JOURNAL OF INFORMATION, KNOWLEDGE AND RESEARCH IN ELECTRONICS AND COMMUNICATION ENGINEERING
SPEECH ENHANCEMENT IN NOISY ENVIRONMENT WITH HIDDEN MARKOV MODEL
1 KOMAL R. BORISAGAR , 2 DR. G.R.KULKARNI
1 Research Scholar in JJT University, Jhunjhunu-Rajasthan, India.
2 Principal, C.U. Shah Engineering College of Engineering and Technology,
Wadhwan city 363030 Gujarat – India.
,
ISSN: 0975 –6779| NOV 11 TO OCT 12 | VOLUME – 02, ISSUE - 01 Page 1
JOURNAL OF INFORMATION, KNOWLEDGE AND RESEARCH IN ELECTRONICS AND COMMUNICATION ENGINEERING
ISSN: 0975 –6779| NOV 11 TO OCT 12 | VOLUME – 02, ISSUE - 01 Page 1
JOURNAL OF INFORMATION, KNOWLEDGE AND RESEARCH IN ELECTRONICS AND COMMUNICATION ENGINEERING
ABSTRACT:This paper proposes a hidden Markov model (HMM)-based speech enhancement method, aiming at reducing non stationary noise from speech signals. The system is based on the assumption that the speech and the noise are additive and uncorrelated. Cepstral features are used to extract statistical information from both the speech and the noise. A priori statistical information is collected from long training sequences into ergodic hidden Markov models.For the taken speech and noise first state wise model is prepared and using wiener filter at each state noise is removed in straight forward way. In the present concept both speech and noise signal are modelled combined and then given for the noise reduction state wise.
ISSN: 0975 –6779| NOV 11 TO OCT 12 | VOLUME – 02, ISSUE - 01 Page 1
JOURNAL OF INFORMATION, KNOWLEDGE AND RESEARCH IN ELECTRONICS AND COMMUNICATION ENGINEERING
ISSN: 0975 –6779| NOV 11 TO OCT 12 | VOLUME – 02, ISSUE - 01 Page 1
JOURNAL OF INFORMATION, KNOWLEDGE AND RESEARCH IN ELECTRONICS AND COMMUNICATION ENGINEERING
Key Words: Hmm, Ar, Enhancemet, Wiener Filter
ISSN: 0975 –6779| NOV 11 TO OCT 12 | VOLUME – 02, ISSUE - 01 Page 1
JOURNAL OF INFORMATION, KNOWLEDGE AND RESEARCH IN ELECTRONICS AND COMMUNICATION ENGINEERING
I.Introduction
Hidden Markov models are used for the statistical modeling of non-stationary signal processes such as speech signals, image sequences and time-varying noise. An HMM models the time variations and/or the space variations of the statistics of a random process with a Markovian chain of state-dependent stationary subprocesses. An HMM is essentially a Bayesian finite state process, with a Markovian prior for modelling the transitions between the states, and a set of state probability density functions for modelling the random variations of the signal process within each state. [1]
The implicit postulation in using an HMM for noise is that noise statistics can be modelled by a Markovian chain of N different stationary processes [3]. A stationary noise process can be modelled by a singlestate HMM. For a nonstationary noise, a multi-state HMM can model the time variations of the noise process with a finite number of quasi-stationary states. In general, the number of states required to accurately model the noise depends on the
nonstationary character of the noise. [3]
II.Hidden Markov Model And Speech
The performance of HMMs trained on clean signals deteriorates rapidly in the presence of noise, since noise causes a mismatch between the clean HMMs and the noisy signals [4]. The noise-inducedmismatch can be reduced, either by filtering the noise from the signal or by combining the noise.
Consequently thesignal models in the noisy signal. The model combination method, illustrated in Figure .In this method HMMs of speech are combined with an HMM of noise to form HMMs of noisy speech signals [5]. In the power-spectral domain, the mean vector and the covariance matrix of the noisy speech can be approximated by adding the mean vectors and the covariance matrices of speech and noise models.
Decomposition of State Sequences of Signal and Noise
The HMM-based state decomposition problem can be stated as follows: given a noisy signal and the HMMs of the signal and the noise processes, estimate the underlying states of the signal and the noise [6]. HMM state decomposition can be obtained using the following method:
(1) Given the noisy signal and a set of combined signal and noise models, estimate the maximum-likelihood (ML) combined noisy HMM for the noisy signal. [7]
(2) Obtain the ML state sequence of from the ML combined model [7].
Figure 1 Concept speech and noise stages merging
(3) Extract the signal and noise states from the ML state sequence of the ML combinenoisy signal model. The ML state sequences provide the probability density functions for the signal and noise processes. The ML estimates of the speech and noise pdfs may then be used to obtain a MAP estimate of the speech signal. Alternatively the mean spectral vectors of the speech and noise from the ML state sequences can be used to program a state-dependent [7].
V. Experiment METHODOLOGY, SIMULATION Results And Comments
Noise reduction using HMM requires wide training vector. Used numbers of hidden stages for the utterance of word “one” are three with that number of noise stages are two. A useful way of interpreting and using HMMs is to consider each state of an HMM as a model of a segment of a stochastic process. Each state must have a mechanism to accommodate the random variations in different realizations of the segments that it models. The state transition probabilities provide a mechanism for connection of various states, and for the modelling the variations in the duration and time-scales of the signals in each state. Following steps are carried out in the implementation
- Combine the signal and noise models to form the noisy signal models.
- Given the noisy signal and the set of combined noisy signal models, obtain the ML combined noisy signal model.
- From the ML combined model, obtain the ML state sequence of speech and noise.
- Use the ML estimate of the power spectra of the signal and the noise to program the wiener filter equation .
- Use the state-dependent Wiener filters to filter the signal.
Figure 2 Original Speech Signal
Here above simulation result shows noisy speech in that in the initial part large amount of noise can be detected and it shows utterance of “one”
Figure 3 data vector given to HMM
Here figure 3 shows five state data to create states of HMM. To solve the purpose here utterance “A” is repeated and for forming different state, selection state vector is chosen as five.
Figure 4 Mu vectors
Figure 4 shows different states state vector of mu for creating state of noise. Here air hum is chosen as a noise. So for that value of noise vector to create noise states in hidden markov model here different values of mu are taken.
Figure 5State observation vector given in HMM
With the variation in the noise stages and speech stages values of the mu is changing continuously which is plotted in the mentioned result of figure5.
Figure 6 Cleaned speech signal
Simulation result in figure 6 show status of speech signal after applying noise reduction using HMM. It can be observed that using applied concept within a utterance as well as after and beyond utterance speech signal is cleaned.
Application of HMM for noise reduction gives better performance as can be seen in the simulation results. Wiener filtering at each stage gives reduction in the noise very keenly at higher precision level. Obtain result because of this is considerable.
VI. Conclusion
The proposed speech concept is able to reduce non-stationary noise sources. Using number of stages three for word utterance “one”and noise guassian mixture stages three significant amount of noise can be reduced. In enhancement problems, where speech is degraded by an impulsive noise source, such as a babble noise, the proposed speech enhancer is found to substantially reduce the influence of the noise.
VII. Future Work
The speech recognizer and enhancer is implemented in Matlab and because of that it runs slow. Implementing the speech recognizer in C or assembler will be desire to get a faster execution time. Moreover by changing model structure in view of number of states or mixture more accurate result can be demonstrated.
VIII. References
[1] S. F. Boll, “Suppression of aucostic noise in speech using spectral subtraction,” IEEE Transactions on Acoustics, Speech, and Signal Processing, vol. ASSP-27(2), pp. 113– 120, April 1979.
[2] Y. Ephraim, D. Malah, and B. H. Juang, “On the application of hidden markov models for enhancing noisy speech,” IEEE Transactions on Acoustics, Speech, and Signal Processing, vol. 37, no. 12, pp. 1846–1856, December 1989.
[3] S. J. Young M. J. F. Gales MJF, “Cepstral parameter compensation for hmm recognition in noise. speech communication,” Speech Communication, vol. 12, no. 3, pp. 231–239, July 1993.
[4] Y. Ephraim and M. Rahim, “On second order statistics and linear estimation of cepstral coefficients,” IEEE Transactions on Speech and Audio Processing, vol. 7, no. 2, pp. 162–176, March 1999.
[5] L. R. Rabiner, “A tutorial on hidden markov models and selected applications in speech recognition,” Proceedings of the IEEE, vol. 77, no. 2, pp. 164–171, February 1989.
[6] Mikael Nilsson, Mattias Dahl and Ingvar Claesson, “HMM based speech enhancement applied in non-stationary noise using Cepstral features and log-normal approximation” , Publication in Blekinge Institute of Technology, Department ofTelecommunications and Signal Processing.
[7] Saeed V. Vasaghi, “Advanced Digital Signal Processing and Noise Reduction” , 3rd edition , Wiley , 2006.
ISSN: 0975 –6779| NOV 11 TO OCT 12 | VOLUME – 02, ISSUE - 01 Page 1