SPEECH RECOGNITION IN SKI SLOPE SENSORINEURAL HEARING LOSS Estudo do reconhecimento de fala nas perdas auditivas neurossensoriais descendentes

(1) Universidade Federal de Minas Gerais (UFMG), Belo Horizonte, Minas Gerais, Brazil. (2) Universidade Federal de Minas Gerais (UFMG), Belo Horizonte, Minas Gerais, Brazil. (3) Speech-language pathology and Audiology course, Universidade Federal de Minas Gerais (UFMG), Belo Horizonte, Minas Gerais, Brazil. Conflict of interest: non existent Because consonants are sounds of lower intensity than vowels, they become more difficult to be detect, especially by individuals with hearing impairment, who, consequently, have difficulties in discriminating words3. Hearing losses are classified according to the pure-tone hearing thresholds for air and bone conduction and with the tests of speech. These make it possible to define the type and degree of hearing loss, in addition to measuring speech discrimination4. The sensorineural hearing losses deriving from lesions in the cochlea and/or cranial nerve VIII (vestibulocochlear nerve) up to the brainstem auditory nuclei and are characterized by alterations in the hearing thresholds of air and bone conduction4. Biochemical alterations, failure in the cellular mechanism of the cochlea or diseases of the inner ear and retrocochlear auditory pathways, or even alterations caused by aging, can lead to  INTRODUCTION


INTRODUCTION
In order for verbal communication to be efficient, good speech reecognition is crucial.This depends on the acoustical characteristics of the words and good auditory perception, as well as suprasegmental features.Vowels, whose range of frequency varies between 0.4 and 0.5 kHz, concentrate the greatest amount of acoustical speech energy, while intelligibility relies on consonant sounds, which have a sound spectrum with frequencies above 2 kHz, and less distribution of acoustical energy 1,2 .
To be included in the study sample, the test records should belong to individuals older than 18 years, with a type A tympanometric curve 13 , and ski slope sensorineural hearing losses 14 of a mild to severe degree 15 .The hearing loss was regarded as "ski slope" when the difference between the means of the frequencies of 0.25-2 kHz and 3-8 kHz was greater than15 dBHL 14 .Cases of conductive and mixed hearing loss as well as sensorineural hearing loss with flat or ascending audiometric configurations were excluded.In addition, cases with incomplete data such as the lack of acoustic immittance or speech audiometry results were excluded.
The study sample included the audiometric evaluation of 55 ears of 19 men and 11 women aged between 26 and 91 years (mean, 66.5 years; median, 71 years).Because these patients were regularly seen at the Audiology Outpatient Clinic, it was possible to obtain their informed consent with explanations on the theme of the study and its aims.
The values found in the SRPI and SRT studies were correlated with the following averages for the pure-tone hearing thresholds: • Average 1: mean of the frequencies of 0.5, 1 and 2 kHz • Average 2: mean of the frequencies of 0.5, 1, 2 and 3 kHz • Average 3: mean of the frequencies of 0.5, 1, 2, 3 and 4 kHz • Average 4: mean of the frequencies of 0.5, 1, 2 and 4 kHz In addition, the SRPI values were correlated with the differences, in decibels, of the pure-tone hearing thresholds between frequency octaves across the following ranges: For the statistical analysis, the level of significance was set at 0.05 and the confidence intervals were constructed with 95% of statistical confidence.Since the sample data were quantitative and continuous, parametric statistical tests were used.Pearson's correlation test was used to estimate the degree of correlation of SRT and SRPI with the differences of thresholds between the frequency octaves and audiometric averages.To validate the analyses, the correlation matrix pictured below was used: compromised sound transduction, which, in turn, will impair an individual's hearing and speech comprehension 5 .
In sensorineural hearing losses, high frequency hearing thresholds are proportional to the extent of damage to the ciliated cells at the base of the cochlea 6 .Losing the cilia of the internal ciliated cells in that region leads to a reduction or absence of afferences 7 .In these cases, the input of acoustical energy in that frequency region, which is contributive to speech intelligibility, is compromised-among other factors, by an increase in the amplification of the sound energy 8 .For that reason, it is believed that the greater the deficit in the high frequencies in relation to the low frequencies, the worse the word recognition performance.
Speech recognition and the use of amplification in the different audiometric configurations is an issue that has been debated by several authors.One study demonstrated that in the hearing losses with the greatest impairment in the low frequencies, speech recognition improves regardless of the degree of loss when that region is amplified 9 .On the other hand, for more severe hearing losses with a flat configuration, amplifying the low frequency regions is more beneficial to intelligibility than amplifying the high frequency regions 10,11 .In ski slope hearing losses, the amplification over the range of high frequencies is very important for speech recognition 8,10,12 .
The objectives of the present study were (a) to assess whether the increase in the difference of the pure-tone hearing thresholds between the frequency octaves interfere with speech recognition, and (b) to assess whether hearing losses in the frequencies of 3 kHz and 4 kHz influence the results of speech recognition tests.

METHODS
The present retrospective and comparative study was approved by the Research Ethics Committee of the Universidade Federal de Minas Gerais under Opinion ETIC 0653.0.203.000-10.
The study was based on secondary data and involved a review of the medical records of the patients who were seen at the Auditory Health Care Service of the UFMG Hospital das Clínicas, Anexo São Geraldo, from March through July 2011.The audiometric tests of patients with sensorineural hearing loss showing a ski slope audiometric configuration, candidates for hearing aids, were selected.The following tests were analyzed: pure-tone audiometry (study of the speech recognition threshold -SRT-and study of the speech between the frequency octaves (in decibels), the SRT and the SRPI (in percent values).
Table 2 depicts the correlations between the SRT and SRPI values with the hearing threshold differences between the frequency octaves.
Table 3 shows the correlations established between the tone averages and SRT, and SRPI.

RESULTS
Table 1 shows the descriptive analysis of the quantitative variables including the averages of the hearing thresholds by frequency, the tone averages, the differences of the pure-tone hearing thresholds

DISCUSSION
The SRPI and SRT values were correlated to the averages of the pure-tone hearing thresholds, and the SRPI, to the differences of pure-tone hearing thresholds between frequency octaves in order to assess how frequencies of 3 and 4 kHz influence speech recognition tasks.
For the variables described in Table 1, and analyzing the average results found by frequency level, a progressive increase was noted in the threshold as the tone frequency was increased, a characteristic of hearing losses with a sloping audiometric configuration.Regarding the analysis of the differences of the pure-tone thresholds between the frequency octaves, the increase in these values leads to sharper drops in the audiometric curve.Thus, a more marked fall was observed between the frequency ranges of 0.5 -2 kHz and 1 -3 kHz.
The average value found for SRT is consistent with Average 1 (0.5, 1 and 2 kHz).However, when the frequencies 3 kHz and/or 4 kHz were included in this average, an increase of 7, 11 and 8 dB HL was noted in Averages 2, 3 and 4, respectively, which render them inconsistent with the SRT value.This finding corroborates the literature, which reports that high frequencies outside the range of 0.5 -2 kHz are less relevant to the SRT 16 .
In the overall sample, the mean SRPI value was low (56.4%).The great number of errors made in the speech test was likely due to an impairment over the high frequencies.The literature reports that individuals with sensorineural hearing losses for high-pitched sounds have difficulties in speech recognition due to decreased acoustic information 3 .However, the present study encompasses an aging population, with a mean age of 66.5 years and median of 71 years.This may have had an influence on the poor levels of speech recognition.Presbycusis, a form of age-related hearing loss, is often associated with difficulties in inteligibility due to the organic and physiological changes that take place in the auditory system over the years.
In addition, cognitive aspects should be taken into consideration, since this population frequently present cognitive alterations worsened by the hearing deficits.
The analysis of the data on the correlation between the SRT and the differences between the octaves showed a statistically significant (p= 0.002) moderate correlation (-44.1%) for the difference between intervals 1-2 kHz.This means that these variables are inversely proportional, i.e., the greater the difference between the frequencies, the worse the SRT values.Regarding the other ranges, no significant correlations were found.With respect to the SRPI, there was no statistically significant correlation with the differences between the frequency octaves (p > 0.05) (Table 2).
Table 3 correlates the averages of thresholds to the SRT and SRPI values.The p-value was statistically significant across correlations.Regarding SRT, a very good correlation (>80%) with the analyzed averages was found.Pearson's correlation was 93% for Average 1, i.e., the frequencies 0.5 kHz, 1 kHz, 2 kHz contribute significantly to the speech thresholds.The average of those frequencies is used in most classifications of the degree of hearing loss 16,17 .However, when the frequencies of 3 kHz and 4 kHz were added, this correlation was reduced, although it remained very good according to the scale used in the present study.
Regarding SRPI (Table 3), a good correlation was found with the averages in this study, except for Average 1 (0.5, 1 and 2 kHz), whose correlation was moderate.This result shows that the inclusion of the frequencies 3 and 4 kHz in the tone average improves the correlation with the SRPI values and, consequently, reinforces its importance for speech recognition.
One study correlated the average for the speechrelated frequencies (0.5 kHz, 1 kHz and 2 kHz) and the average of 3, 4 and 6 kHz with the Lists of Sentences in Portuguese (LSP) test.Statistically significant relationships were found only with the first average.However, according to the authors, sounds and thus the actual impairment in communication caused by those losses is not described 17 .

CONCLUSION
In the present study, the differences in the pure-tone hearing thresholds between the frequency octaves, i.e., the rise in the audiometric curve slope had no significant influence on the speech recognition task (SRPI).Speech recognition as evaluated by the SRT and SRPI tests shows very good correlation with the averages of the frequencies between 0.5 kHz and 4 kHz.The inclusion of the frequencies 3 kHz and 4 kHz in the tritone average of speech (0.5, 1 and 2 kHz) was important for the determination of the speech recognition percent indices.

ACKNOWLEDGMENTS
We are grateful to the speech-language pathologists and audiologists at the Audiology Outpatient Clinic of the Hospital de Clínicas, Universidade Federal de Minas Gerais, for their assistance in the data collection; to the speech-language pathologist and audiologist Letícia Penna, for reading the manuscript and helping with suggestions, and to the professionals who assisted in the statistical analysis and preparation of the abstract.
this does not imply that the frequencies 3, 4 and 6 kHz have no bearing on speech recognition; rather, it means that there are factors that should be considered in addition to the audibility of those frequencies 18 .
Another study confirmed that high frequency acoustic information-above 3 kHz-could be quite useful in terms of speech comprehension for people with flat sensorineural hearing loss and in high frequencies up to 70 dBHL 19 .
Although other studies highlight the importance of the frequencies 0.5 kHz to 2 kHz, it cannot be affirmed that frequencies below 0.5 kHz and above 2 kHz are not imortant for speech recognition.In the present study, both 3 kHz and 4 kHz were found to be important for speech discrimination.Other studies also indicated that the frequencies between 4 and 6 kHz contribute to the recognition of consonants 20,21 .

Table 2 -Correlation of the speech recognition threshold (SRT) and the speech recognition percent index (SRPI) with the differences in hearing thresholds (in decibels) between the frequency octaves.
Correlation test, p<0.05.(*) moderate correlation, (**) poor correlation -accoding to the classification scale.