Application of Perceptual Filtering Models to Noisy Speech Signals Enhancement

Table 2

Recognition rate for the proposed perceptual generalized spectral subtraction method based on the gammatone and the gammachirp filter banks decomposition (GSS_GT and GSS_GC) compared with the spectral subtraction rule (GSS).

Babble noise
Methods/SNR	0	5	10	15
GSS	16.88%	31.25%	44.19%	64.96%
GSS_GT
MCRA	30.02%	39.76%	51.56%	80.98%
MCRA2	28.12%	36.88%	47.69%	64.25%
IMCRA	30.68%	45.77%	63.21%	84.24%
MS	26.63%	33.86%	50.03%	79.58%
GSS_GC
MCRA	33.80%	41.62%	56.23%	89.01%
MCRA2	29.99%	37.80%	48.89%	71.53%
IMCRA	33.89%	56.07%	68.21%	91.22%
MS	27.38%	35.96%	55.83%	83.98%
Car noise
Methods/SNR	0	5	10	15
GSS	47.88%	61.25%	65.19%	73.56%
GSS_GT
MCRA	67.90%	82.83%	77.63%	88.98%
MCRA2	58.61%	77.76%	67.42%	74.09%
IMCRA	70.68%	86.40%	87.87%	90.09%
MS	57.08%	70.85%	75.29%	88.08%
GSS_GC
MCRA	69.09%	87.99%	79.80%	89.97%
MCRA2	68.78%	85.60%	79.04%	80.43%
IMCRA	71.08%	88.86%	89.89%	91.95%
MS	59.18%	72.15%	78.90%	89.01%