I'm looking for a tool which prints the phonemes it recognized (maybe
pocketsphinx can do that with API, sphinx3 is too heavy. but i think sphinx3
is able to do that with the option -mode allphone)
The goal is to store and recognize proper nouns with their phonemes. DTW works
very well in this case... but it is completely speaker dependent.
Regards,
Laurent
If you would like to refer to this comment somewhere else in this project, copy and paste the following link:
Voice command with a specific and simple vocabulary. Pocketsphinx works very
well in this case.
But there is a another tricky part i try to have, the user stores his name in
a database with his voice and later an administrator can remove his name from
this database with his voice.
I see 3 possibilties:
DTW: but it s speaker dependent.
Create a dicitonnary only with the alphabet and then spell the name. Once i have the written name i can produce the phoneme transcription with espeak or lia_phon, then i can create a dictionnary with names and phonemes for recogniton. Problem; the recognition of alphabet is not very efficient (pronounication of letters are maybe too much similar),
Advantage: speaker independent and the system can pronounce the name by using
a tts like mbrola.
Maybe a last possibility but i don't know if it is possible easily. The system stores the phonemes of a familly name (for example) from the user's speech. When the administrator speaks for a recognition of a username, the system tries to compare the administrator's phonemes between all previously stored phonemes by the users. This solution could the best one.
Don't know if it s clear???
Regards,
Laurent
If you would like to refer to this comment somewhere else in this project, copy and paste the following link:
In such situations names are usually transribed with g2p code and added both
to the grammar and to the dictionary. It can be done with lia_phon for
example.
If you would like to refer to this comment somewhere else in this project, copy and paste the following link:
Usually all possible names are already stored in the dictionary. At least most
common ones.
As an alternative you can build a grammar consisting of all possible
syllables/subword parts and use it for recognition. Also you can use a grammar
with phones like this one:
\<phones> = aa | b | c | ae | d .... ;
If you would like to refer to this comment somewhere else in this project, copy and paste the following link:
Hi,
I'm looking for a tool which prints the phonemes it recognized (maybe
pocketsphinx can do that with API, sphinx3 is too heavy. but i think sphinx3
is able to do that with the option -mode allphone)
The goal is to store and recognize proper nouns with their phonemes. DTW works
very well in this case... but it is completely speaker dependent.
Regards,
Laurent
Pocketsphinx doesn't support this as a part of the public API. Basically it's
because such recognition is not really reasonable/practical.
The proper usage of pocketsphinx depends on the type of applicatoin you are
going to implement. What exactly do you develop?
Voice command with a specific and simple vocabulary. Pocketsphinx works very
well in this case.
But there is a another tricky part i try to have, the user stores his name in
a database with his voice and later an administrator can remove his name from
this database with his voice.
I see 3 possibilties:
DTW: but it s speaker dependent.
Create a dicitonnary only with the alphabet and then spell the name. Once i have the written name i can produce the phoneme transcription with espeak or lia_phon, then i can create a dictionnary with names and phonemes for recogniton. Problem; the recognition of alphabet is not very efficient (pronounication of letters are maybe too much similar),
Advantage: speaker independent and the system can pronounce the name by using
a tts like mbrola.
Maybe a last possibility but i don't know if it is possible easily. The system stores the phonemes of a familly name (for example) from the user's speech. When the administrator speaks for a recognition of a username, the system tries to compare the administrator's phonemes between all previously stored phonemes by the users. This solution could the best one.
Don't know if it s clear???
Regards,
Laurent
In such situations names are usually transribed with g2p code and added both
to the grammar and to the dictionary. It can be done with lia_phon for
example.
But when the user tells his name how i can store it in phoneme format?
Usually all possible names are already stored in the dictionary. At least most
common ones.
As an alternative you can build a grammar consisting of all possible
syllables/subword parts and use it for recognition. Also you can use a grammar
with phones like this one:
\<phones> = aa | b | c | ae | d .... ;
yes, i will try...
something like:
<name> = series of phones
<phones> = aa | b | ae ... </phones></name>
I don't know the syntax for the grammar for the moment.
keep you inform
thx
laurent