Estimation of Phoneme-Specific HMM Topologies for the Automatic Recognition of Dysarthric Speech

Table 4

Recognition accuracies obtained with different training schemes for the baseline SD ASR systems across all dysarthric speakers.

Speaker	Training scheme 1
	SD Bakis-1	SD Bakis-2	SD Ergodic	Average scheme 1
FB	77.73	72.93	76.42	75.69
MH	73.80	69.43	73.36	72.20
BB	75.69	63.30	69.27	69.42
LL	69.20	67.86	66.96	68.01
JF	60.89	48.00	55.56	54.82
RL	50.67	48.44	43.56	47.56
RK	44.30	15.35	25.44	28.36
BK	13.27	6.64	26.99	15.63
BV	48.64	50.91	43.64	47.73
SC	44.55	19.55	27.73	30.61
Speaker	Training scheme 2
	SD Bakis-1	SD Bakis-2	SD Ergodic	Average scheme 2
FB	72.49	63.76	65.94	67.40
MH	66.38	58.95	57.64	60.99
BB	69.72	54.13	62.39	62.08
LL	64.73	57.14	62.05	61.31
JF	49.78	43.11	52.00	48.30
RL	51.11	42.67	44.89	46.22
RK	38.16	39.04	39.47	38.89
BK	31.42	18.58	36.28	28.76
BV	55.91	53.18	47.73	52.27
SC	44.09	43.18	48.18	45.15