This is a Seq2Seq model trained on a karaoke dataset to predict syllables with pitch and timing from song lyrics.
As of writing, the model has only been trained on 1/2 of the full dataset. Expect the quality to improve later.
The Huggingface demo seems to produce outputs with a small sequence length. So what you see on the right will only make a prediction for the first two syllables.
- Downloads last month