Lombard effect, intelligibility, ambient noise and willingness to spend time and money in a restaurant in older people


With protocol approval from the Office for the Protection of Research Subjects Review Board at the University of Illinois Urbana-Champaign (IRB #19479), a total of 31 people were recruited for the study. The experiment was performed in accordance with guidelines and regulations suggested by the University of Illinois Urbana-Champaign Office for the Research Subject Protection Review Board. Inclusion criteria required that the participant be over the age of sixty and have normal hearing ( 25dB HL); moreover, participants must be proficient English speakers, since the intelligibility test will be given in English and non-smokers. Participants signed an informed consent form, completed a demographic data form, and underwent a hearing test before the end of the experiment. Hearing testing included otoscopy, tympanometry, speech reception thresholds, and air and bone conduction audiometry. An initial exclusion criterion, in order to continue the Lombard effect part of the study, involved pure tone air conduction results showing asymmetries greater than 10 dB HL, at two or more consecutive frequencies, or a deviation air-bone greater than 15 dB HL, without prior medical consultation with an otolaryngologist. Among the 31 participants (14 men and 17 women), 10 had normal hearing, 10 mild HL and 11 moderate to severe HL.


Once participants confirmed their age, hearing screening was performed, including hearing history review, otoscopy, tympanometry, air conduction pure-tone audiometry with 250 insertion phones at 8000 Hz (with inter-octaves if necessary), speech reception thresholds in quiet, and pure tone thresholds in unmasked or masked bone conduction. Over-ear headphones were used to reconfirm pure tone air conduction deviations of 10 dB HL or more between the ears at two or more consecutive frequencies.

After completing the hearing test, the individual continued with the second part of the experiment. This included in order the Lombard effect test, the speech intelligibility in noise test and scored on a visual analog scale their perceptual responses of (1) communication disturbance, (2) willingness to spending time and (3) willingness to spend money. In a restaurant. Participants completed all components of the study in one or two forty-five to sixty minute sessions, resulting in scheduling conflicts and/or participant fatigue.

Categorization of hearing loss

After the hearing assessment, the participants were grouped into three categories. These categories were determined by the collective averages of pure tones of 2000 Hz, 4000 Hz and 8000 Hz from both ears per participant. The average right and left ear audiometric thresholds by participant category are shown in Figure 4. The calculated average high frequency was categorized into: (1) normal hearing [<25 dB HL](2) mild sensorineural hearing loss [26–40 dB HL]3) moderate, severe and profound sensorineural hearing loss [41+ dB HL]. Categorization was based on ASHA39 and Clark40 degrees of hearing loss.

Figure 4

Average hearing thresholds by hearing loss category for left and right ear.


The Lombard effect Participants were provided with the head-mounted microphone connected to a portable recorder. Then, the participants were asked to read in the presence of a pre-recorded restaurant noise the first six sentences of “The Rainbow Passage”22, which is used by speech therapists to assess and monitor voice and articulation disorders. Pre-recorded restaurant noise included the babbling of real restaurant patrons and the clinking of crockery and cutlery. Noise was presented at eleven levels from 35 dB(A) to 85 dB(A) with a 5 dB interval in random order. Participants were instructed, by the undergraduate student assistant, with the following: “Every time I [the listener] want you to pretend we’re talking in a restaurant and tell me the story. Make sure I understand you equally well every time.

Speech intelligibility in noise Participants performed a speech intelligibility in noise test for each of eleven random noise conditions. Our intelligibility test was designed using the 250–300 Northwestern University 6, NU-6 standard41, words accessible to all audiologists. These words were recorded using the voice of an undergraduate student assistant; the words were cut individually and then randomized into eleven lists of twenty words. These recordings were then played through a Head and Torso Simulator (HATS) (GRAS, Holte, Denmark), placed at a distance of one meter from the participant, to ensure consistency of the speech material. Word lists were presented at a level of 60 dB(A), corresponding to normal vocal effort at one meter distance42. Participants had to repeat the word spoken by the HATS. Intelligibility scores were measured as percentage of correct words.

Disruption, time and budget scores Self-reported communication impairment and willingness to spend time and money were measured on paper-based visual analog scales. The participant was instructed to place a vertical tick on the horizontal line regarding their amount of disturbance, time, and money per noise condition. The paper document was then scanned and uploaded for measurement, in centimeters, using Adobe Acrobat Pro DC software and the Adobe Software Measurement Tool (version 2022.001.20085, https://www.adobe. com/). The responses were then recorded in an Excel spreadsheet according to the participant’s identification number.

Analysis of voice recordings

Participant audio recordings were segmented using ELAN software (version 5.8, https://archive.mpi.nl/tla/elan)43. Each of the eleven segments of the “Rainbow Passage” was recorded according to participant ID number and noise level condition. MATLAB (R2017a) was used for speech signal analysis. In each condition, the equivalent SPL was measured. For each condition, the mean SPL value was obtained per subject. For each subject, the average SPL among the conditions was calculated and subtracted from each average SPL value for that subject (called Delta SPL). This intra-subject centering was performed in order to evaluate the variation of the subject’s vocal behavior in the different noise conditions compared to his typical vocal behavior (mean value of SPL per subject). Each participant was identified by ID number, gender, age, and type of hearing loss. As a first preliminary assessment of the reliability of the recordings, the voice-to-noise ratio (VNR) was evaluated. To assess the VNR, a mixture of Gaussian models was used. In statistics, a mixture model is a probabilistic model for representing the presence of subpopulations within an overall population, without requiring a set of observed data to identify the subpopulation to which an individual observation belongs. In our case, the time history of all the SPLs recorded by the microphone represents the global population. The two subpopulations are voice levels and noise levels. The overall population distribution was modeled as the sum of two Gaussian distributions, where two mean values ​​represented the mean voice and noise levels. The average VNR among the different noise conditions was 21.2 dB with a standard deviation of 2.0 dB. In the worst case (VNR = 19.3 dB), the contribution of the background noise on the overall level (noise and voice) was around 0.05 dB. This result confirms that the effect of noise on the equivalent level was acoustically negligible.

Equipment and room measurements

The study was carried out in a standardized soundproof cabin44. Reverberation time was measured in the sound booth from impulse responses (IR) generated by balloon pops45. The four IRs were recorded in two source positions and two microphone positions using an NTI Measurements M2211 microphone (Class 1 frequency response) and analyzed in third-octave bands using an audio analyzer and NTI XL2 acoustics. The reverberation time (T20) at mid frequencies in the room was 0.05 s, while the background noise was 22.5 dB(A). The hearing test will be performed using a calibrated tympanometer (TympStar, Grason-Stadler, Eden Prarie, MN), audiometer (GSI 61, Grason-Stadler, Eden Prarie, MN), telephone inserts ( 3M EA-RTONE 5A 410-5002), a TDH-50 headset and a bone oscillator. Participants’ speech will be recorded using a head-mounted microphone (Beta 54 WBH54, Shure, Niles, IL) and connected to an audio interface (US-, TASCAM [DR44], Montebello, California). The same restaurant noise was emitted from two directional speakers (100 dB High Performance Speaker, E3 Diagnostics, Arlington Heights, IL) placed 45 to the participants’ sides.

Comments are closed.