Male voices are recognized quite well, even in a noisy environment. (from 1-30 KeywordThreshold confidently)
Female voices recognize very bad.(use 1-50 KeywordThreshold)
What other settings do I need to use?
Other issues:
What is the permissible range of sensitivity?
How to use the parameter noise? How does he work?
Distant speech recognition requires a very good microphone array or specialized acoustic model. Our current models are for close talking microphones.
Our latest model cmusphinx-ru-5.2 is more accurate than the model you are using. You should use it with 16khz.
remove_noise parameter depends on the model and enables noise removal. cmusphinx-ru-5.2 must be used with enabled noise removal.
Keyword spotting does not use pronunciation variants from the dictionary, only the first one. If you want to use variants, you need to include all them in keyword list.
If you would like to refer to this comment somewhere else in this project, copy and paste the following link:
Thanks Nicholay. I will try.
If I use a keyword list, do I need to specify a ".setKeywordThreshold"?
What is the allowable range of threshold in keyword list? ( /1e-5/ ... /1e-200/)
The list and the dictionary are correct? ttps://bitbucket.org/KirsanovAndreii/raw_file/downloads/
Do i need to fill the file "noisedict"?
What does the parameter do ".setBoolean("-allphone_ci", true)"? You sometimes recommend it?
Do I need to use this?:
.setFloat("-lw", 2.0)
.setFloat("-beam", 1e-20)
.setFloat("-pbeam", 1e-20)
.setFloat("-vad_threshold", 3.0)
How file ru.lm is used? I deleted it from the directory assets (Android)
Last edit: Kirsanov Andreii 2018-02-01
If you would like to refer to this comment somewhere else in this project, copy and paste the following link:
Nikolay hello.
I use a PocketSphinx to recognize one key phrase(two words) (redistribution code from Demo Android)
Decoder settings:
Male voices are recognized quite well, even in a noisy environment. (from 1-30 KeywordThreshold confidently)
Female voices recognize very bad.(use 1-50 KeywordThreshold)
What other settings do I need to use?
Other issues:
What is the permissible range of sensitivity?
How to use the parameter noise? How does he work?
raw file and dictionary: https://bitbucket.org/KirsanovAndreii/raw_file/downloads/
Last edit: Kirsanov Andreii 2018-01-30
Distant speech recognition requires a very good microphone array or specialized acoustic model. Our current models are for close talking microphones.
Our latest model cmusphinx-ru-5.2 is more accurate than the model you are using. You should use it with 16khz.
remove_noiseparameter depends on the model and enables noise removal. cmusphinx-ru-5.2 must be used with enabled noise removal.Keyword spotting does not use pronunciation variants from the dictionary, only the first one. If you want to use variants, you need to include all them in keyword list.
Thanks Nicholay. I will try.
If I use a keyword list, do I need to specify a ".setKeywordThreshold"?
What is the allowable range of threshold in keyword list? ( /1e-5/ ... /1e-200/)
The list and the dictionary are correct? ttps://bitbucket.org/KirsanovAndreii/raw_file/downloads/
Do i need to fill the file "noisedict"?
What does the parameter do ".setBoolean("-allphone_ci", true)"? You sometimes recommend it?
Do I need to use this?:
.setFloat("-lw", 2.0)
.setFloat("-beam", 1e-20)
.setFloat("-pbeam", 1e-20)
.setFloat("-vad_threshold", 3.0)
How file ru.lm is used? I deleted it from the directory assets (Android)
Last edit: Kirsanov Andreii 2018-02-01