Hi my name is Gaby Im mexican and I need help, I read the manuals and tutorials for the sphinx but I need more help
I read the messegaes write Omar, and I need a similar system, but my system is only for two words: ADELANTE ATRAS. I will train a continous acustic model for the mexican spanish for application in the Sphinx 4, but need more help for the acustic model, for example:
I dont know what is the sintax for the transcrip file, if only I want to recognize 2 words, and I dont have idea for the Sound files.This means that: What need talk in a sound file and how many sounds file I need?.
I think that in the Sound file many persons need record many ocurrences of the two words, because the system need be speaker independent, for example:
ADELANTE ATRAS ADELANTE ATRAS ADELANTE ATRAS
How long need be each sound file?
I only need this system, and think that you make before a similar system, can you help me?
What is the structure of the phone list for this type of system, because is not dependent of context, I think that the only two phones in the phone list must be ADELANTE ATRAS (plus SIL), Am I right?
Can you show me an example for this type of system?. I think that the TDIGITs is more complex and I only need to recognize two words
I read the Tiny doc for the scripts_pl and I am not secure if need run all the script, I read in this forum and found that only need run 00 and 02, Is that rigth? and The Sphinx 4 work fine only with this type of acustic model?
Im using linux, and I record 50 utterances (tree people record this), the structure of each utterance (transcription file) is:
Hi my name is Gaby Im mexican and I need help, I read the manuals and tutorials for the sphinx but I need more help
I read the messegaes write Omar, and I need a similar system, but my system is only for two words: ADELANTE ATRAS. I will train a continous acustic model for the mexican spanish for application in the Sphinx 4, but need more help for the acustic model, for example:
I dont know what is the sintax for the transcrip file, if only I want to recognize 2 words, and I dont have idea for the Sound files.This means that: What need talk in a sound file and how many sounds file I need?.
I think that in the Sound file many persons need record many ocurrences of the two words, because the system need be speaker independent, for example:
ADELANTE ATRAS ADELANTE ATRAS ADELANTE ATRAS
How long need be each sound file?
I only need this system, and think that you make before a similar system, can you help me?
What is the structure of the phone list for this type of system, because is not dependent of context, I think that the only two phones in the phone list must be ADELANTE ATRAS (plus SIL), Am I right?
Can you show me an example for this type of system?. I think that the TDIGITs is more complex and I only need to recognize two words
I read the Tiny doc for the scripts_pl and I am not secure if need run all the script, I read in this forum and found that only need run 00 and 02, Is that rigth? and The Sphinx 4 work fine only with this type of acustic model?
Im using linux, and I record 50 utterances (tree people record this), the structure of each utterance (transcription file) is:
<s> ADELANTE ATRAS ADELANTE ATRAS ADELANTE ATRAS ADELANTE ATRAS ADELANTE ATRAS <s> (archivo01)
...
...
<s> ADELANTE ATRAS ADELANTE ATRAS ADELANTE ATRAS ADELANTE ATRAS ADELANTE ATRAS <s> (archivo50)
The phone list is:
ADELANTE
ATRAR
SIL
The main dictionary is
ADELANTE ADELANTE
ATRAS ATRAS
The fill dictionario is
<s> SIL
</sil> SIL
SIL SIL
The control file is (The scripy 00 shows a error in the sintay of this file, I think need more things Is rigth?):
archivo01
archivo02
....
archivo50
The sound files were record in RAW and were convert with the wave2feat aplication to cepstra vector
Thanks a lot, Gaby
Please can anybody help me?
I need this information, please only tell me if my ideas for a Acustic models are right.
Gaby