VoxForge
VoxForge
Hello,
I need to compare some voice speeches of different people to see if they say the same and if they pronounce it correctly. For this I am using Julius but the problem is that it always tries to match the speech with the grammar even if the word it is said isn't in it. I thought that maybe using dictation I could get better results, because I don't need to have an accurate output (it doesn't have to be the same as what it is said on the speech), if the output is the same for two speeches of different people where they say the same and correctly it's enough.
What do you think? Is to possible to say to Julius not to give an output if what was said is not part of the grammar?
I hope you understand what I want to do.
Thank you very much and sorry for my English.
--- (Edited on 4/3/2009 4:46 am [GMT-0500] by Visitor) ---
Hi blizarazu,
>it always tries to match the speech with the grammar even if the word it is
>said isn't in it.
You need to add more words in your grammar. See this thread: One word grammar, always recognized? (lrascao's last post)
Ken
--- (Edited on 4/3/2009 12:45 pm [GMT-0400] by kmaclean) ---
Hi Ken,
I tried doing it like it is said in that thread and it works quite well when it only has to recognise one word, but with a whole sentence in the grammar always fails, only recognises correctly if it has two possible words for each word in the sentence. Anyway the biggest problem for what I need to do is that even if I say the same sentence almost never has the same output.
I've been searching for an English language model for Julius but I couldn't find any. Is there any? I would like to see if I get better results with a language model.
Thank you for your help
--- (Edited on 4/23/2009 6:05 am [GMT-0500] by Visitor) ---
Hi blizarazu,
>searching for an English language model for Julius
Try Keith Vertanen's English Gigaword Language Model.
See this post for more info on creating language models for Julius: How to create reverse 3 gram for julius. Note that reverse n-grams are no longer required for the current version of Julius (from lee.akinobu post on Julius forum):
Old Julius-3.x requires two N-gram models to run recognition: forward 2-gram for the first pass, and backward 3-gram for the second pass. The backward 3-gram should be trained from the same corpus with the normal N-gram, with its word order reversed and with the same cut-off value.
Julius-4 can do recognition with only a forward N-gram or a backward N-gram. Word probabilities for reverse direction on each pass are calculated from the given N-gram by Bayes assumption. Since the second pass produces final output, we recommend using backward N-gram.
Their combinations like older versions are still supported, in that case forward 2-gram in the given forward N-gram will be used on the first pass, and the given backward N-gram will be used on the second pass.
Ken
--- (Edited on 5/3/2009 9:15 pm [GMT-0400] by kmaclean) ---