Multitask Learning with Local Attention for Tibetan Speech Recognition

Table 3

Dialect ID recognition accuracy (%) of two-task models.

Architecture	Model	Lhasa-Ü-Tsang	Changdu-Kham	Amdo Pastoral
DialectID model		97.88	92.24	97.9
WaveNet-CTC with dialect ID	D-S	98.57	95.23	99.6
	S-D	99.01	97.61	99.41
Attention (5)-WaveNet-CTC	D-S	100	89.28	94.52
	S-D	0	0	0
WaveNet-Attention (5)-CTC	D-S	100	98.8	99.41
	S-D	100	94.04	98.06