Deep Learning for Prosodic Features Detection in Human Speech

Wandb project

SSPNet Vocalization Corpus

Model	Laughter F-score	Filler F-score
wav2vec 2.0 + BLSTM	0.5454	0.6921
BCResNet	0.4871	0.667
BCResNet + BLSTM	0.5019	0.6733
wav2vec 2.0 + BCResNet + BLSTM	0.5515	0.6658

LibriStutter

Model	Prolongation F-score	Phrase repetition F-score	Sound repetition F-score	Word repetition F-score
wav2vec 2.0 + BLSTM	0.6783	0.5598	0.616	0.5025
BCResNet + BLSTM	0.8078	0.6013	0.7275	0.4799
wav2vec 2.0 + BCResNet + BLSTM	0.6167	0.6635	0.7124	0.442