wav2vec2-base-it-latin
This model is a fine-tuned version of wav2vec2-base-it-voxpopuli
The dataset used is the poetaexmachina-mp3-recitations, all of the 2-series texts (vergil) and every tenth 1-series text (words from Poeta Ex Machina's database of words with scansions).
It achieves the following results on the evaluation set:
- Loss: 0.1943
- WER: 0.398