I am trying to build a model from scratch for a few sentences and a language that is not supported by Common Voice.
If i want to train a model with 10 sentences only from scratch, how much data (in hours) do i need for a sufficient result ?
I tried building the model with one voice and had no result, with multiple voices and still no result.
I think your description is close to what @elpimous_robot did for his robot TUTORIAL : How I trained a specific french model to control my robot
That being said, could you please describe what you intend to build ?
A simple Albanian Speech-to-text application that detects my commands and displays the commands to text.
Simple as that.
Just testing the abilities of DeepSpeech
You could try to experiment with transfer learning from the english model, but you will require a few hours, at least.
If you intend to have a limited scope of commands, no free-text, you will get much better result if you also build a language model from your commands / words.
To get more data, you might want to try and find other people to work on Albanian for Common Voice: https://voice.mozilla.org/
Thank you for the fast replies.