Working on new language , mainly African language.

[shekharmeena2896]

the dataset for the African languages is fairly limited and fine-tuning on hundreds of dataset is a huge task. let me know any one want to discuss this further

for shona language of Zimbabwe , there is data in google fleurs dataset which about 12 hours of shona data. will that be enough for it?

[anurag12-webster]

well im also looking to train for a new language

[shekharmeena2896]

which language? have u got any leads how to implement it?

[anurag12-webster]

which language? have u got any leads how to implement it?

i want to add the brazilian-portuguese (br-pt) language i have curated dataset on my side but for pretraining authors suggest 1k hours least let's see what can be done i have only ~180hrs

[amuvarma13]

which language? have u got any leads how to implement it?

i want to add the brazilian-portuguese (br-pt) language i have curated dataset on my side but for pretraining authors suggest 1k hours least let's see what can be done i have only ~180hrs

180 hours is probably good especially if you use the es-it pretrained model as a base - if you want specific voices you will need atleast 50-100 rows of a specific speaker.

Also you are probably better off creating a separate discussion as I suspect there will be other people who are interested in brazilian portugese.

[dfernandesbsolus]

which language? have u got any leads how to implement it?

i want to add the brazilian-portuguese (br-pt) language i have curated dataset on my side but for pretraining authors suggest 1k hours least let's see what can be done i have only ~180hrs

Are you done? :D

[amuvarma13]

the dataset for the African languages is fairly limited and fine-tuning on hundreds of dataset is a huge task. let me know any one want to discuss this further

for shona language of Zimbabwe , there is data in google fleurs dataset which about 12 hours of shona data. will that be enough for it?

You can start by training the base english model and seeing what kind of results you get - sometimes the results have been good - if they arent good you may have to look into finding a larger dataset for pretraining.