I’m a bit confuesed about the vocabulary.txt.
Is it just every line a new transcript and does this need to match any order or only every transcript of the wave files in no special order or just a bunch of text to get the statistics of letters?
Greetings
1 Like
The vocabulary is used for the language model and has nothing to do with the wavs. Usually, take as much written material in your language as you can. And it is always about statistics of words, not letters
See this repo for an example how to do that even though model generation is still 0.6 based so you can see what type of data to use. For newer versions >0.7 you build a scorer as described in the docs.
1 Like