Multitask Learning with Local Attention for Tibetan Speech Recognition

Table 4

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

Architecture	Model	Lhasa-Ü-Tsang	Changdu-Kham	Amdo Pastoral
SpeakerID model		67.75	93.13	95.31
WaveNet-CTC with speaker ID	S-S1	68.32	92.85	97.48
	S-S2	71.15	95.23	96.12
Attention (5)-WaveNet-CTC	S-S1	0	0	0
	S-S2	60.64	77.38	85.85
WaveNet-Attention (5)-CTC	S-S1	70.35	92.85	97.48
	S-S2	69.40	100	96.70