Better Speech Perception in Noise With an Assistive Multimicrophone Array for Hearing Aids

Transcription

1 Better Speech Perception in Noise With an Assistive Multimicrophone Array for Hearing Aids Heleen Luts, Jean-Baptiste Maj, Wim Soede, and Jan Wouters Objective: To evaluate the improvement in speech intelligibility in noise obtained with an assistive real-time fixed endfire array of bidirectional microphones in comparison with an omnidirectional hearing aid microphone in a realistic environment. Design: The microphone array was evaluated physically in anechoic and reverberant conditions. Perceptual tests of speech intelligibility in noise were carried out in a reverberant room, with two types of noise and six different noise scenarios with single and multiple noise sources. Ten normal-hearing subjects and 10 hearing aid users participated. The speech reception threshold for sentences was measured in each test setting for the omnidirectional microphone of the hearing aid and for the hearing aid in combination with the array with one and three active microphones. In addition, the extra improvement of five active array microphones, relative to three, was determined in another group of 10 normal-hearing listeners. Results: Improvements in speech intelligibility in noise obtained with the array relative to an omnidirectional microphone depend on noise scenario and subject group. Improvements up to 12 db for normal-hearing and 9 db for hearing-impaired listeners were obtained with three active array microphones relative to an omnidirectional microphone for one noise source at 90. For three uncorrelated noise sources at 90, 180, and 270, improvements of approximately 9 db and 6 db were obtained for normal-hearing and hearing-impaired listeners, respectively. Even with a single noise source at 45, benefits of 4 db were achieved in both subject groups. Five active microphones in the array can provide an additional improvement at 45 of approximately 1 db, relative to the three-microphone configuration for normal-hearing listeners. Conclusions: These improvements in signal-to-noise ratio can be of great benefit for hearing aid users, who have difficulties with speech understanding in noisy environments. (Ear & Hearing 2004;25; ) Laboratory for Experimental Otorhinolaryngology, Katholieke Universiteit Leuven, Leuven, Belgium (H.L., J.-B.M., J.W.); SCD- SISTA, Katholieke Universiteit Leuven, Leuven, Belgium (J.- B.M.,); ARDEA, Leiden, The Netherlands (W.S.); and Leiden University Medical Centre, Leiden, The Netherlands (W.S.). DOI: /01.aud ba Speech understanding in noisy backgrounds is often very difficult for hearing-impaired (HI) listeners. Therefore, noise reduction strategies are developed and implemented in hearing devices. These techniques can be classified into two different approaches: a first, based on a single microphone, and a second, based on multiple microphones. A wellknown single microphone beamformer approach is a hardware directional microphone. Some studies have shown that the directional microphone may give a speech reception threshold (SRT) improvement of approximately 3 db in difficult listening conditions (Hawkins & Yacullo, 1984; Leeuw & Dreschler, 1991). Based on the hardware directional microphone design, software directional microphones have been applied (Thompson, 1999). This method uses two omnidirectional microphones as the front and the rear entrance comparable to a hardware directional microphone. It acts basically the same as a single directional microphone but with the advantage that the delay can be set with an analogue delay circuit or with a digital signal processor with a delay section. An improvement of the SRT of approximately 3 db in similar conditions has been confirmed (Wouters, Litiere & van Wieringen, 1999). More recently, adaptive directional microphones also based on two omnidirectional microphones have been developed and implemented in commercial hearing aids (Luo, Yang, Pavlovic & Nehorai, 2002; Ricketts & Henry, 2002). With more than two microphones, Soede developed a fixed beamformer technique (Bilsen, Soede & Berkhout, 1993; Soede, Berkhout & Bilsen, 1993a; Soede, Bilsen & Berkhout, 1993b; Soede, Bilsen, Berkhout & Verschuure, 1993c), in which five cardioid microphones were spaced on a 10-cm-long endfire array (an endfire array has its microphones collinear with the target direction). This technique was compared with an omnidirectional microphone system, and an average SRT improvement of 7 db was obtained in a diffuse noise field with the endfire array configuration. Approaches similar to the Soede strategy have been investigated with rectional microphones (Desloge, Rabinowitz & Zurek, 1997; Kates, 1993; Stadler & Rabinowitz, 1993), and evaluations of these different microphone arrays in reverberant conditions have shown significant improvements of speech intelligibility in noise 0196/0202/04/ /0 Ear & Hearing Copyright 2004 by Lippincott Williams & Wilkins Printed in the U.S.A. 411

2 412 EAR & HEARING / OCTOBER 2004 (Desloge et al., 1997; Kates & Weiss, 1996; Saunders & Kates, 1997). Derived from the fixed beamformer strategies, adaptive beamformers have also been developed (Greenberg & Zurek, 1992; Vanden Berghe & Wouters, 1998). Evaluations of adaptive beamformers for two-microphone, behind-the-ear devices show significant improvements in the near-field over directional microphones for one or two noise sources (Maj, Wouters & Moonen, 2003; Wouters & Vanden Berghe, 2001). However, for more practical situations with numerous noise sources such as cocktail parties, with multiple speakers or situations with diffuse, reverberated noise, further research needs to be done. It is to be expected that it will be difficult to suppress adequately a multitude of independent noise sources with a two-microphone system. The microphone array used in this study is a prototype developed by Etymotic Research, based on the Delft array, a so-called fixed delay-and-sum beamformer array as described by Soede (Soede et al., 1993a). Five microphones are placed in a straight line, forming an endfire array; signal processing is done with analog low-voltage transistor electronics. Sound waves coming from the front, the sides, or the rear will have different arrival times. Each microphone signal is electronically delayed relative to the adjacent one. When a sound wave is coming from the front, the acoustical input is received out of phase, but each microphone signal is delayed differently according to the sound travel time, resulting in a high output signal after summation. A sound wave coming from the side arrives simultaneously at each side microphone. After delay of the microphone signals, signals are summed, yielding a lower output. A delay-and-sum beamformer array is the most effective when the length of the overall array of microphones is larger than the wavelength of the sound. This means that the beam directivity changes with frequency. The overall length of the array in this study is 11.6 cm. The array is used in combination with a behindthe-ear hearing aid. The system has a hardware zoom function that makes it possible to adapt the beam width by using one, three, or five microphones. In some situations, a high directivity is not necessary or can even be undesirable or dangerous. Therefore, the user must be able to switch off the directionality of the microphone by choosing between a directional and an omnidirectional position. This array is meant as an assistive listening device to be used with a hearing aid or a cochlear implant. The hardware and signal processing of the new array is significantly different from the Delft array. The new array consists of five bidirectional microphones (also called dipole or figure-eight microphones), in contrast with the Delft array that consisted of five cardioid microphones. The bidirectional microphone was chosbn because it has the narrowest beam in the frontal direction. Furthermore, there is no need for an external acoustic delay on the back port, as in (hyper-)cardioid microphones. This difference in hardware results in a higher directivity at the low frequencies. The difference in signal processing also affects the directivity versus frequency of the array. In the Delft array, the electronic circuit provided a delay that was the same for all frequencies. That delay in the Delft array was optimized to get the highest Directivity Index (DI) at 4000 Hz. However, a shortcoming of this approach is that this delay was not optimal for the lower frequencies. In the new design of the circuitry, a frequency-dependent delay is obtained with a phase-shifting circuit. This circuit was optimized to reach the highest possible directivity in the frequency range between 500 and 4000 Hz, based on the use of dipole microphones. This optimization results in a significant improvement of the directivity of the array for frequencies between 500 and 1600 Hz of 2 to 3 db for an array of five microphones with the same length. The directivity at 3000 Hz is nearly the same. Based on this optimized circuit, a three-microphone array could be defined with an average directivity comparable to the Delft array with five microphones. However, the new three- and five-microphone arrays have a lower directivity for 4000 and 5000 Hz (see Fig. 5). Since the new array has a different directional behavior compared with the Delft array, as a result of differences in microphone hardware and signal processing, the question was whether the present design would give comparable results. The hardware zoom function of the new array offers the possibility to compare the microphone array of one, three, and five active microphones with an external omnidirectional hearing aid microphone. Furthermore, over the past 10 years, it became clear that the unweighted DI or signal-to-noise ratio (SNR), as used in the papers of 1993, do not accurately predict the performance of a device. Nowadays, we use the intelligibility-weighted DI as an estimation of expected performance. The microphone array was evaluated physically in anechoic and reverberant conditions and perceptually in reverberant conditions. Listening tests were carried out with two groups of 10 normalhearing (NH) subjects and a group of 10 HI subjects. Two noise materials were used in six different noise scenarios. Speech reception thresholds in noise were measured for 27 test conditions per person.

3 EAR & HEARING, VOL. 25 No Hz * 1kHZ 2kHz A\erage - Free-field Distance (m) Fig. 1. Theoretical free-field level of a sound source, as a function of distance, compared with the measured level of one single loudspeaker in the test room. Measurements were relative to the sound level at 25 cm from the loudspeaker. Physical Evaluation Physical measurements were partly performed stand-alone in an anechoic room and partly with the array on a mannequin head in a reverberant test room, the same room that was used for the perceptual measurements with the subjects. In anechoic conditions, all recordings were made with a pink noise in stand-alone configuration. The DI and intelligibility-weighted polar diagrams were calculated for the array using one, three, and five active microphones. In the reverberant room, the microphone array was positioned on a mannequin. A stationary speech-weighted noise of the Dutch sentences spoken by the female speaker, the same material as used in the perceptual evaluation (Versfeld, Daalder, Festen & Houtgast, 2000), was used for the recordings. Intelligibility-weighted polar diagrams were calculated for one, three, and five active microphones in the array. Experimental Setup The reverberant test room had a volume of 70 m 3 and a reverberation time (T60) of 0.76 seconds. The critical distance was measured with a sound level meter (Rion NA-27) in one-third octave bands. Figure 1 shows the theoretical freefield level of a sound source compared with the measured level in the test room of one single loudspeaker. The measurements were relative to the sound level at 25 cm from the loudspeaker. The critical distance was approximately 2 meters over the average of 500, 1000, and 2000 Hz. In anechoic and reverberant conditions, the signal at the output of the microphone array was connected to and amplified by a Larson-Davis 2200C amplifier and was directly led to the LynxOne sound card (24-bit analog-to-digital conversion) for recording and analysis. The presented signals were sent through a loudspeaker (Yamaha CBX-S3) located 1 meter from the microphone array (in stand-alone configuration) or the center of the head of the mannequin in the azimuth plane. The level at the loudspeaker was adjusted to get a sound level of 70 db SPL at the microphone array (in stand-alone configuration) or at the center of the head of the mannequin. The output of the array with one, three, and five active microphones was recorded for different locations of the loudspeaker, corresponding to angles between 0 and 360 in steps of 15. Performance Measures The DI is a measure to define the directivity of hearing aids, and changes in DI appear to be correlated to changes in speech intelligibility in noise (Beranek, 1954; Ricketts, 2000a). With 0 the azimuth coordinates and.1) the elevation coordinates, the directivity index equals: DI(f) = 10.Log 27r I 0 0 Tr 47r X P(f,0,0)I2 I P ( f, 19 4)1 2. 1sin 01. d0 do where the 1 P(f 0,4)) 1 2 is the magnitude of the mean squared sound pressure, at frequency f, of the output signal of the hearing aid when the sound source is located at the coordinate (0,4o). If symmetry is assumed in the vertical plane and there is reasonable symmetry around the horizontal plane, the DI can be calculated from only the 1 P(f, = 0)1 2 values recorded at discrete angles of the horizontal plane by using the simplified formula (Beranek, 1954; Ricketts, 2000b): DI(f) = 10.Log 180 /AO 4 x 57.3 x IP( f0,0)12 E IP( /AO = 0)1 2. 1sin AO (1) (2) where f are the center frequencies from 160 Hz to 5000 Hz of the one-third octave bands. Because the measurements are carried out with the device in stand-alone configuration and only the azimuth angles are measured, Equation (2) is considered in this study. An intelligibility-weighted version of the DI can be defined as: 16 = E I i DIi (3) i =1 L

4 414 EAR & HEARING / OCTOBER 2004 where I i are the weights for the importance of the i-th third octave band (ANSI, 1997). The performance of the microphone array is assessed by measuring the intelligibility-weighted SNR defined as: SNR, s ll = E I i A i SNR i (4) i=1 where SNR i is the SNR measured (in db SPL) in the i-th third octave band. I and A i are the weights for the importance of the band and the audibility function, respectively, as described by the speech intelligibility index SII (ANSI, 1997). The A i values were calculated in accordance with the ANSI S3.5 (1997) standard, using the one-third octave band procedure and assuming the standard speech spectrum level produced at a normal vocal effort (62.4 db SPL). If the SNRs H is calculated for different locations of a sound source, a polar diagram can be drawn. A polar diagram shows the sensitivity of the signals (i.e., the output of the noise reduction algorithms) as a function of the direction of the source (i.e., azimuth). The plotted values are relative to the angle of the direction 0 (in front of the head), which is considered as the angle of incidence of the speech source. Perceptual Evaluation Subjects Three groups of subjects participated in the experiments: two groups of 10 NH volunteers and 10 HI subjects. The age range of the first NH group was between 18 and 23 years. In the second NH group, ages varied from 23 to 42 years. In both groups, audiometric thresholds were less than 20 db HL at all octave frequencies between 250 and 8000 Hz. The HI subjects had bilateral mild to moderate pure sensorineural hearing losses and were using a hearing aid with telecoil on their right ear. Six subjects used a Widex Senso C-8, two a Widex Logo L-6-T, one a Danavox 133-PP, and one a Phonak Pico SC. These subjects varied in age between 37 and 73 years. They were selected from the patient population at the Ear, Nose, and Throat Department of the K.U. Leuven University Hospital. During the speech-in-noise tests, only the right ear was tested, so the subject's left ear had to be worse or equal to their right ear. Since the test settings had some constraints, the speech intelligibility and thus the hearing thresholds of the subject's test ear had to be within certain limits. A maximum pure tone average of 50 db HL was permitted. Mean hearing thresholds of the right ears are presented in Figure 2. The pure tone average ranged from 33 to 45 db m Frequency (Hz) Fig. 2. Mean pure-tone thresholds and standard deviations in db HL for the right ear of the 10 hearing-impaired subjects. HL, with an average of 39 db HL and a standard deviation of 5 db. Experimental Setup The microphone array prototype was attached to the leg of a pair of glasses and was connected to a special transmission unit (Sennheiser EZI120) with a transmitting telecoil that was positioned close to the hearing aid, which was equipped with a receiving telecoil. The signals were transmitted wirelessly from coil to coil. The NH subjects used the microphone array in combination with a behind-the-ear Siemens Prisma HdO+ on their right ear. The settings of the hearing aid were equal for all NH subjects. Audiometric thresholds of 20 db HL were imported; noise reduction and directional microphone characteristics were switched off. Sound was delivered through an ER3 14B ear tip that was attached to the hearing aid with an adapted piece of tubing and an elbow. The HI subjects used their own hearing aid and earmold; old tubing was replaced, and batteries were checked. Each hearing aid contained an omnidirectional microphone. If the used hearing aid had a volume control, the volume was set at a comfortable level for the subject and kept constant for the complete test session. To use the microphone array, the telecoil had to be switched on and the microphone switched off. Only the right ear was tested with the microphone array. A monaural listening situation was created by an earplug (Bilsom 303S-30 or 303L-30) and by an ear cap (Bilsom 2301) at the other ear. The experimental setup is shown in Figure 3. The perceptual tests were performed in the reverberant test room and in the same conditions as in part of the physical evaluation (described above). Five identical loudspeakers (Yamaha CBX-S3) were placed around the testing person. The middle of each loudspeaker was on the same height as the microphone array (140 cm), and the distance between the loudspeakers and the middle of the head was 1 meter. Subjects were seated on a height-adjustable chair. The subject and the test leader were in different

5 EAR & HEARING, VOL. 25 No Test leader CONTROL ROOM Sony CD991 Sony CD991 PC Madsen OB 822 Amplaid 309 TEST ROOM Subject hearing aid + array Fig. 3. Experimental setup. Reverberation time (T 60 ) of the test room is 760 ms. rooms. The response of the subject was recorded, amplified, and transferred to the control room, where the test leader administered the experiment. The speech source was always at 0 azimuth. A computer (with a 16-bit sound card) controlled the speech audiometry test and the output of the speech material; a MADSEN OB822 audiometer (right channel) was used to adjust the level of the speech. Two SONY CD991 CD players were used for presenting the noise signals from CD. The level of the noise at 0, 45, and 90 was controlled by the same MADSEN OB822 audiometer (left channel). An AM- PLAID 309 audiometer controlled the level of the noise at 180 and 270. The different noise sources were calibrated separately to get a constant sound level of 60 db SPL in the middle of the head of the subject. As a result, a sound level of 63 db SPL and 64.8 db SPL was obtained with two and three noise sources, respectively. When multiple noise sources were used with the same spectrum, the noise signals were presented through different CD players so that the noise was uncorrelated. METHODS Hearing thresholds for both ears were determined for all octave audiometric frequencies between 250 and 8000 Hz with the Hughson-Westlake method (5 up-10 down). A MADSEN Midimate 622 audiometer with TDH-39 headphones was used. The array was evaluated in two stages. In a first stage, the benefit with one and three active array microphones was determined with 10 NH and 10 HI subjects. SRTs in noise were measured for Dutch VU sentences (female speaker). These sentences were developed by Versfeld et al. (2000) and ate an extension of the materials of Plomp and Mimpen (1979). The SRT is defined as the signal level in db SPL that is required to understand 50% of the presented sentences. An adaptive testing procedure was used: The noise level was kept constant at 60 db SPL and the level of the sentences was adjusted in steps of 2 db. In each condition, 13 sentences were presented. The level of the last 10 sentences was averaged to obtain the SRT. SRTs were measured for the omnidirectional hearing aid microphone (Omni), the microphone array with one active bidirectional microphone (1 Mic), and with three active microphones (3 Mic). SNRs at 50% intelligibility were deduced from the SRT in noise by subtracting the noise level from the speech SRT. Two types of background noise were used in the tests: a stationary speech-weighted noise, with a spectrum equal to the average spectrum of the sentences, and a nonstationary multitalker babble noise. The multitalker babble was taken from the compact disk Auditory Tests (Revised) edited by Auditec of St. Louis. The speechweighted noise was presented in six different configurations: one noise source at 0, 45, 90, or 180, two independent noise sources at 90 and 270, or three independent noise sources at 90, 180, and 270. The multitalker babble noise was presented in three noise configurations: one noise source at 0, 90, and 180. In total 27 conditions were tested per subject. Each sentence list was presented just once per subject to avoid learning effects. Because of time constraints, the perceptual evaluation considered the differences between the omnidirectional microphone and the array with one and three active microphones for the different test conditions, since the largest effects were expected. To define the additional improvements in speech recognition that can be provided by extending the array to five microphones (5 Mic), some further tests were carried out. A new group of 10 NH subjects participated in the perceptual tests. Three and five active array microphones were compared. A speechweighted noise of 60 db SPL was presented in two noise configurations: one noise source at 45 and three independent noise sources at 90, 180, and 270. The test settings and methods were the same as for the NH group described above. Results were presented as differences relative to the three-microphone configuration. Physical Evaluation RESULTS The intelligibility-weighted polar diagrams of the three microphone configurations in anechoic conditions are shown in Figure 4. In stand-alone configuration or mounted on eyeglasses on the mannequin head, the microphones of the array are oriented in an endfire configuration. Small asymmetries in the array housing are responsible for the patterns in the

6 416 EAR & HEARING / OCTOBER Fig. 4. Directional response patterns for 1 ( - ), 3 ( ), and 5 (- - -) active array microphones, SII-weighted, in stand-alone configuration, anechoic room, pink noise. top half and the bottom half that are not symmetrical. Between the angles 330 and 15, the three microphone configurations perform almost identically. The 3 Mic or 5 Mic at all angles outperform the 1 Mic configuration. The differences between the 3 Mic and 5 Mic are small between the angles 270 and 30 and at the angle 90. The biggest SNR improvement between these two microphone configurations is found at 60 and is approximately 6 db. The DI of the different microphone configurations as a function of frequency is depicted in Figure 5. CE ` Frequency (Hz) Fig. 5. Directivity index (DI) as a function of frequency for 1 ( - ), 3 ( ), and 5 (- - -) active array microphones. The DI was measured for one-third octave bands. Dotted line represents the Delft array (Soede et al., 1993a). Fig. 6. Directional response patterns for 1 ( - ), 3 ( ), and 5 (- - -) active array microphones, MI-weighted, on manikin, reverberant room (1 60 = 760 msec), speech-weighted noise. The higher the number of microphones, the higher the DI. For each microphone configuration, the DI increases as a function of the frequency but drops after a specific frequency. These knee point frequencies are 1 khz, 1.8 khz, and 3 khz for the 1 Mic, the 3 Mic, and the 5 Mic, respectively. As described by Meyer (2001), the increase and decrease in DI of multimicrophone arrays goes in steps. The main differences between the three microphone configurations are located at the high frequencies. Below 500 Hz, the difference between the 1 Mic and the 3 Mic does not exceed 1 db. For the difference between the configurations 3 Mic and 5 Mic, this is the case up to 1.6 khz. The DI A' of the 1 Mic, 3 Mic, and 5 Mic configurations are 4.0 db, 7.1 db, and 8.5 db, respectively. Figure 6 shows the polar diagrams in reverberant conditions. Between the angles 270 and 30, the three configurations of the microphones give the same performance in weighted SNR. The difference between the 3 Mic and the 5 Mic configuration is small. The 5 Mic configuration performs slightly better than the 3 Mic between the angles 30 and 60 and between the angles 90 and 120. However, these improvements in weighted SNR do not exceed 1 db. The 1 Mic performs worse than the two other configurations between the angles 30 and 270. The difference between the 1 Mic and the 3 Mic or the 5 Mic configuration is approximately 5 db between 120 and 255.

7 EAR & HEARING, VOL. 25 No TABLE 1. Mean signal-to-noise ratio at 50% score (db) and 95% confidence intervals of the mean for the 27 tested conditions for 10 normal-hearing and 10 hearing-impaired listeners Noise Noise Configuration NH Mean 95% CI Mean 95% CI Stationary speech-weighted 0 Omni -3.2 [-4.4; -1.9] 4.8 [2.9; 6.8] 1 Mic 2.4 [1.8; 3.0] 6.4 [4.7; 8.1] 3 Mic -2.4 [-3.7; -1.0] 3.0 [0.9; 5.0] 45 Omni 0.5 [-1.1; 2.1] 7.2 [4.3; 10.2] 1 Mic 0.4 [ -0.8; 1.5] 7.4 [3.5; 11.3] 3 Mic -4.1 [ -5.3; -2.8] 2.3 [0.4; 4.3] 90 Omni 2.9 [1.2; 4.6] 9.3 [5.6; 13.0] 1 Mic -1.2 [-2.1; -0.3] 5.0 [2.2; 7.8] 3 Mic -9.1 [ -10.3; -7.9] -0.1 [ -2.6; 2.4] 180 Omni -2.9 [ -5.5; -0.4] 3.3 [ -0.1; 6.6] 1 Mic -0.3 [ -1.7; 1.1] 6.4 [2.8; 10.0] 3 Mic -7.6 [-8.9; -6.3] 1.0 ( -0.7; 2.7] Omni 0.9 [ -1.4; 3.3] 6.3 [3.4; 9.2] 1 Mic -2.8 [-4.1; -1.6] 3.7 [0.8; 6.5] 3 Mic -9.1 [ -9.9; -8.3] -0.9 [ -2.4; 0.5] Omni -0.3 [ -1.9; 1.3] 4.3 [1.6; 6.9] 1 Mic -3.5 [ -4.8; -2.3] 1.3 [ -0.3; 2.9] 3 Mic -8.9 [ -10.2; -7.6] -2.3 [-3.6; -1.0] Nonstationary multitalker babble 0 Omni 2.5 [1.9; 3.1] 9.0 [7.3; 10.6] 1 Mic 5.6 [5.0; 6.2] 10.0 [8.3; 11.7] 3 Mic 4.2 [2.9; 5.5] 11.9 [8.9; 15.0] 90 Omni 8.2 [6.5; 9.8] 15.6 [12.7; 18.5] 1 Mic 2.7 [1.4; 4.0] 9.4 [6.6; 12.3] 3 Mic -3.0 [ -4.9; -1.0] 6.9 [4.8; 8.9] 180 Omni 2.8 [0.3; 5.4] 9.9 [6.6; 13.2] 1 Mic 1.1 [-0.1; 2.3] 7.5 [6.6; 8.3] 3 Mic -3.4 [ -4.5; -2.4] 6.3 [4.4; 8.2] HI NH = normal-hearing; HI = hearing-impaired; Omni = omnidirectional hearing aid microphone; 1 Mic = one active array microphone; 3 Mic = three active array microphones. Perceptual Evaluation Evaluation of 1 Mic and 3 Mic The average SNRs at 50% score for all measured conditions are shown in Table 1. The average absolute difference in SNR between the NH and HI group over all measured conditions is 6.8 db (95% CI, 5.4; 8.1 db) (with the first and the second numbers presenting the lower and upper limit of the confidence interval). The SNRs in the conditions with multitalker babble noise are on average 4.8 db (95% CI, 3.9; 5.7 db) and 5.5 db (95% CI, 4.7; 6.3 db) higher than these obtained with the stationary speech-weighted noise for NH and HI subjects, respectively. An interfering multitalker babble noise makes speech understanding more difficult relative to a stationary speechweighted noise. The mean improvements of the SNR for 1 Mic and 3 Mic relative to the omnidirectional microphone are given in Figure 7. Because the NH and the HI groups showed some difference, results are analyzed separately. For both groups, two analyses of variance with repeated measurements are carried out with SPSS 10.0 software: a first analysis on the factors microphones (improvement of 1 Mic and 3 Mic relative to Omni) and noise configuration (one noise source at 0, 45, 90, or 180, two noise sources at 90 and 270, and three noise sources at 90, 180, and 270 ) and a second analysis on the factors microphones (improvement of 1 Mic and 3 Mic relative to Omni), noise (speech-weighted noise and multitalker babble noise), and noise configuration (one noise source at 0, 90, or 180 ). All performed analyses show interaction effects between the different factors (p < 0.05). Consequently, SNR improvements of the different test conditions are compared separately. With a single stationary speech-weighted noise at 90, azimuth HI listeners achieve an average improvement relative to Omni of 4.3 db (95% CI, 0.7; 9.2 db) and 9.4 db (95% CI, 5.9; 12.8 db) for 1 Mic and 3 Mic, respectively. With three active microphones, improvements of 10 db are obtained with two and three independent noise sources. Even with a noise source at 45, an improvement of 5.0 db is achieved. Averaged over the six conditions measured with stationary speech-weighted noise, the 3 Mic condition gives an average additional improvement relative to 1 Mic of 4.5 db (95% CI, 2.7; 6.4 db). In multitalker babble noise, a maximum improvement relative to Omni of 8.7 db (95% CI, 5.3; 12.1

8 418 EAR & HEARING / OCTOBER 2004 Env E O E z I ' PT NH S 0 S 45* S 90 S 180 S ' S ' B 0 B 90 B 180 -r) 0 0. E cc z U) _, L Hi S 0 S 45 S 90 S 180 S ' S B 0 B 90 B 180 Noise configuration a 1 Mic III 3 Mic Fig. 7. Mean improvements of one (1 Mic) and three (3 Mic) active array microphones relative to the omnidirectional microphone of the hearing aid and standard deviations (in db) for six noise configurations with stationary speech-weighted noise (S) and three noise configurations with nonstationary multitalker babble noise (B) for 10 normal-hearing (NH) and 10 hearing-impaired (HI) subjects. db) is obtained in the 3 Mic condition with one noise source at 90 azimuth. Evaluation of 5 Mic The additional perceptual measurements carried out in the second NH group show that extending the array to five active microphones can provide an improvement of 0.8 db (95% CI, 0.1; 1.8 db) relative to 3 Mic with one stationary speech-weighted noise source at 45. With three noise sources at 90, 180, and 270, an additional improvement of 1.2 db (95% CI, 0.4; 2.0 db) is obtained. DISCUSSION Improvements in speech intelligibility in noise obtained with the array relative to an omnidirectional microphone depend on the noise scenario and the subject group. With a single stationary speechweighted noise at 90 azimuth, 1 Mic gives an improvement of 4.2 db averaged over both subject groups. This is comparable to results of other studies of the effect of a directional microphone (Amlani, 2001; Hawkins & Yacullo, 1984; Leeuw & Dreschler, 1991; Wouters et al., 1999). When speech and noise are both presented at 0 azimuth, speech intelligibility in noise with 1 Mic is worse than with the Omni for the NH subjects as well as for the HI subjects for both types of noises. This might be the result of a leveling effect. The delay-and-sum beamformer adds the different microphone signals of the sound that comes from the frontal direction in phase. This results in a higher output for 3 Mic and 5 Mic in comparison with 1 Mic, of approximately 5 and 7 db, respectively. In the prototype array that was used in the study, the gain of 1 Mic was corrected internally for 2 to 3 db but not for the complete 5 to 7 db. The output of 1 Mic of the frontal sound will thus be 3 to 5 db softer than the output of 3 Mic or 5 Mic. At a noise level of 60 db SPL and with the poorer performance of the 1 Mic, one comes close to the baseline performance of the NH listeners, as the speech level for almost perfect speech understanding in quiet is at least 45 db SPL (the SRT for sentences in quiet is approximately 25 to 30 db SPL) (Plomp, 1986). A similar poor result occurs with noise at 180. This will be related to the bidirectional elements of the array. For multitalker babble noise, the SRTs are on average 5 db higher compared with the conditions with speech-weighted noise. As a consequence, the level of the speech was sufficiently high to be above the SRT in quiet. This explains why for both NH and HI subjects, the SRT for 1 Mic has deteriorated by on average 3 db compared with the Omni in conditions with speech-weighted noise at 180 and not with multitalker babble noise. The same effect, although less distinct, can be seen with one noise source at 0. The deterioration of the SRT for 1 Mic compared with the Omni is smaller with multitalker babble noise than with speech-weighted noise. The improvement in speech intelligibility between the 1 Mic and the 3 Mic configuration tends to be

9 EAR & HEARING, VOL. 25 No underestimated by the physical measurements. The DI and the polar diagrams show that the biggest SNR improvement is found between the 1 Mic and the 3 Mic configuration and a smaller improvement between the 3 Mic and 5 Mic configuration. However, the weighted SNR improvements between the 1 Mic and the 3 Mic configuration on the polar diagram are 1.2 db, 2.4 db, and 5.1 db for a single noise source at the angles 45, 90, and 180, respectively (see Fig. 6), against the corresponding SNR improvements of 4.5 db, 7.9 db, and 7.3 db obtained with the perceptual measurements of the NH group and 5.1 db, 5.1 db, and 5.4 db for the HI group. For the difference between the 3 Mic and the 5 Mic, the perceptual results (an improvement of 0.8 db with speech-weighted noise at 45 ) agree well with the physical measurements and with the SNR estimates derived from the polar diagrams (differences relative to 0, see Fig. 6). A difference between the 5 Mic and the 3 Mic of 0.9 db, 0.5 db, and 0.4 db was found for one noise source at 45, 90, and 180 respectively. The free-field DI A./ of the microphone array is 4.0 db, 7.1 db, and 8.5 db for the 1 Mic, 3 Mic, and 5 Mic, respectively. This DI A/ calculation is within 1.5 db, compared with the results of the listening tests with three noise sources at 90, 180, and 270 for the NH and HI subjects. The performance of an array can be improved by changing the type of microphones. Stadler & Rabinowitz (1993) showed that bidirectional microphones give a better directivity (7.4 db) than omnidirectional (3.9 db), cardioid (6.3 db), and hypercardiod (6.8 db) microphones with a delay-and-sum beamformer array of five microphones with a span of cm in endfire configuration. The DI A' of the Delft array (Soede et al., 1993a) with cardioid microphones, calculated from the polar diagrams for 0.25, 0.5, 1, 2, 3, 4, and 5 khz is 7.6 db. The new array with bidirectional microphones comes close to this DI AD with only 3 active microphones. Testing of HI subjects appears to be required in this type of study. The audiometric thresholds and the speech intelligibility of the NH subjects are very similar between subjects. In the HI group, there is a larger variability in degrees of hearing loss and in abilities of speech understanding in noise. In addition to this, the NH subjects all used a Siemens Prisma hearing aid, whereas the HI listeners wore their own hearing aids, which included a variety of models. These factors could explain the bigger diversity (larger standard deviation) in the results of the HI group in comparison to the NH group. Standard deviations range from 1.0 to 4:4 db and from 1.6 to 6.9 db for NH and HI listeners, respectively. The differences between 1 Mic and 3 Mic are in general smaller with multitalker babble noise than with speech-weighted noise, on average 3.9 db for the NH and only 0.6 db for the HI. The difference between NH and HI subjects could be explained by the limited ability of listeners with sensorineural hearing loss to take advantage of temporal and spectral dips that are present in the multitalker background (Festen & Plomp, 1990; Peters et al., 1998). CONCLUSION In this study, a real-time fixed endfire array with one, three, or five microphones to be used in combination with a hearing aid was evaluated physically and by perceptual tests in a realistic reverberant listening environment with NH and HI subjects. The array with three active microphones can provide average improvements in SRTs, relative to an omnidirectional hearing aid microphone, up to 12 db for NH listeners and 9 db for HI listeners. With five microphones, an additional benefit of approximately 1 db can be achieved. With the help of this array, the speech reception in noisy backgrounds of HI subjects is appreciably increased. ACKNOWLEDGMENTS We thank L. Cleuren for her help with the experimental part of this study. This work is supported by the Fund for Scientific Research-Flanders (Belgium), through projects and G We thank Etymotic Research (U.S.A.) and M. Killion for supporting this study. Address for correspondence: H. Luts, Laboratory for Experimental Otorhinolaryngology, Faculty of Medicine, Katholieke Universiteit Leuven, Kapucijnenvoer 33, B-3000 Leuven, Belgium. Received March 12, 2003; accepted March 11, 2004 REFERENCES Amlani, A. M. (2001). Efficacy of directional microphone hearing aids: a meta-analytic perspective. Journal of the American Academy of Audiology, 12, ANSI. (1997). Methods for calculation of the speech intelligibility index. Acoustical Society of America, ANSI S3.5. Beranek, L. L. (1954). Acoustics. New York: McGraw Hill Electrical and Electronic Engineering Series, McGraw Hill. Bilsen, F. A., Soede, W., & Berkhout, A. J. (1993). Development and assessment of two fixed-array microphones for use with hearing aids. Journal of Rehabilitation Research and Development, 30, Desloge, J. G., Rabinowitz, W. M., & Zurek, P. M. (1997). Microphone-array hearing aids with binaural output 1. Fixedprocessing systems. IEEE Transactions on Speech and Audio Processing, 5, Festen, J. M., & Plomp, R. (1990). Effects of fluctuating noise and interfering speech on the speech- reception threshold for impaired and normal hearing. Journal of the Acoustical Society of America, 88, Greenberg, J. E., & Zurek, P. M. (1992). Evaluation of an adaptive beamforming method for hearing aids. Journal of the Acoustical Society of America, 91,

10 420 EAR & HEARING / OCTOBER 2004 Hawkins, B., & Yacullo, W. (1984). Signal-to-noise ratio advantage of binaural hearing aids and directional microphones under different levels of reverberation. Journal Speech Hearing Disorders, 49, Kates, J. M. (1993). Superdirective arrays for hearing aids. Journal of the Acoustical Society of America, 94, Kates, J. M., & Weiss, M. R. (1996). A comparison of hearing-aid array processing techniques. Journal of the Acoustical Society of America, 99, Leeuw, A. R., & Dreschler, W. A. (1991). Advantages of directional hearing aid microphones related to room acoustics. Audiology, 30, Luo, F. L., Yang, J., Pavlovic, C., & Nehorai, A. (2002). Adaptive null-forming scheme in digital hearing aids. IEEE Transactions on Signal Processing, 50, Maj, J. B., Wouters, J. & Moonen, M. (2003). Noise reduction results of an adaptive filtering technique for dual-microphone behind-the-ear hearing aids. Ear and Hearing, in revision. Meyer, J. (2001). Microphone array for hearing aids taking into account the scattering of the head. IEEE Workshop on Applications of Signal Processing to Audio and Acoustics, New Paltz, October 21-24, Peters, R. W., Moore, B. C. J., & Baer, T. (1998). Speech reception thresholds in noise with and without spectral and temporal dips for hearing-impaired and normally hearing people. Journal of the Acoustical Society of America, 103, Plomp, R. (1986). A signal-to-noise ratio model for the speechreception threshold of the hearing-impaired. Journal of Speech and Hearing Research, 29, Plomp, R., & Mimpen, A. M. (1979). Improving the reliability of testing the speech reception threshold for sentences. Audiology, 18, Ricketts, T. (2000a). Directivity quantification in hearing aids: fitting and measurement effects. Ear and Hearing, 21, Ricketts, T. (2000b). The impact of head angle on monaural and binaural performance with directional and omnidirectional hearing aids. Ear and Hearing, 21, Ricketts, T., & Henry, P. (2002). Evaluation of an adaptive, directional-microphone hearing aid. International Journal of Audiology, 41, Saunders, G. H., & Kates, J. M. (1997). Speech intelligibility enhancement using hearing-aid array processing. Journal of the Acoustical Society of America, 102, Soede, W., Berkhout, A. J., & Bilsen, F. A. (1993a). Development of a directional hearing instrument based on array technology. Journal of the Acoustical Society of America, 94, Soede, W., Bilsen, F. A., & Berkhout, A. J. (1993b). Assessment of a directional microphone array for hearing-impaired listeners. Journal of the Acoustical Society of America, 94, Soede, W., Bilsen, F. A., Berkhout, A. J., & Verschuure, J. (1993c). Directional hearing aid based on array technology. Scandinavia Audiology Supplement, 38, Stadler, R. W., & Rabinowitz, W. M. (1993). On the Potential of Fixed Arrays for Hearing-Aids. Journal of the Acoustical Society of America, 94, Thompson, S. C. (1999). Dual microphones or directional-plusomni: which is the best? The Hearing Review, 3, Vanden Berghe, J., & Wouters, J. (1998). An adaptive noise canceller for hearing aids using two nearby microphones. Journal of the Acoustical Society of America, 103, Versfeld, N., Daalder, L., Festen, J. M., & Houtgast, T. (2000). Extension of sentence materials for the measurement of the speech reception threshold. Journal of the Acoustical Society of America, 107, Wouters, J., Litiere, L., & van Wieringen, A. (1999). Speech intelligibility in noisy environments with one- and two- microphone hearing aids. Audiology, 38, Wouters, J., & Vanden Berghe, J. (2001). Speech recognition in noise for cochlear implantees with a two- microphone monaural adaptive noise reduction system. Ear and Hearing, 22,