Customizing model

[JRMeyer]

>>> sibtainraza158
[June 21, 2019, 6:54pm]

Hi Guys, I am transcribing files having duration of two minutes. I have
customized the language model using my own vocabulary. I have few things
to ask. slash
What should be the order of my language model (kenlm)? slash
My audio files contain words like zero,one,two, three...nine
,twenty,thirty,forty ,fifty etc and a ,b ,c,d,e,f,...z etc. slash
e.g transcript

Your account number is four w nine seven ten five p eight h zero zero
zero f b you have transferred five thousand nine hundred sixty eight
dollars and ninety cents your remaining balance is four thousand eighty
six dollars and forty cents.

Should I change the alpha and beta parameters ? slash
I have tried to find the optimal values of alpha and beta but it is a
trade of between certain words. Not working as i expected. slash
What should be length of sentences in my vocab file? slash
Thanks Alot

[This is an archived TTS discussion thread from discourse.mozilla.org/t/customizing-model]

[JRMeyer]

>>> lissyx
[June 24, 2019, 2:27pm]



> I have customized the language model using my own vocabulary.

How did you customized it, adding your own content or just changing the
content with your own ?



> My audio files contain words like zero,one,two, three...nine
> ,twenty,thirty,forty ,fifty etc and a ,b ,c,d,e,f,...z etc. slash
> e.g transcript
>
> Your account number is four w nine seven ten five p eight h zero zero
> zero f b you have transferred five thousand nine hundred sixty eight
> dollars and ninety cents your remaining balance is four thousand
> eighty six dollars and forty cents.

So according to your example transcript, the audio content is not just
numbers.



> Should I change the alpha and beta parameters ?

That's possible.



> I have tried to find the optimal values of alpha and beta but it is a
> trade of between certain words. Not working as i expected.

It's always a trade-off. May I ask the accent of your speakers ? it's
not impossible this has a play in your issue here.

[Archived Post]

[JRMeyer]

>>> sibtainraza158
[June 24, 2019, 3:19pm]

Thank you for Replying . slash
Accent is American and it works fine with the pre trained model. I made
my own language model (no previous vocabulary). For me only account
number and amount matters. But there are one or two mistakes in account
number and in amount as well. slash
I added alphabet characters digits pronounciation(one,two,thirty,forty
etc) and amounts like hundred, thousand etc. slash
Did i make any wrong assumptions for language model?

[Archived Post]

[JRMeyer]

>>> sibtainraza158
[June 24, 2019, 3:20pm]

and yes account number contains letters as well.

[Archived Post]

[JRMeyer]

>>> lissyx
[June 24, 2019, 3:38pm]



> Did i make any wrong assumptions for language model?

No, but it's possible the current dataset does not contain enough
numbers, and thus accoustic recognition is not perfect enough to help
the LM ?

[Archived Post]

[JRMeyer]

>>> sibtainraza158
[June 24, 2019, 3:42pm]

should i finetune on my data? slash
If yes I do not have smaller files of 5 to 7 seconds. slash
At maximum i have 100 audio files slash
are 3 epochs enough to finetune ?

[Archived Post]

[JRMeyer]

>>> lissyx
[June 25, 2019, 6:21am]

I can't tell for you, you need to explore your dataset ... But 100 audio
files with 5 to 7 seconds seems like not a lot, I'm not sure you can get
anything ...

[Archived Post]