Fine Tuning a model with new voices

[michalis_p (michalis@indigital.gr)]

Hi I have a well trained set in Greek and now I want to expand the model with more voice tones (female / male / elder / etc).
I have the audio recordings in wav and the transcripts from usage of the current model which I will use as input.
According to the documentation I will have to give train, dev and test csv files.
Do I have to separate the data on dev, train and test sets or I can use only 1 csv for the fine tunning?

Then another question is, if I want to expand the vocabulary or create a new vocabulary (I have a vocabulary and a model for medical use and I want to create a new one for layer usage) I have to create a new scorer or I have to make a new training set?

[lissyx ((slow to reply) [NOT PROVIDING SUPPORT])]

@michalis_p

Do I have to separate the data on dev, train and test sets or I can use only 1 csv for the fine tunning?

no, you still need the three

@michalis_p

if I want to expand the vocabulary or create a new vocabulary (I have a vocabulary and a model for medical use and I want to create a new one for layer usage)

your description is unclear, but i think you just want to create a new scorer here

[michalis_p (michalis@indigital.gr)]

Hi lissyx,
thank you for your reply.
So should I separate the files and spread them test, train and dev?
Or I can put the new wav and text on train and then use the previous dev and test?

So in case I need of a new vocabulary I can only make a new scorer and keep the same model.