ucsb_henley_gate_post

Overview for the 2017 SPIN Workshop

Here you can read Eriksholm researchers' contributions to the SpiN workshop in Oldenbrug, Germany, January 5-6, 2017.

Learning effects in the Danish HINT

Bramsløw L1, Simonsen LB2, Hischou ME1,3, Hashem R1,3, Hietkamp RK1

1) Eriksholm Research Centre, Rørtangvej 20, DK-3070 Snekkersten
    labw@eriksholm.com  +45 4829 8928    rhie@eriksholm.com  +45 4829 8921
2) University of Copenhagen, Faculty of Humanities, Karen Blixens Vej 4, 2300 Copenhagen S, Denmark
3) University of Southern Denmark, Faculty of Humanities, Campusvej 55, 5230 Odense M, Denmark

Abstract

The benefit from hearing aid signal processing can be tested in the laboratory by means of a speech test, such as the open-set Hearing In Noise Test (HINT, Nilsson et al, 1994). The Danish HINT (Nielsen & Dau, 2011) consists of only 200 test sentences organised in 10 lists. These HINT lists are used heavily in the newly developed Competing Voices Test (Bramsløw et al, 2016) in which pairs of sentences are presented. Thus, reuse of the material will occur multiple times within visits. This can lead to learning of the sentences and possibly confound the test outcome. The change in performance over time has two components; 1) a practice effect from becoming more trained in the listening task (or alternatively, fatigue) and 2) a learning effect due to memorization of the specific sentences.

This poster presents practice and learning effects of the Danish HINT sentences from two experiments. In experiment 1, we used varying degrees of exposure at 3 visits with 3 weeks interval; selected lists were reused 1-9 times altogether. Ten elderly listeners were tested using a default adaptive HINT procedure aiming at 50% correct sentence score in speech shaped noise. Results showed a maximum within visit learning effect of 1.5 dB and a between visit learning effect of 1.1 dB. These effects can be attributed mainly to learning, as practice effect across varying lists was not statistically significant.

In experiment 2, we investigated the effect of exposing at 80% sentence score vs 50% score in two visits with 2 and 4 repetitions of lists. This test used 15 elderly listeners. While the previous learning effects were confirmed, there was no added learning due to the 80% sentence score exposure with its higher word recognition.

The implications of the results for future similar speech tests are discussed. Clever experimental design is proposed to compensate for practice and learning effects.

Find the poster here

Effect of low-frequency gain on speech intelligibility and sound quality in a competing voices situation

Lochner J1, Santurette S1,2, MacDonald E1, Bramsløw L3

1) Hearing Systems Group, Department of Electrical Engineering, Technical University of Denmark
2) Department of Otorhinolaryngology, Head & Neck Surgery and Audiology, Rigshospitalet
3) Eriksholm Research Centre, Oticon A/S

Abstract

In hearing aids, low-frequency (LF) gain in the octave bands centered at 125, 250, and 500 Hz is often reduced either because of open fittings or to minimize upward spread of masking. However, pitch cues extracted from LF resolved harmonics are considered important for speech intelligibility (SI) in competing voices situations. Furthermore, the sound quality is typically described as more natural with higher LF gain. Here, the effects of modifying LF gain on SI and sound quality for a female target voice masked by either a same-sex or different-sex masker voice were assessed.

Ten young normal-hearing (NH) listeners and 11 older hearing-impaired (HI) listeners with sensorineural and moderate LF hearing loss participated. Two conditions with an increase in LF gain of either 4 dB or 8 dB relative to the prescribed-gain condition were tested. In the prescribed-gain condition, the HI listeners were fitted individually on both ears using a linear CAMEQ rationale, while no gain was applied for the NH listeners. A self-scored closed-set Danish matrix-type SI test (Dantale II) was used. The gain was applied directly to the stimuli and the speech stimuli were presented via headphones. The percentage of correctly identified words was evaluated at four fixed signal-to-noise ratios (SNRs). All NH listeners were measured at the same absolute SNR while all HI listeners were measured at constant SNRs relative to their individual speech reception thresholds.

For NH listeners, no significant effect of increased LF gain on SI was found. At 0 dB SNR in the prescribed-gain condition, the HI listeners scored significantly lower with the same-sex versus different-sex masker, whereas the scores were similar for the two maskers at SNRs different from 0 dB. The HI listeners did not benefit either from higher LF gains in terms of SI. However, an interaction between LF gain and masker type (same sex versus different sex) was found. This is because a 4-dB increase in LF gain was beneficial in the different-sex masker condition but detrimental in the same-sex masker condition.

A subjective quality rating of the different LF gain conditions was also performed by the same HI listeners using the speech and masker material from the SI test at 0 dB SNR. The quality ratings were significantly higher with increased LF gain compared to the prescribed-gain condition.

Overall, the results suggest that a higher LF gain does not affect SI but may be beneficial in terms of sound quality.

Find the poster here

The Bimodal Fusion Questionnaire: Subjective Self-Assessment of Binaural Fusion

Janssen NA1, Bramsløw L2, Riis SK3, Marozeau J1

1) Hearing Systems Group, Department of Electrical Engineering, Technical University of Denmark
2) Eriksholm Research Centre, Oticon A/S
3) Department of Mathematical Modelling, Technical University of Denmark

Abstract

In recent years, implantation criteria for cochlear implants have changed and we see more and more so-called bimodal patients wearing a cochlear implant (CI) on one side and a hearing aid (HA) on the other side. Bimodal users thus listen with two devices that deliver sound in a very different manner due to their working principles. The HA and CI devices are often manufactured by different companies, use different processing strategies and are typically fitted by different professionals. Today, no universally accepted bimodal fitting scheme is available and many patients experience two sound percepts that are not matched in pitch and loudness.

From this the question arises, whether bimodal users fuse sounds from both ears, like normal hearing persons do. Are they perceiving just one sound object or do they perceive a more complex acoustic scene with two simultaneous sound objects?

In order to assess this question, 9 bimodal users were interviewed based on a self-assessment questionnaire. This questionnaire was divided into two parts: First, their auditory experiences such as speech perception in noise, distance perception and listening effort were assessed using the SSQ5 questionnaire (see Mertens 2013). Then, they were asked to rate the difference in perception for sounds of a single sound source across their two ears and additionally rate a possible difference in pitch and loudness, as well as preference for localization and sound quality. The latter was rated for six everyday situations, each in the context of having a conversation: Quiet, a busy restaurant, nature with wind, a busy street at rush-hour, listening to music and shopping in a supermarket.

The everyday situations do not yield significantly different results and are therefore considered together. Altogether, the answers indicate that a 4 out of 9 participants report to perceive a single fused sound object despite significant differences across ears in pitch and loudness. While this does not seem to affect perceived speech perception in noise or localization abilities, bimodal users that report a low difference in perception also report less cognitive effort.
More data will be collected in order to confirm the bimodal distribution of the two groups perceiving either fusion or no fusion

Find the poster here