I'm studying this paper - Hazen, Timothy J., Stephanie Seneff, and Joseph Polifroni. "Recognition confidence scoring and its use in speech understanding systems." Computer Speech & Language 16.1 (2002): 49-67.
In that, on page 6, various features for word level confidence scores are given. I don't understand the meaning of "across all acoustic observations" phrase that the author uses many times. For example,
"Mean acoustic score: The mean log-likelihood acoustic score across all acoustic
observations in the word hypothesis"
Does he mean that acoustic score (a log-likelihood ratio), for a word hypothesis, is averaged over all the frames? Suppose a word has 10 frames, does he mean that the all the acoustic scores of all frames are added and then divided by 10?
Thanks.
If you would like to refer to this comment somewhere else in this project, copy and paste the following link:
Thanks. I just needed to confirm if that was what author wanted to say. Frankly I think he should have written an equation rather than a long sentence.
If you would like to refer to this comment somewhere else in this project, copy and paste the following link:
Hi,
I'm studying this paper - Hazen, Timothy J., Stephanie Seneff, and Joseph Polifroni. "Recognition confidence scoring and its use in speech understanding systems." Computer Speech & Language 16.1 (2002): 49-67.
In that, on page 6, various features for word level confidence scores are given. I don't understand the meaning of "across all acoustic observations" phrase that the author uses many times. For example,
"Mean acoustic score: The mean log-likelihood acoustic score across all acoustic
observations in the word hypothesis"
Does he mean that acoustic score (a log-likelihood ratio), for a word hypothesis, is averaged over all the frames? Suppose a word has 10 frames, does he mean that the all the acoustic scores of all frames are added and then divided by 10?
Thanks.
Yes, why not. It's not a good confidence measure though.
Thanks. I just needed to confirm if that was what author wanted to say. Frankly I think he should have written an equation rather than a long sentence.