Sub-band temporal modulation envelopes and their normalization for automatic speech recognition in reverberant environments

In this paper, we propose a new temporal filter design method based on minimum KL divergence criterion for robust recognition of noisy and reverberant speech. The main idea is to optimize the filter parameters by minimizing the KL divergence of two distributions, of which one is the feature distribution in the test environment, and another is the feature distribution represented by the acoustic model. The minimization of the KL divergence reduces the mismatch between the acoustic model and the test data. Experimental results on Aurora-5 task shows that the new filter design outperforms other filter design methods significantly in noisy and reverberant test conditions. In addition, the proposed filtering of feature trajectories is shown to be complementary to linear transformation of feature vectors, which is popular in feature processing.

show abstract

“…reverberation. Temporal filters are reported to be effective in dealing with both additive noise [8,10,11] and reverberation [12,13,14].…”

Section: Introductionmentioning

confidence: 99%

Temporal filter design by minimum KL divergence criterion for robust speech recognition

Xiao

Chng

2013

2013 IEEE International Conference on Acoustics, Speech and Signal Processing

View full text Add to dashboard Cite

show abstract

“…For better alignment or matching, normalization of the subband temporal modulation envelopes may be used [17]. The main factors responsible for the stagnation in the fields of speech recognition are environmental noise, channel distortion, and speaker variability [10], [11], [12]. Let us consider simple speech recognition as a pattern matching problem.…”

Section: Speech Recognitionmentioning

confidence: 99%

Effect of MFCC Based Features for Speech Signal Alignments

Singh¹,

Khanna²,

Lehana³

2013

IJNLC

View full text Add to dashboard Cite

show abstract

“…Although, a lot of effort has been put for improving the speech recognition, until now the performances of such systems are unappealing in real world tasks [8]. The main factors responsible for the stagnation in the fields of speech recognition are environmental noise, channel distortion, and speaker variability [7] [9] [10]. For alignment, segmentation is used and it is a technique by means of which the boundaries between words, syllables, or phonemes in spoken languages are identified [1].…”

Section: Introductionmentioning

confidence: 99%

“…The most common method for training of the speaker recognition system is hidden Markov model (HMM) and its latest variant is (HMM-GMM) [6]. For better alignment or matching, normalization of the subband temporal modulation envelopes may be used [7]. Although, a lot of effort has been put for improving the speech recognition, until now the performances of such systems are unappealing in real world tasks [8].…”

Section: Introductionmentioning

confidence: 99%