Reverberant speech recognition exploiting clarity index estimation

Parada, Pablo Peso; Sharma, Dushyant; Naylor, Patrick A.; Waterschoot, Toon van

doi:10.1186/s13634-015-0237-7

EURASIP Journal on Advances in Signal Processing

Table 5 WER (%) obtained with the reverberant part of the evaluation data set

From: Reverberant speech recognition exploiting clarity index estimation

	Sim.				Real
	R1		R2		R3		R1
	Near	Far	Near	Far	Near	Far	Near	Far
Clean-cond.	17.91	25.67	42.85	83.70	54.22	89.08	90.19	88.15
Multi-cond.	20.60	21.09	23.70	38.72	28.08	44.86	58.45	55.44
NIRA-CART
Clean&Multi cond.	18.67	21.59	23.83	38.72	28.15	44.86	58.45	55.44
C50FV	20.62	20.74	23.12	39.14	28.19	46.61	58.19	55.50
C50HLDA	18.38	19.99	21.34	37.03	27.44	42.55	55.92	54.09
MS3	18.08	19.82	21.92	35.94	27.35	40.25	55.64	53.51
MS3+C50HLDA	17.16	19.40	20.60	32.67	25.37	36.32	53.53	51.49
MS5	16.32	18.52	20.49	36.34	25.85	40.62	55.35	53.41
MS5+C50HLDA	16.44	17.93	19.91	32.51	24.45	37.62	53.66	51.65
MS8	16.72	19.32	20.79	34.02	26.50	39.31	53.24	53.11
MS8+C50HLDA	15.72	18.26	19.79	30.76	24.16	35.85	52.06	50.03
MS11	16.50	18.99	21.14	34.75	25.85	39.09	57.87	55.37
MS11+C50HLDA	16.10	17.79	19.95	31.58	23.90	36.21	54.77	51.49
MS14	16.50	19.06	21.37	34.64	24.83	39.50	55.61	54.66
MS14+C50HLDA	15.88	17.93	19.73	30.78	22.39	35.86	52.67	51.96
MS18	16.25	19.13	21.19	34.96	24.94	39.40	56.50	55.20
MS18+C50HLDA	15.64	18.23	19.79	31.15	22.83	36.15	53.78	52.46
NIRA-BLSTM
Clean&Multi cond.	18.01	21.08	23.70	38.72	28.08	44.86	58.45	55.44
C50FV	20.52	20.50	23.07	39.25	28.07	46.58	58.70	55.13
C50HLDA	18.40	19.69	21.31	37.16	27.18	42.83	55.35	53.78
MS3	16.93	18.87	21.79	35.99	27.25	39.98	54.52	52.36
MS3+C50HLDA	15.96	18.18	20.05	32.38	25.00	35.93	53.37	51.76
MS5	16.01	18.25	20.50	36.32	26.02	40.42	54.90	52.84
MS5+C50HLDA	16.16	17.15	19.50	32.67	24.07	37.25	53.40	50.74
MS8	16.01	18.50	20.13	34.12	25.39	39.00	53.66	52.63
MS8+C50HLDA	15.66	17.01	19.50	30.65	23.22	35.68	52.22	49.70
MS11	15.88	17.94	20.07	34.75	24.96	38.87	55.76	53.71
MS11+C50HLDA	14.79	16.79	18.85	31.50	22.85	36.09	53.27	50.57
MS14	15.66	17.42	20.08	34.23	24.25	38.95	55.41	53.21
MS14+C50HLDA	14.35	17.40	18.48	30.95	22.61	35.32	52.48	51.82
MS18	15.06	17.52	19.86	34.14	24.83	39.14	57.11	54.90
MS18+C50HLDA	14.81	16.66	18.93	31.02	22.49	35.93	54.07	51.22

The first two rows correspond to the baseline methods, and the remainder are the methods proposed in this work. R1, R2 and R3 represent room numbers 1, 2 and 3, respectively. Best performance results in each column are shown in italics

Back to article page