Speaker adaptation based on regularized speaker-dependent eigenphone matrix estimation

EURASIP Journal on Audio, Speech, and Music Processing

Table 2 Average tonal syllable recognition rate (%) after eigenphone-based speaker adaptation using lasso

λ ₁	Number of adaptation sentences
	1	2	4	6	8	10
10	52.25	56.04	58.06	59.06	60.24	61.27
	(0.61)	(0.43)	(0.23)	(0.16)	(0.12)	(0.04)
20	53.88	56.55	58.54	59.36	60.24	60.83
	(0.83)	(0.63)	(0.42)	(0.33)	(0.26)	(0.23)
30	53.63	55.96	57.70	59.19	60.05	60.81
	(0.91)	(0.74)	(0.54)	(0.44)	(0.37)	(0.34)
40	53.82	55.18	57.30	58.90	59.75	60.49
	(0.95)	(0.820)	(0.65)	(0.61)	(0.49)	(0.42)

The number of eigenphones (N) was fixed to 100. λ₂=λ₃=0, and λ₁ was varied between 10 and 40. The average overall sparsity is shown in parentheses.