Wav2vec2 makes it possible for low resource languages to build high quality acoustic models using only unlabelled audio. Finetuning this with a couple of hours of labelled data gives you a pretty good start to ASR.
What sort of interface would be required to leverage a wav2vec model (finetuned or not) with MFA instead of Kaldi?
mdatsev
commented
1 year ago
Wav2vec2 makes it possible for low resource languages to build high quality acoustic models using only unlabelled audio. Finetuning this with a couple of hours of labelled data gives you a pretty good start to ASR.
What sort of interface would be required to leverage a wav2vec model (finetuned or not) with MFA instead of Kaldi?