I'd like to analyze the perplexity of a hypothesis in a transcription by calculating the perplexity of that hypothesis against the language model that I'm using, with the aim of having a more realistic estimate of how accurate the transcription will be, does it make sense or the confidence-score function in pocket-sphinx already does this for confidence estimation?
In other words, if I calculate the perplexity on the language model, will I get useful information that I can use for the confidence estimation in pocket-sphinx, or the confidence function in pocket-sphinx already does this in its algorithm?
(please let me know if the question isn't clear)
If you would like to refer to this comment somewhere else in this project, copy and paste the following link:
I'd like to analyze the perplexity of a hypothesis in a transcription by calculating the perplexity of that hypothesis against the language model that I'm using, with the aim of having a more realistic estimate of how accurate the transcription will be, does it make sense or the confidence-score function in pocket-sphinx already does this for confidence estimation?
In other words, if I calculate the perplexity on the language model, will I get useful information that I can use for the confidence estimation in pocket-sphinx, or the confidence function in pocket-sphinx already does this in its algorithm?
(please let me know if the question isn't clear)
No
Last edit: Nickolay V. Shmyrev 2015-03-05
Confidence estimation and perplexity estimation of the language model are not related.