A COMPARISON OF DIFFERENT METHODS FOR NOISE REDUCTION IN HEARING AIDS Jean-Baptiste Maj E-mail:
[email protected]
Jan Wouters Lab. Exp. ORL, KULeuven, Kapucijnenvoer 33 B 3000 Leuven
Marc Moonen ESAT-SISTA, KULeuven, Kardinaal Mercierlaan 94 B 3001 Leuven
E-mail:
[email protected]
E-mail:
[email protected]
ABSTRACT
speech-processing algorithm for hearing aids with two microphones [5][6]. Relative to the omnidirectional microphone, the A2B can improve the speech intelligibility by 7dB [6]. The A2B has the advantage that we can in combination with different microphone configurations. Here, the reason why we will compare the A2B for a selection of such combinations (HA1, HA2 and HA3)
This paper presents a comparison of different methods for noise reduction in Behind-The-Ear (BTE) hearing aids with two microphones. In this work we study three microphone configurations. A first hearing aid (HA1) uses two omnidirectional microphones (2× ×HAO), a second hearing aid (HA2) uses, a directional microphone (HA2D), created with the dual microphone configuration, and one of both omnidirectional microphones (HAO), and a third (HA3) has a hardware directional (HA3D) and an omnidirectional microphone (HAO). With each hearing aid a two-stage adaptive beamformer (A2B) can be added. To make a comparison between different methods (HAO, HA2D, HA3D, HA1+A2B, HA2+A2B and HA3+A2B), we will measure directivity index (DI) and signal-to-noise ratio (SNR).
INTRODUCTION The most common complaint of hearing aid users related to that difficulty on understanding speech in noisy condition. Present-day, the hearing aids generally use a directional microphone to improve speech intelligibility in noise. The first directional microphones were hardware directional microphones and some studies have shown that directional microphones improve speech intelligibility with respect to omnidirectional microphones more specifically they can provide an improvement of 3 to 5dB in signal-to-noise ratio (SNR) in difficult listening situations [1][2]. Another solution to create directional patterns is to use a microphone array with a fixed beamformer in a broadside or endfire array configuration, by means of which an improvement of more than 4dB has been obtained in diffuse noise [3]. Recently, hearing instruments with two omnidirectional microphones, employing the socalled ‘dual-microphone technique’, have been introduced [4]. Recent studies have presented good results with a two-stage adaptive beamformer (A2B)
NOISE REDUCTION SYSTEMS The directional microphone (HA3D) (figure 1) has two inlets that allow sound to enter both the front and rear acoustical cavities and arrive on either side of the microphone diaphragm. The acoustic time delay is selected so that if a sound strikes the rear inlet first, the arrival time of this sound will be reduced so that it reaches the diaphragm of the microphone at the same time as the sound entering the front inlet, causing cancellation. Time delay acoustical network Rear microphone port
Diaphragm Front microphone port
Effective port spacing
Figure 1: Hardware directional microphone (HA3D)
In the dual-microphone configuration, a directional microphone (HA2D) is created with two omnidirectional microphones. The directional microphone is produced as the difference between the front microphone and the delayed signal of the rear microphone (figure 2). Front microphone d
w0 + w1
Rear microphone
τ
-
Directional microphone
Figure 2: Dual microphone technique (HA2D)
Speech reference
Front microphone
+
Σ
Delay
Delay
+
Σ
+
Result
−
− + Σ
Filter
Adaptive filter
Rear microphone Noise Reference
Figure 3: Two-stage adaptive beamformer
The microphone parameters are the interport distance d and the internal delay τ. The two port microphone can be viewed as a two-element endfire array with frequency-dependent weights w1(f)=e-j2πfτ and w0(f)=1 [7]. A two-stage adaptive beamformer (A2B) has been developed by Vanden Berghe and Wouters [5] for hearing aids with two microphones. The signal processing strategy is shown figure 3. Both microphone signals contain speech as well as noise. The first section of the system improves the noise reference by eliminating speech. The second section consists of an unconstrained adaptive noise cancellation [8] and models the difference between the noise reference and the noise portion of the delayed speech reference.
Department at the KULeuven. To calculate the SNR, we recorded the signals in similarly conditions, but we used a reverberant room with a reverberation time of 0.5s and angles 0° to 345° in steps of 15° for the noise signal. Furthermore, the hearing aid, in this case, was not stand-alone but mounted on a manikin. DIRECTIVITY INDEX We have used the directivity index to compare the different noise reduction systems with respect to their spatial characteristics. The directivity index indicates the extent to which a sound field consisting of uncorrelated plane waves uniformly distributed in space, is attenuated in comparison with sound transmitted in the direction of the maximum array gain. The directivity index is calculated using:
DI ( f ) = 10. log10 Q( f )
SIGNALS For recordings, we utilized a speech-weighted noise based on the multilanguage long-term average speech spectrum (LTASS [9]). To calculate the directivity index (DI see below), we recorded a stationary signal at 0° representing the speech signal and a second stationary signal at angles ranging from 15° to 180° in steps of 15° for the noise.
with Q( f ) =
4π P0 180 / ∆θ
2.π .
∑ n =1
2
P(θ n ) . sin θ n .∆θ 2
P(θ) : System output energy for sound at the angle θ. P0: energy of sound at azimuth 0°. n: point number of measurement
Loudspeaker for the speech
1m Loudspeaker for the noise Stand-alone hearing aid Figure 4:Configuration to record the signals
These measurements were made with stand-alone hearing aids in the anechoic room of the Physics
According to this definition, the directivity index of an omnidirectional microphone in anechoic conditions is DI(f)=0dB for all frequencies. A directional microphone with a directivity pattern equivalent to a cardioid microphone, has a directivity index DI(f)=4.7dB [7]. The following figures 5, 6 and 7 show the directivity index for each system. Each hearing aid presents a DI of the omnidirectional microphone around 0dB, which confirms that we have true omnidirectional microphones. It appears that the directional microphone (HA3D) gives better results compared to
the dual microphone technique (HA2D) and the omnidirectional microphone. 14 12 10 8
D.I. (dB)
6 4 2 0 -2 -4 -6 -8 2 10
Omnidirectional microphone (HA1) Omnidirectional microphone HA1+A2B 3
10 Frequency (Hz)
4
10
Figure 5: DI for the hearing aid HA1 and the beamformer A2B 14
Figure 5 and 6 show a negative DI for the directional microphone (HA2D) and the beamformer for some frequencies. Indeed, in the low frequencies 250Hz to 350Hz the two-noise reduction systems perform less than the omnidirectional microphone. Thompson argues that it is necessary that the sensitivity of the two microphones of the HA1 be adequately matched in both magnitude and phase to use the dual microphone technique (HA2D) [4]. It appears that we need matched microphones for the beamformer too. The best solution is obtained when we add the directional microphone with the two-stage adaptive beamformer (HA3+A2B). The directional microphone (HA3D) gives a better estimation of speech than the dual-microphone technique (HA2D) and the omnidirectional microphone for all frequencies. A better speech reference obviously yields better results for the beamformer. SIGNAL-TO-NOISE RATIO
12
We have defined the signal-to-noise ratio as:
10 8
SNR = 10. log10
D.I. (dB)
6 4
-2 -4
Directional microphone (HA2) Omnidirectional microphone HA2+A2B
-6 3
10 Frequency (Hz)
4
10
Figure 6: DI for the hearing aid HA2 and the beamformer A2B 14 12 10 8 6 D.I. (dB)
2
and noise the signal for an angle between 15° and 345°.
0
4 2 0 -2 -4
Directional microphone (HA3) Omnidirectional microphone HA3+A2B
-6 -8 2 10
2
with speech the signal 0° Figure 5 shows theatSNR effects of each
2
-8 2 10
∑ speech ∑ noise
3
10 Frequency (Hz)
Figure 7: DI for the hearing aid HA3 and the beamformer A2B
4
10
Figure 8, 9 and 10 show the SNR effect for each system against one noise source with a speechweighted spectrum (15° to 345°) relative to the speech (angle 0°). The curve of the omnidirectional microphone is not flat around zero due to the effect of the diffraction around the head of the manikin. As shown in the DI-curves, for all angles, the hardware directional microphone (HA3D) performs better than the HA2D configuration and the omnidirectional microphone. The dual-microphone technique performs worse than the omnidirectional microphone between angles 15° and 60°. Furthermore, when we add the beamformer to the hearing aid HA3, we obtain an improvement of 15 dB between the omnidirectional microphone and the beamformer at the angle 90°. For comparison, with the hearing aid HA1, at the same angle, we only obtain an improvement of 7dB. A better estimation of the speech reference can improve the SNR between the beamformer and the omnidirectional microphone by more than two dBs.
14
CONCLUSION
HA1+A2B Omnidirectional microphone Omnidirectional microphone (HA1)
12 10
SNR (dB)
8 6 4 2 0 -2 -4
0
45
90 135 180 225 270 Angle relative to the direction of the speech (°)
315
360
Figure 8: SNR for the hearing aid HA1 and the beamformer 14
REFERENCES
12 10
SNR (dB)
8 6 4 2 0 HA2+A2B Omnidirectional microphone Directional microphone
-2 -4
0
45
90 135 180 225 270 Angle relative to the direction of the speech (°)
315
360
Figure 9: SNR for the hearing aid HA2 and the beamformer A2B 14 12 10 8 SNR (dB)
The DI and the SNR analysis show that the best noise reduction configuration is obtained with the hearing aid HA3 (hardware directional and omnidirectional microphone) plus the two-stage adaptive beamformer signal processing. Indeed, the hardware directional microphone gives a better speech reference for the two-stage adaptive beamformer strategy. A similar scheme with a directional microphone (HA2D) based on the dual microphone technique is found to give poor performance because it has an inverse directivity (negative value of the directivity index on figure 4) at low frequencies.
6 4 2 0 HA3+A2B Omnidirectional microphone Directional microphone (HA3)
-2 -4
0
45 90 135 180 225 270 315 Angle relative to the direction of the speech source (°)
Figure 10: SNR for the hearing aid HA3 and the beamformer A2B
360
[1] A. R. Leeuw, W. A. Dreschler. “Advantages of directional hearing aid microphones related to room acoustics”. Audiology 30, 330-344, 1991. [2] D. B. Hawkins, W. S. Yacullo. “SNR advantage of binaural hearing aids and directional microphones under different levels of reverberation”. Journal of Speech and Hearing Disorders, 49, pp. 278-286. [3] W. Soede, F. Bilsen, A. Berkhout, “Developement of a Directional Hearing aid Instrument based on array technology”, The Journal of the Acoustical Society of America., 94(2), 785-798, 1993. [4] S. C. Thompson. “Dual microphones or directionalplus-omni: which is the best”. The hearing review, vol. 3, high performance hearing solutions, p. 31-35, 1999. [5] J. Vanden Berghe, Jan Wouters. “An adaptive noise canceller for hearing aids using two nearby microphones”. The Journal of the Acoustical Society of America, 103(6), 3621-3626, 1998. [6] J. B. Maj, J. Wouters, M. Moonen. “Noise Reduction in hearings aids with two microphones”. Submitted to Eusipco-2000. [7] R. W. Stadler, W. Rabinowitz. “On the potential of fixed arrays for hearings aids”. The Journal of the Acoustical Society of America., 94(3), 1332-1342, 1993. [8] B. Widrow, S. Stearns, Adaptive Signal Processing, Prentice-Hall, 1985. [9] D. Byrne and al. “An International Comparison of Long-Term Average Speech Spectra”, The Journal of the Acoustical Society of America, 96 (4), 2108-2120, 1994. [10] L. Beranek, Noise and Vibration Control, Institute of Noise Control Engineering, 1988.
