I have been using a small (JSGF-based) grammar with good effect. However, my only problem is that noises (such as breathing, coughing, etc.) unfortunately show up as words.
I wanted to record myself making such noises, and then add them to the dictionary. Is that necessary, or are there any runtime settings that could be used instead?
Furthermore, if this is indeed a good idea, how would these non-phoneme based "words" be added to the dictionary? WOuld they be also added to the JSGF file?
Thanks!
Colin
If you would like to refer to this comment somewhere else in this project, copy and paste the following link:
I have been using a small (JSGF-based) grammar with good effect. However, my only problem is that noises (such as breathing, coughing, etc.) unfortunately show up as words.
You can collect and provide test data to get help on this issue.
I wanted to record myself making such noises, and then add them to the dictionary. Is that necessary, or are there any runtime settings that could be used instead?
It is not easy to just add sounds to the model. Our models have specific sounds for breath and those should be detected with some reliability. You might only need tuning on the parameters like fillprob which control fillers insertion probability
Furthermore, if this is indeed a good idea, how would these non-phoneme based "words" be added to the dictionary? WOuld they be also added to the JSGF file?
There is noisedict in acoustic model already containing fillers from the model. Fillers are usually inserted after every word in the grammar. If you want some different behavior, you can add fillers to the grammar itself but then you need to remove them from noisedict.
If you would like to refer to this comment somewhere else in this project, copy and paste the following link:
Hi,
I have been using a small (JSGF-based) grammar with good effect. However, my only problem is that noises (such as breathing, coughing, etc.) unfortunately show up as words.
I wanted to record myself making such noises, and then add them to the dictionary. Is that necessary, or are there any runtime settings that could be used instead?
Furthermore, if this is indeed a good idea, how would these non-phoneme based "words" be added to the dictionary? WOuld they be also added to the JSGF file?
Thanks!
Colin
You can collect and provide test data to get help on this issue.
It is not easy to just add sounds to the model. Our models have specific sounds for breath and those should be detected with some reliability. You might only need tuning on the parameters like
fillprobwhich control fillers insertion probabilityThere is noisedict in acoustic model already containing fillers from the model. Fillers are usually inserted after every word in the grammar. If you want some different behavior, you can add fillers to the grammar itself but then you need to remove them from noisedict.