy and paste the following link:
I used the Scale score S and the total score T and the acoustic score A and the total Language score L as feed to self organizing map neural network SOM after normalizing their values but the neural network does not converges
Then after investigating I read that these values are scaled by the S factor and the real values for the score could be obtained by summing up the T , A and L values to the scaling value the S value.
So I sum
A+S
T + S
L +S
And feed only the three new numbers to the self organizing map
Also the neural network did not converges
Does any one have any clue.
If you would like to refer to this comment somewhere else in this project, copy and paste the following link:
My idea regarding using semi supervised training came from a paper I read and Mr. Arthur coauthor of this fantastic paper
“investigating on ensemble based semi-supervised acoustic model training” http://www.cs.cmu.edu/~archan/papers/eurospeech2005_ensemble.pdf
my question is how can I get this four parameters
The inputs consist of four features representing both language model and acoustic
model information:
1. LM-backoff-mode
2. Utterance-levelposterior-probability
3. Word-level-posterior-probability
4. Frame-level-posterior-probability
could you please help me
thank you in advance.
If you would like to refer to this comment somewhere else in this project, copy and paste the following link:
Hi Hiyassat,
Your question requires a detail reply. Would you give me another 3 days before sending my reply? I was unfortunately caught up with some paper writing at this point.
Arthur
If you would like to refer to this comment somewhere else in this project, copy and paste the following link:
I want to use semi supervised training using neural network on the basis of likelihood and posterior probabilities
If any one has this neural network could he pleased help me and we can do the research together and perhaps we can publish it in one good journal or conference
I used the Scale score S and the total score T and the acoustic score A and the total Language score L as feed to self organizing map neural network SOM after normalizing their values but the neural network does not converges
Then after investigating I read that these values are scaled by the S factor and the real values for the score could be obtained by summing up the T , A and L values to the scaling value the S value.
So I sum
A+S
T + S
L +S
And feed only the three new numbers to the self organizing map
Also the neural network did not converges
Does any one have any clue.
Hi,
You could modify sphinx 3's s3_am routine to get the score and run decoding.
Of course, you still need to train the NN. I see as the big problem and perhaps you are the first one try to do it in Sphinx. :-)
Arthur
My idea regarding using semi supervised training came from a paper I read and Mr. Arthur coauthor of this fantastic paper
“investigating on ensemble based semi-supervised acoustic model training”
http://www.cs.cmu.edu/~archan/papers/eurospeech2005_ensemble.pdf
my question is how can I get this four parameters
The inputs consist of four features representing both language model and acoustic
model information:
1. LM-backoff-mode
2. Utterance-levelposterior-probability
3. Word-level-posterior-probability
4. Frame-level-posterior-probability
could you please help me
thank you in advance.
Dear Arthur
Please help?
Hi Hiyassat,
Your question requires a detail reply. Would you give me another 3 days before sending my reply? I was unfortunately caught up with some paper writing at this point.
Arthur
Any help please!!!