EURASIP Journal on Advances in Signal Processing

Table 4 Recognition accuracy (%) achieved by MSE ^(o) and MSE for Aurora-2 clean-condition training task, where MSE ^(o) is MSE employing nearly error-free VAD results (MSE in the oracle condition)

From: Enhancing the magnitude spectrum of speech features for robust speech recognition

SNR	Set A		Set B		Set C		Average
SNR	MSE^(o)	MSE	MSE^(o)	MSE	MSE^(o)	MSE	MSE^(o)	MSE
20 dB	97.70	97.31	98.19	97.55	96.91	96.19	97.74	97.18
15 dB	95.76	94.47	96.78	95.54	93.60	90.98	95.73	94.20
10 dB	89.36	87.25	92.76	89.54	82.86	78.67	89.42	86.45
5 dB	72.43	69.97	79.23	73.47	60.47	54.53	72.76	68.28
0 dB	41.44	39.80	52.10	43.33	30.73	26.73	43.56	38.60
average	79.33	77.76	83.81	79.89	72.91	69.42	79.84	76.94

Back to article page