Creating An Acoustic Model

[ElMokhtar Ahmed]

Hi everyone,
i want to create my personal acoustic model but have some questions that i should find their answers before .
who is better record each word in the dictionary or a whole sentence in a wav file ?
also the volume of wav files how it should be high or low ?
can i train in noisy places to adapt my model ? ,if yes what is the tag of noise(and other tags if possible)
can i have the acoustic databases of the official acoustic models ?

thank you !

[Nickolay V. Shmyrev]

who is better record each word in the dictionary or a whole sentence in a wav file ?

If you intent to recognize words in the dictionary, record them. Otherwise if you intent to use this model for dictation, record sentences

also the volume of wav files how it should be high or low ?

Volume should be average, you only need to care to avoid cclipping

can i train in noisy places to adapt my model ? ,if yes what is the tag of noise(and other tags if possible)

You can use any symbol or word to designate noise. Many databases use [noise] to designate noise in recordings.

can i have the acoustic databases of the official acoustic models ?

It is not clear what do you mean by that.

[ElMokhtar Ahmed]

thanks
I mean that if there is a way to get the acoustic datas (the wav files) that officials acoustic models are trained with, or can i adapt the existing acoustic models , if yes which is the best solution, create my own model or adapt an existing one.
I have a small vocab (<100 words) and i want to get nearly to 100% accuracy.

[Nickolay V. Shmyrev]

is a way to get the acoustic datas (the wav files) that officials acoustic models are trained with

Yes, you can find TEDLIUM corpus in google, it represents current acoustic model well.

or can i adapt the existing acoustic models

Yes, you can

create my own model or adapt an existing one.

It is better to create your own model.

[ElMokhtar Ahmed]

thanks so much