my apologise for being misleading in my question. Perhaps I am still confusing myself over the difference of Acoustic Model and Language Model. But I am slowly getting there. Thanks for the excellent guidance. Yes, I do mean to train acoustic model.
Most of what I am training focus on multiple words in a sentence. However, for every 200 sentences transcription, I'll probably have about 38 single word (or digit) transcription. So it will focus heavily on the sentences. But it will still be necessary to learn the digits.
Robbie, does that mean having number in the sentence tag isn't a problem?
Thanks,
Suwandy
If you would like to refer to this comment somewhere else in this project, copy and paste the following link:
No worries about the confusion. To be sure, I've never trained my own acoustic model. That said, as long as the digit is in the dictionary you should be fine.
Robbie
If you would like to refer to this comment somewhere else in this project, copy and paste the following link:
Hello,
I never trained a accustic mode but i wrote some transcripts for speaker adaption. I think you can only use the words standing in your dictionary.There should be no problem if you use only one word. I suggest you to dont use the <S> </S> because you can timealign the transcript file with the align program later... and it adds the <S> </S>.
Its enough to add the filename without ending.
shio
If you would like to refer to this comment somewhere else in this project, copy and paste the following link:
I think you might be referring to how to train an acoustic model? Because for language models, you don't need to map to any files. Furthermore, if you language model consists of one word per sentence, then you are going to end up with a lousy language model (use a grammar instead).
If you are referring to an acoustic model, although I've never done this step, I don't see any problem with having only one word per sentence.
Regards,
Robbie
If you would like to refer to this comment somewhere else in this project, copy and paste the following link:
my apologise for being misleading in my question. Yes I do mean to train acoustic model.
Most of what I am training focus on multiple words in a sentence. However, for every 200 sentences transcription, I'll probably have about 38 single word (or digit) transcription. So it will focus heavily on the sentences. But it will still be necessary to learn the digits.
Robbie, does that mean having number in the sentence tag isn't a problem?
Thanks,
Suwandy
If you would like to refer to this comment somewhere else in this project, copy and paste the following link:
Hi,
I am trying to train my own language model. and I am at the point of trying to make a transcription file for my training files.
I know that there must be a one-one mapping of transcriptions to training files, and that the syntax is normally
[code]
<S> SENTENCE </S> (name of file transcription correspond to)
[end code]
Will this work if I use it on a digit? Or one worded word? Say...
[code]
<S> 1 </S> (name of file)
<S> heat </S> (name of file)
[end code]?
Or is there a different syntax I have to use? Help is very much appreciated
Thanks a lot,
Suwandy
Hi,
my apologise for being misleading in my question. Perhaps I am still confusing myself over the difference of Acoustic Model and Language Model. But I am slowly getting there. Thanks for the excellent guidance. Yes, I do mean to train acoustic model.
Most of what I am training focus on multiple words in a sentence. However, for every 200 sentences transcription, I'll probably have about 38 single word (or digit) transcription. So it will focus heavily on the sentences. But it will still be necessary to learn the digits.
Robbie, does that mean having number in the sentence tag isn't a problem?
Thanks,
Suwandy
No worries about the confusion. To be sure, I've never trained my own acoustic model. That said, as long as the digit is in the dictionary you should be fine.
Robbie
Hello,
I never trained a accustic mode but i wrote some transcripts for speaker adaption. I think you can only use the words standing in your dictionary.There should be no problem if you use only one word. I suggest you to dont use the <S> </S> because you can timealign the transcript file with the align program later... and it adds the <S> </S>.
Its enough to add the filename without ending.
shio
I think you might be referring to how to train an acoustic model? Because for language models, you don't need to map to any files. Furthermore, if you language model consists of one word per sentence, then you are going to end up with a lousy language model (use a grammar instead).
If you are referring to an acoustic model, although I've never done this step, I don't see any problem with having only one word per sentence.
Regards,
Robbie
Hi,
my apologise for being misleading in my question. Yes I do mean to train acoustic model.
Most of what I am training focus on multiple words in a sentence. However, for every 200 sentences transcription, I'll probably have about 38 single word (or digit) transcription. So it will focus heavily on the sentences. But it will still be necessary to learn the digits.
Robbie, does that mean having number in the sentence tag isn't a problem?
Thanks,
Suwandy