Authors Simonsen1
Christian Stender Jens-Christian Britze Kijne1 Lise Bruun Hansen1 Anne Specht Petersen1 Søren Laugesen2 Filip Marchman Rønne2 Niels Søgaard Jensen2 Oticon A/S Kongebakken 9 DK-2765 Smørum www.oticon.com 1
2
Eriksholm Research Centre Rørtangvej 20 DK-3070 Snekkersten www.eriksholm.com
_____________ Contact Christian Stender Simonsen
[email protected]
• 10 kHz bandwidth
Adaptive Speech-Reception Threshold (SRT) measures are popular for good reasons. They do, however, have drawbacks related to the unbounded nature of the Signal-to-Noise Ratio (SNR) at which the SRT is achieved [1].
(reference)
• 2.5 kHz bandwidth
SII suggests a 20% difference in speech recognition between these settings.
(LACK OF) ECOLOGICAL VALIDITY: Often the SRT is much lower than the SNR found in realistic listening conditions [4,5]. If the test involves aided listening, the hearing aid may therefore be subjected to conditions for which it was never intended. This has the potential to cause misleading results.
TARGET SPEECH: Danish HINT sentences [8], all consisting of five words presented from 0°. Target level fixed at 70 dB SPL (C). MASKER SPEECH: Running speech (reading from a fairytale), speech pauses cut down to 65ms. Two female or Figure 1: Sketch of the loudspeaker setup. two male talkers, used in pairs symmetrically around the listener at three different angle pairs (see Figure 1): • +/- 45° • +/- 30° • +/- 15° WORD or SENTENCE scoring
TEST CONDITIONS: • 3 masker positions, word or sentence scoring, female/male maskers adds up to 3 x 2 x 2 = 12 different test conditions. • However, only 4 test conditions selected [6] to span test adequately: [15mS], [30mS], [30mW], [45fW] o f and m denote female and male masker o S and W denote sentence and word scoring o Numbers ‘15’, ‘30’ and ‘45’ denote masker azimuth angles
10 11 12
Training Test Retest
1.9 dB (±1.6)
6 -0.7 dB (±1.0)
2
30mW
FixedIndividual SNR SFS Change HA setting Individual FixedSFS SNR
6
1 Reference / 1 Lowpass 1 1T Reference / 1 Lowpass 1
2
-2.8 dB (±1.8)
4.1 dB (±1.2)
0
4.0 dB (±0.8)
0 2.4 dB (±1.8)
-2 -0.3 dB (±1.3)
-4
–
Lowpass mean: 57%
Current effect: F(1, 13)=23.001, p=.00035 Vertical bars denote 0.95 confidence intervals Include: v2='+3dB' AND (v4='Test') 80 75 70 65 60 55
Table 3: Effect table for the experimental contrast
50 45 Reference
Mean Mean±SD Raw Data
-6 45fW
30mW 30mS SFS condition
Mean Mean±SD Raw Data
-4
15mS
HA setting
-6 30mS
15mS
SFS condition
15mS
Although we have a +3-dB shift of SRT for 30mS the current validation compared to the 30mW previous validation, the relative effects are 45fW comparable.
Nominal SRT shift
Discussion FIVE TEST SUBJECTS WERE ‘LOST’ because the initial target SNR of 0dB SNR turned out to be too low, which was proposed based on the previous validation [7]. The SNR difference was probably based on a difference in reverberation in the two test facilities. The first validation study was performed in an anechoic sound studio, and the present study used an EBU3276 [9] compliant studio (standard listening room). A +3-dB difference between the two validation studies was found.
THE BEAUTY OF THE SFS TEST for hearing aid evaluations lies in that all listeners are evaluated at the same SNR. However, the method of obtaining same SNR across listeners does come at a cost: frontal plane, where directionality processing in hearing aids are often not active. Nevertheless, using different masker locations might introduce other hearing aid processing differences.
• Although female and male maskers were spectrally matched, it still sounds
Table 2: Effect on SRT between baseline and test trials SFS condition
Lowpass
• All masker locations used in the SFS test are situated in a narrow beam in the
-2
30mW
Reference mean: 72%
Figure 5:Fixed-SNR measurements of the two settings used in test.
So even though we have some flexibility in the SFS test, we still need to have a good indication of target SNR for all test subjects before the experiment.
Previous validation
4
-2.5 dB (±1.7)
45fW
Method
Baseline Test Retest
Figure 2: Effect on SRT between baseline and test trials – current validation
4
To validate the SFS test with emphasis on: 1. Do the four SFS conditions change test difficulty as expected? 2. What is the reliability of the SFS test? 3. Can the SFS test measure an expected contrast?
7 8 9
SRT MANIPULATOR EFFECTS re ‘Baseline’ is evaluated by examining the differences between baseline (30mW) and Figure 3: Effect on SRT between baseline the other adaptive-SNR conditions (runs 2-6). and test trials – previous validation
8
–
1 1 1
Results
SFS TEST: To address these problems, the SFS (Spatial Fixed-SNR) speech-inspeech intelligibility test is proposed. It uses a fixed-SNR paradigm. Percentcorrect scores within the informative 20-90% range are obtained for the individual by selecting among four test conditions with different test difficulty [6]. Thus, the SFS test is aimed at within-subjects comparisons. The SFS Test was validated in a previous experiment [7]. The current experiment is set up with different experimental contrast and using different test facilities (reverberant versus anechoic facility).
Aim of the study
Reference
SNR
VALIDITY is evaluated by examining the experimental contrast in the fixed-SNR paradigm. It was expected to find a mean of the reference of 70% correct and 20% below for the low-pass version. This corresponds to our findings:
Break
(low-passed version)
SNR CONFOUNDS: Aided hearing-impaired listeners often show a wide spread in SRT. Therefore, the hearing aids under test will be subjected to very different SNRs among different listeners. These differences in SNR can affect hearing-aid signal processing and can in turn potentially confound the test results [2,3].
SRT B manipula C tor test D
4 5 6
Results, continued
% correct
EXPERIMENTAL CONTRAST: The SFS test is ideal for comparing different settings. In the previous validation Table 1: Validation study protocol. The order of runs 3,4,5,6 experiment, the experimental as well as blocks {7,8,9} and {10,11,12} were balanced across contrast did not behave listeners. as expected. In the current Run Trial SFS Paradigm HA setting #HINT type conditio lists validation, Oticon Alta Pro n RITE hearing aids were 1 Training 30mW 2T programmed with two 2 Baseline 30mW 1 different settings: 3 A 1 Adaptive-
Background
SPECTRAL MATCHING: all targets and maskers were spectrally matched to a female speech spectrum.
Presented at the International Hearing Aid Research Conference IHCON 2014 Granlibakken, Tahoe City August 13-17, 2014
compared to previous validation (see discussion) !
SRT shift (dB)
_____________
• However 5 TPs not part of fixed-SNR analysis due to +3-dB adjustment
SRT shift (dB)
The Spatial FixedSNR (SFS) test: presentation and validation
A novel speech intelligibility test is presented, in which individual test subjects can be evaluated in different conditions with varying difficulty, such that the test signal-to-noise ratio (SNR) is the same for all subjects.
LISTENERS: N=19 (10 males and 9 females) hearing impaired listeners with sensorineural hearing loss and average age 71 participated. Pure tone average (4PTA) hearing losses ranged from 34 to 60 dB HL
Current Previous validation validation SRT shift SRT shift
+5.0 dB
+4.1 dB
+4.0 dB
+2.5 dB
+1.9 dB
+2.4 dB
0 dB
-0.7 dB
-0.3 dB
-2.5 dB
-2.5 dB
-2.8 dB
SRT MANIPULATOR EFFECTS – SRT vs 4PTA: The effect of hearing loss on SRT was examined by regression Figure 4: Effect on SRT vs. 4PTA. analysis of the four test conditions as a function of 4PTA in Figure 4. The comparable slope of the four lines, indicates that the SRT manipulators work effectively across PTA.
like a male/female. Thus, temporal differences still exists in some form.
• Running an SFS test takes more time than running an adaptive SRT test and
requires a priori knowledge of target SNR.
TRANSLATION OF THE SFS TEST to other languages than Danish should be almost straightforward, since the HINT material is available in several languages. However number of words in each sentence might cause design challenges. The Danish HINT material exclusively includes sentences with five words.
Conclusion The Spatial Fixed-SNR (SFS) test was validated (again): 1. The four proposed SFS conditions were able to change test difficulty, such that informative %-correct data could be measured for all listeners 2. The reliability of the SFS test was found to be 8.5%, which apparently is lower than for the standard HINT. Furthermore, the result corresponds almost spot-on to the result from the first validation. 3. With the SFS test a 15% highly significant difference was detected between the reference and the lowpass setting. The experimental contrast was created with an expected 20% difference as target.
TEST RELIABILITY is calculated by computing test-retest standard deviations for the Fixed-SNR results: •
SDfixed-SNR = 8.5%
•
SDfixed-SNR = 8.6% in previous validation
For standard Danish HINT a test retest SD of 0.92 dB was found[8]. The slope of the psychometric function for the SFS test was estimated in [6] to be 13.7%/dB yielding 0.62 dB. Thus, potentially better reliability for the fixed-SNR paradigm.
REFERENCES [1] Naylor G (2010). Limitations of Speech Reception Threshold (SRT) as an outcome measure in hearing-aid research. International Hearing Aid Research Conference (IHCON), Lake Tahoe, CA, USA, oral presentation. [2] Bernstein JGW (2011). Controlling signal-to-noise ratio effects in the measurement of speech intelligibility in fluctuating maskers, In: Dau T, Jepsen ML, Poulsen T, Dalsgaard JC. (Eds.) Speech Perception and Auditory Disorders. Proceedings of the Third International Symposium on Audiological and Auditory Research, ISAAR. Centertryk A/S: Denmark. [3] Naylor G, Burmand Johannesson R (2009). Long-term signal-to-noise ratio at the input and output of amplitude-compression systems. J. Am. Acad. Audiol. 20, p. 161-171. [4] Pearsons KS, Bernett RL, Fidell S (1977). Speech levels in various noise environments. Bolt, Beranek and Newman. Inc. Canoge Park, California. [5] Smeds K, Wolters F, Rung M (2012). Realistic signal-to-noise ratios. International Hearing Aid Research Conference (IHCON), Lake Tahoe, CA, USA, poster presentation. [6] Rønne FM, Laugesen S, Jensen NS, Hietkamp RK, Pedersen JH (2013). Magnitude of speech-reception-threshold manipulators for a spatial speech-in-speech test that takes signal-to-noise ratio confounds and ecological validity into account. ICA 2013 Montreal, Canada, vol. 19, 1–8. [7] Laugesen S, Rønne FM, Jensen NS, Sorgenfrei MG (2013). Validation of a spatial speech-in-speech test that takes Signal-to-Noise Ratio (SNR) confounds into account , ISAAR proceedings. [8] Nielsen JB, Dau T (2011). The Danish hearing in noise test. Int. Journ. Aud. 50, p. 202-208. [9] EBU (1998) Technical Document Tech 3276: Listening Conditions for the Assessment of Sound Programme Material: Monophonic and Two-Channel Stereophonic, 2nd ed. European Broadcast Union.