In situ Performance analysis for Noise Suppression in Hearing prosthesis

M. S. Godwin Premi, G. Merlin Sheeba, Z. Mary Livinsa, G. Mary Valantina

School of EEE, Sathyabama Institute of Science and Technology, Chennai, India

*Corresponding Author E-mail: msgodwinpremi@gmail.com

ABSTRACT:

Human beings, wearing hearing prosthetics will have hard time in segregating voice of interest from background noise in surrounded noisy environment, such that the intelligibility of the hearing prosthetics has to be improved. In this paper, a variable bandwidth band pass filter is proposed so that the intelligibility of the speech will get enhanced. The proposed algorithm is also analyzed in different noisy environments. This work is implemented in LAB View and Field Programmable Gate Arrays.

KEYWORDS: Binary Mask, Hearing Aid, Noise Reduction, Variable Bandwidth Band Pass Filter

1. INTRODUCTION:

In a noisy background hearing impaired people find difficult to eliminate the unwanted noise from the actual signal required to be heard. The cocktail party problems does not affect the normal human beings but its affects impaired as it has a difficulty in recognizing and eliminating prior high reverberations from noise separations which takes place between auditory system and human brain in a stipulated time^[1].

In recent researches speech enhancements have been improved to greater extent but their intelligibility problem exists^[4]. Even the noise reduction technique in the Cochlear implants through the analysis of wind noise. They took a comparative study on impaired patients and found speech intelligibility decreases with respect to wind velocity^[5].

The overall level of wind noise was greater than the difference in the input of two micro phone policies and audio processing segments (Fig.1) basically uses how human speech bring up accordingly to the uniform wrapped frequency^[6]. The analysis on degraded speech has also been carried out, by weighing each frequency region in the specified band segment^[7]. This is done by the comparison of Ideal Weiner Mask and Ideal Binary Mark algorithms through a single channel hearing prostheses.

A comparative study was done on white noise and the babble noise and the voices were processed by wide frequency compression and spectral subtraction sequences^[8].

A comparative study was done on white noise and the babble noise and the voices were processed by wide frequency compression and spectral subtraction sequences where it explains acoustical and perceptual sounds in a hearing aid system^[9]. Even noise reduction algorithm was in much higher frequency channels use a specific predefined coder and a specific threshold is fixed for SNR value to comparison of 3 different noises which can with an improvement in the sentences^[10].

2. OVERVIEW OF MASKING TECHNIQUE:

Generally the binary mask by a true SNR to a preset threshold value is obtained. But, in speech signal, time and frequency are being less closely associated, background noise is also considered to be less closely associated in the same domain and after that the desired signal can be denoised^[11].

Here, the binary mask to the target with noise ingested signal where the average is being taken for the noise signal in the particular bandwidth is applied. Even later it is shown that the binary masked pattern in speech and noise signal^[12] is used.

The signals are being subjected to spectral analysis and then classification is being done in it, output are classified such as if the band mentions it as ‘0’ the presence of noise in the band as ‘1’ the presence of target with noise in that particular signal. The target with noise and the noise signal are being masked together.

3. SPECTRAL ANALYSIS:

In order to analyse the bandwidth of a signal, spectral analysis is done for both the data base as well as the real-time usage of target with noise and noise signal. To eliminate the noise from the speech signal the average spectra of the noise is calibrated this was done using FFT analysis the time duration is calculated for 5ms and the time to frequency analysis is being done further. Based on this analysis and the frequency in which the speech signal is being available, the specific bandwidth is utilized and these signals are being correlated for signals with noise and without noise.

4. MATERIALS AND METHODS:

Filter analysis in Lab View:

Basically, filtering process brings in a variation for the required content of the signal. Here it is considered that the noise and target with noise filtering involves two segments and they are removing noise and decimation. The desired content can be removed from the raw signal by considering the sampling frequency which plays a prominent role if, greater the frequency component in a particular band determines the sapling rate. Depending upon the bandwidth of the speech signal variable band pass filters are being used here, this increases the intelligibility of the hearing prosthesis.

Application of transform:

Hilbert transform is applied for target with noise signal, which transforms the phase of signal by 90°. To overcome time-domain aliasing Hilbert transform can be formulated, and there are various acoustic frequency band and the time corresponding to these frequency band varies. These acoustic waveforms are processed through filters and the resultant is being the envelope and they are sent to the binary masking segment.

The transform is used to particular band of speech signal to extract the region of interest, and will calculate the Fourier transform for the speech signal in entire bandwidth is being considered. It will eliminate the negative frequencies in the second phase, and the bandwidth of the region of interest is subjected to inverse Fourier transform and the resultant will be a signal with real and imaginary parts. So, the Hilbert transformed pairs are complex valued signal.

Time-Frequency Analysis:

Analysis of signals is an adherent one to be followed to effectiveness of the bandwidth and even the manipulation of energy is in a higher sequences. In general many speech signals consists of spectral components in relation to acoustics of the audio signal. Fourier analysis plays a prominent role in estimation of signal parameters like magnitude, distribution of the signal energy level but timing sequences are not being available in these forms. In further ways to extract the time of a particular signal, every component is subjected to a transform in those sequences. In speech signal considered here the time domain signal shows the amplitude of pitch against the milliseconds. Time framing is also another part of the speech signal analysis where normally each content of the signal is broken up into frames as the length of the speech signal might be longer and it cannot be received by the ender user in immediate response. So, frame conversion process is carried and overlapping is done later.

Speech Intelligibility:

The ability to understand what is being spoken by the end user and they respond to it accordingly is where the articulation factor lies. In regards to the hearing impaired patients the articulation is the major factor were the subject understand and responds accordingly depending upon the processor advancements in the hearing aid. Articulation rates vary depending upon the filters being used.

Further Spectrograms are carried out for the target with noise input signal and the SNR value is to be comparatively lower for the desired target signal this process is being performed for both the target and noise in various frequency bands. These results revel that the intelligibility has be considerably increased for the desired target signal with a degraded noise signal at specific angle.

5. RESULTS AND DISCUSSION:

In the speech signal band target with noise spectrum (Fig 2.a) is been available is subjected to Hilbert transformation in lab view and the average of noise signal is been taken then the T/F array values are being sent to the FPGA for Binary making analysis same, after the analysis these signal target with noise and target is subjected to the Band stop filter. The signal is denoised from this analysis and required target signal is synthesized the in Lab view. The overall experimental block setup is been displayed below as shown in (Fig.2b)

These are the modules are being designed in the Lab View environment.

a) The two sources noise and target with noise is taken into consideration

b) Time versus Frequency conversion is carried out and binary masking algorithm is implement for noise signal

c) The above algorithm is implemented for 105 bandwidth frequency

d) Then the spectrum for voice signal is calculated and bandwidth for Time vs Frequency conversion is recalculated

e) The step ‘ii’ is again repeated for the new bandwidth

f) The results shows the noised get cancelled for the proposed bandwidth

Fig.2a Target with Noise Spectrum

Fig.2b Block diagram for Noise Suppression technique

Time-Frequency Conversion:

The band pass filter is applied to the given noise signal (Fig. 3a) thereby average (Fig.3b) value of noise is taken and band pass output is analyzed for every 100ms.If the amplitude is greater than the noise average it is marked that the particular frequency is present at that time. The extracted target+noise signals are being subjected to Hilbert transform to estimate the amplitude modulation envelope-AM Envelope- depending upon the usage of variable band pass filters. In this analysis the highest bandwidth of the frequency range is divided by the number of band pass filters could be used. Initially all these sequences are being done to the noise signal and their average is being found out. Initially, noise and target are analyzed in the frequency band of 105Hz (Fig.3c) and later in reduced (1000/28≈40Hz) frequency. The 1000Hz is the maximum frequency of the target which is been used as the input source. Further, the same time versus frequency analysis is being done in reduced frequency band of 40Hz for the target and the noise signal (Fig.3d). From these results it is confirmed that the reduced bandwidth will provide efficient noise reduction than the 105Hz analysis. The resultant array value of each Band pass filter is being fed to the FPGA through the DAQ NIUSB6501.After completion of binary mask the output is fed again through DAQ to Lab View the masked frequency band will cancel the noise in the target. The Lab View coding for Bandstop implementation is shown in Fig.8. The resultant array of band stop filters array values are being fed to the binary masking is implemented in FPGA. Verilog coding is written for the binary mask algorithm. Then simulated using the Modelsim and implemented in Xilinx Spartan 3E FPGA kit. The simulation results are shown in (Fig 4 and Fig.5).

6. CONCLUSION:

On comparison of frequency band of noise and target values it is noted that some of the noise and data are in same frequency so, if the noise bandwidth is masked the voice will also get cancelled this decreases the intelligibility of the conversion but, at the same time if the time versus frequency conversion is carried out in the reduced bandwidth there is a considerable amount of change in the target and noise signal. So, the masking of the noise frequency band will not affect the target and results in efficiency of noise clearance. Further, this process can be implemented for the processors completely in the canal (CIC). This entire process could be computationally increased to nanoseconds. So, the impaired gets the stimuli instantaneously and responds accordingly.

7. REFERENCES:

1. Valerie Hanson and Kofi Odame. Real-Time Embedded Implementation of Binary Mask Algorithm for Hearing prosthetics. IEEE Transactions on Biomedical Circuits and Systems., 2014; 8(4): 465-473.

2. Lars Baekgaard, Niels Ole Knudsen; Tayyib Arshad, and Hanne Pernille Andersen; Designing Hearing Aid Technology to Support Benefits in Demanding Situations, Part I, The Hearing Review, 2013.

3. Adnylo; Digital Hearing Aids, 2017.

4. Taal, C.H., Hendriks, C.R., Heusdens, R., Jensen, J. An algorithm for intelligibility prediction of time-frequency weighted noisy speech. IEEE Trans. Audio, Speech, Lang. Process., 2011; 19(7): 2125–2136.

5. Kostas kokkinakis and Casey Cox. Reducing the impact of wind noise on cochlear implant processors with two microphones. The Journal of Acoustical Society of America, 2014; 135(8): 219-225.

6. Park, H.M., Oh. S.H., Lee., S.Y. A bark-scale filter bank approachto independent component analysis for acousticmixtures. Neurocomput., 2009;73(1): 304–314.

7. Madhu, N., Spriet, A., Jansen, S., Koning, R., Wouters, J. The Potential for Speech Intelligibility Improvement Using the Ideal Binary Mask and the Ideal Wiener Filter in Single Channel Noise Reduction Systems: Application to Auditory Prostheses. IEEE Transactions on Audio, Speech and Language Processing, 2013; 21(1): 63-72.

8. Peeters, H., Lau, C.C., Kuk, F. Speech-in-noise potential of hearing aids with extended bandwidth. Hear. Rev., 2011; 18(3): 28–36.

9. Ramesh Kumar, M., Caleb Kronen., Kathryn Arehart., James Kates., Quality of voices processed by hearing aids: Intra-talker differences. Proceedings of Meeting on Acoustics., 2013; 19(1): 1-7.

10. Dawson, W., Stefan Mauger., Adam, A., Clinical evaluation of Signal-to-Noise Ratio–Based Noise Reduction in Nucleus Cochlear Implant Recipients. The Official Journal of American Auditory Society, 2013; 32(3): 382-390.

11. Daniele Giacobello., Mads Græsbøll Christensen., Manohar. N., Søren Jensen., Marc Moonen. Sparse Linear Prediction and Its Applications to Speech Processing, IEEE Transactions on Audio, Speech, and Language Processing., 2012; 20(5): 1-8.

12. Ulrik Kjems., Jesper Boldt., Michael Pedersen., Role of mask pattern in intelligibility of ideal binary-masked noisy speech, J. Acoust. Soc. Am., 2009; 126(3): 1415-1426.

Received on 10.06.2019 Modified on 06.07.2019

Research J. Pharm. and Tech. 2020; 13(1): 178-182.

DOI: 10.5958/0974-360X.2020.00036.0


Fig.3a BPF analysis of Noise Signal	Fig.3b Averaging of Noise Signal

Fig. 3c T Vs F output for Noise and Target signal analyzed in 105Hz frequency	Fig.3d TVs F output for Noise and Target signal analyzed in 40 Hz frequency


Fig.5a Binary Masking Array Output	Fig.5b Final Implementation of Binary Mask algorithm in FPGA