A Triplet Multimodel Transfer Learning Network for Speech Disorder Screening of Parkinson’s Disease

Table 6

Performance of deep models on the IPVS dataset.

Vowel					Vowel					Vowel
Methods	Precision (%)	Recall (%)	F1 score (%)	Acc (%)	Methods	Precision (%)	Recall (%)	F1 score (%)	Acc (%)	Methods	Precision (%)	Recall (%)	F1 score (%)	Acc (%)
BiLSTM	99.24	99.94	99.59	99.60	BiLSTM	99.39	99.83	99.61	99.47	BiLSTM	99.41	99.68	99.55	99.50
LSTM	99.36	99.94	99.65	99.68	LSTM	99.39	100.00	99.69	99.58	LSTM	99.39	99.76	99.57	99.52
BiLSTM (Attn)	99.28	99.89	99.59	99.60	BiLSTM (Attn)	99.63	99.93	99.78	99.70	BiLSTM (Attn)	99.24	99.58	99.41	99.35
HMM	96.34	96.30	96.32	96.50	HMM	97.63	95.56	96.58	95.37	HMM	96.94	93.42	95.14	94.70
CNN	98.82	99.64	99.23	99.26	CNN	99.01	99.76	99.38	99.17	CNN	98.68	99.59	99.13	99.02
Vowel					Vowel					TmmNet
Methods	Precision (%)	Recall (%)	F1 score (%)	Acc (%)	Methods	Precision (%)	Recall (%)	F1 score (%)	Acc (%)	Methods	Precision (%)	Recall (%)	F1 score (%)	Acc (%)
BiLSTM	99.15	99.89	99.52	99.42	BiLSTM	98.91	99.73	99.32	99.22	Vowel	99.89	99.89	99.89	99.90
LSTM	99.23	99.89	99.56	99.48	LSTM	99.21	99.73	99.47	99.40	Vowel	99.86	99.83	99.84	99.79
BiLSTM (Attn)	98.69	99.74	99.21	99.06	BiLSTM (Attn)	99.11	99.89	99.50	99.43	Vowel	99.72	99.61	99.66	99.63
CNN	99.11	99.50	99.30	99.15	CNN	98.70	99.68	99.19	99.07	Vowel	99.95	99.85	99.90	99.87
HMM	97.77	96.23	96.99	96.48	HMM	97.63	97.11	97.37	97.00	Vowel	99.89	99.94	99.92	99.91

The bold values in Table 6 represents the highest results among all the compared methods.