Click here to register.

Italian

Flat
Re: Italian acoustic model
User: kmaclean
Date: 1/27/2016 12:22 pm
Views: 29
Rating: 0

>La mia domanda è: esiste un modo (che io mi son perso) per costruire il file

>con la pronuncia delle parole, a partire da queste, oppure devo farlo per

>forza a mano?

Google translate: My question is, is there a way (that I am lost) to build the file with the pronunciation of words, from these, or do I have to do that by hand?

You might look at a text to speech engine (like MaryTTS) with support for Italian, and then create a script to generate the words you are missing (i.e. using grapheme-to-phoneme conversion).

 

Re: Italian acoustic model
User: gabriele.z
Date: 1/28/2016 3:15 am
Views: 18
Rating: 0

Thank you very much for your answer and sorry if I wrote in Italian.

I have some other questions:

if I want to build a model in order to make sphinx understand, let's say, 30 commands, what kind of audio data should I record?

Generic data, or should I have to record sentences similar to the ones I want my system to understand?

Generally speaking, how do I choose the data to be recorded?

Moreover, if the hour of recording necessary to build a model are reached on voxforge, it will be the voxforge team to build the acoustic model?

 

Regards,

Gabriele

Re: Italian acoustic model
User: nsh
Date: 1/28/2016 3:26 am
Views: 27
Rating: 0

> if I want to build a model in order to make sphinx understand, let's say, 30 commands, what kind of audio data should I record?

You need to record those commands in realistic environment. I.e. you need to record audio as you would already be using the system with the microphone you will use in the system.
That should cover 70% of your data.
You also need to record 30% of generic sentences to allow system to generalize over unseen speech. For generic sentences you can get book texts or news.
For 30 commands you need about 5-10 hours of data recorded from 20-100 speakers.

> Moreover, if the hour of recording necessary to build a model are reached on voxforge, it will be the voxforge team to build the acoustic model?

 

Yes, I can help you with that.

 

Re: Italian acoustic model
User: gabriele.z
Date: 1/28/2016 4:06 am
Views: 18
Rating: 0

Thank you very much.

5-10 hours in total, or per speaker?

Regards,

Gabriele

Re: Italian acoustic model
User: nsh
Date: 1/28/2016 4:16 am
Views: 22
Rating: 0

Total

Re: Italian acoustic model
User: p!Zz@b0y
Date: 1/28/2016 4:26 am
Views: 83
Rating: 0

Hi Gabriele,

if you want we can have a skype call this weekend and I can try to retrieve everythink I did for my thesys job and share it with you, and maybe we can also write down some steps, so that we can share them with others.

Ciao! :-D

Re: Italian acoustic model
User: gabriele.z
Date: 1/28/2016 4:38 am
Views: 363
Rating: 0

@nsh Thank you very much.

 

@p!Zz@b0y It would be great! I will send you my contact in a private message. Grazie! :D

 

Re: Italian acoustic model
User: Cris
Date: 5/14/2016 3:41 pm
Views: 859
Rating: 0

Hi is possibile to stare allo with me?

we clan eventually work in cooperation?

Mg email is cristianofumagalli1977@gmail.com

PreviousNext