Skip to main content

Table 5 Comparison of STOI scores (%) for various algorithms under different SNRs using different types of noise

From: A speech enhancement algorithm based on a non-negative hidden Markov model and Kullback-Leibler divergence

Test type

Method

− 5

0

5

10

Unseen 10 types of noise

Noisy

76.97 (\(\pm \,{ 1.45}\))

84.24 (\(\pm \, 0.96 )\)

90.07 (\(\pm \, 0.68\))

94.16 (\(\pm \,0.49)\)

 

Log-MMSE

75.86 (\(\pm \, 1.54\))

83.67 (\(\pm \, 1.01\))

89.72 (\(\pm \, 0.70\))

93.85 (\(\pm \, 0.48\))

 

OMLSA

75.88 (\(\pm \, 1.52\))

83.58 (\(\pm \, 1.01\))

89.51 (\(\pm \, 0.72\))

93.62 (\(\pm \, 0.55\))

 

Temporal-NMF

77.21 (\(\pm \, 1.45\))

84.39 (\(\pm \, 0.96\))

90.15 (\(\pm \,0.68\))

94.19 (\(\pm \, 0.49\))

 

SLF-NMF

69.35 (\(\pm \, 1.78\))

77.01 (\(\pm \, 1.28\))

82.11 (\(\pm \, 1.09\))

85.72 (\(\pm \, 0.94\))

 

CNMF

77.12 (\(\pm \, 1.51\))

83.02 (\(\pm \, 1.13\))

86.01 (\(\pm \, 1.02\))

89.44 (\(\pm \, 0.91\))

 

NMF-HMM

78.58 (\(\pm \, 1.34\))

84.76 (\(\pm \,0.84\))

88.39 (\(\pm \, 0.58\))

90.88 (\(\pm \, 0.43\))

 

DNS baseline

81.84 (\(\pm \, 1.36\))

86.91 (\(\pm \,1.09\))

91.44 (\(\pm \, 0.75\))

94.67 (\(\pm \, 0.55\))

Unseen office noise

Noisy

49.91 (\(\pm \,{1.33}\))

61.03 (\(\pm \, 1.40 )\)

72.80 (\(\pm \, 1.27\))

82.57 (\(\pm \,1.05)\)

 

Log-MMSE

46.46 (\(\pm \, 1.50\))

58.75 (\(\pm \, 1.57\))

71.09 (\(\pm \, 1.40\))

81.31 (\(\pm \, 1.15\))

 

OMLSA

44.97 (\(\pm \,1.52\))

58.14 (\(\pm \,1.63\))

71.52 (\(\pm \, 1.44\))

82.29 (\(\pm \, 1.14\))

 

Temporal-NMF

49.70 (\(\pm \, 1.46\))

61.79 (\(\pm \, 1.47\))

73.48 (\(\pm \, 1.29\))

83.05 (\(\pm \, 1.05\))

 

SLF-NMF

48.92 (\(\pm \, 1.58\))

60.84 (\(\pm \, 1.54\))

70.95 (\(\pm \, 1.35\))

79.21 (\(\pm \, 1.12\))

 

CNMF

48.43 (\(\pm \, 1.47\))

60.97 (\(\pm \, 1.46\))

71.45 (\(\pm \, 1.12\))

80.03 (\(\pm \, 0.97\))

 

NMF-HMM

50.06 (\(\pm \, 1.72\))

63.02 (\(\pm \, 1.61\))

74.56 (\(\pm \, 1.32\))

82.55 (\(\pm \, 0.88\))

 

DNS baseline

54.22 (\(\pm \, 1.49\))

66.46 (\(\pm \,1.01\))

77.58 (\(\pm \, 0.89\))

86.18 (\(\pm \, 0.50\))

textbfSeen 25 types of noise

Noisy

73.65 (\(\pm \,{ 0.82}\))

81.36 (\(\pm \,1.03 )\)

87.64 (\(\pm \,0.84\))

92.48 (\(\pm \,0.60)\)

 

Log-MMSE

71.96 (\(\pm \, 1.40\))

80.13 (\(\pm \, 1.20\))

87.04 (\(\pm \, 0.94\))

92.08 (\(\pm \,0.68\))

 

OMLSA

73.86 (\(\pm \, 1.38\))

81.58 (\(\pm \,1.18\))

87.90 (\(\pm \, 0.91\))

92.45 (\(\pm \, 0.66\))

 

Temporal-NMF

75.76 (\(\pm \, 1.34\))

83.22 (\(\pm \,1.09\))

89.03 (\(\pm \, 0.88\))

93.46 (\(\pm \, 0.58\))

 

SLF-NMF

65.76 (\(\pm \, 1.58\))

73.49 (\(\pm \, 1.33\))

79.06 (\(\pm \,1.18\))

83.14 (\(\pm \,1.04\))

 

CNMF

76.23 (\(\pm \, 1.38\))

84.12 (\(\pm \, 1.11\))

89.55 (\(\pm \,0.97\))

91.06 (\(\pm \,0.62\))

 

NMF-HMM

81.49 (\(\pm \, 1.66\))

87.02 (\(\pm \,1.35\))

90.28 (\(\pm \,0.77\))

91.84 (\(\pm \, 0.51\))

 

DNS baseline

81.95 (\(\pm \, 1.76\))

87.34 (\(\pm \,1.15\))

91.53 (\(\pm \,0.75\))

94.77 (\(\pm \, 0.53\))