Greenwood Function Research Articles

Feature normalization for robust individual identification of the ortolan bunting (emberiza hortulana L)

This paper presents the use of cepstral moment normalization for the purpose of improving the accuracy of individual identification in the ortolan bunting (emberiza hortulana L). The underlying classification algorithm is based on hidden markov models (HMMs) using Greenwood function cepstral coefficients (GFCCs), and the proposed feature normalization is implemented via a combination of cepstral mean subtraction and cepstral variance normalization, which are equivalent to a statistical z-score normalization across each example utterance. Results indicate a significant reduction in error on the individual identification task, improving accuracy from 89% to 96% in initial experiments, using a single call type over a small population of 6 individuals.

Acoustic to electric pitch comparisons in cochlear implant subjects with residual hearing.

The aim of this study was to assess the frequency-position function resulting from electric stimulation of electrodes in cochlear implant subjects with significant residual hearing in their nonimplanted ear. Six cochlear implant users compared the pitch of the auditory sensation produced by stimulation of an intracochlear electrode to the pitch of acoustic pure tones presented to their contralateral nonimplanted ear. Subjects were implanted with different Clarion electrode arrays, designed to lie close to the inner wall of the cochlea. High-resolution radiographs were used to determine the electrode positions in the cochlea. Four out of six subjects presented electrode insertions deeper than 450 degrees . We used a two-interval (one acoustic, one electric), two-alternative forced choice protocol (2I-2AFC), asking the subject to indicate which stimulus sounded the highest in pitch. Pure tones were used as acoustic stimuli. Electric stimuli consisted of trains of biphasic pulses presented at relatively high rates [higher than 700 pulses per second (pps)]. First, all electric stimuli were balanced in loudness across electrodes. Second, acoustic pure tones, chosen to approximate roughly the pitch sensation produced by each electrode, were balanced in loudness to electric stimuli. When electrode insertion lengths were used to describe electrode positions, the pitch sensations produced by electric stimulation were found to be more than two octaves lower than predicted by Greenwood's frequency-position function. When insertion angles were used to describe electrode positions, the pitch sensations were found about one octave lower than the frequency-position function of a normal ear. The difference found between both descriptions is because of the fact that these electrode arrays were designed to lie close to the modiolus. As a consequence, the site of excitation produced at the level of the organ of Corti corresponds to a longer length than the electrode insertion length, which is used in Greenwood's function. Although exact measurements of the round window position as well as the length of the cochlea could explain the remaining one octave difference found when insertion angles were used, physiological phenomena (e.g., stimulation of the spiral ganglion cells) could also create this difference. From these data, analysis filters could be determined in sound coding strategies to match the pitch percepts elicited by electrode stimulation. This step might be of main importance for music perception and for the fitting of bilateral cochlear implants.

Une anomalie aortique

As otology enters the field of gene therapy and human studies commence, the question arises whether audiograms – the current gold standard for the evaluation of hearing function – can consistently predict cellular damage within the human inner ear and thus should be used to define inclusion criteria for trials. Current assumptions rely on the analysis of small groups of human temporal bones post mortem or from psychophysical identification of cochlear “dead regions” in vivo, but a comprehensive study assessing the correlation between audiometric thresholds and cellular damage within the cochlea is lacking.A total of 131 human temporal bones from 85 adult individuals (ages 19–92 years, median 69 years) with sensorineural hearing loss due to various etiologies were analyzed. Cytocochleograms – which quantify loss of hair cells, neurons, and strial atrophy along the length of the cochlea – were compared with subjects' latest available audiometric tests prior to death (time range 5 h–22 years, median 24 months). The Greenwood function and the equivalent rectangular bandwidth were used to infer, from cytocochleograms, cochlear locations corresponding to frequencies tested in clinical audiograms. Correlation between audiometric thresholds at clinically tested frequencies and cell type-specific damage in those frequency regions was examined by calculating Spearman's correlation coefficients.Similar audiometric profiles reflected widely different cellular damage in the cochlea. In our diverse group of patients, audiometric thresholds tended to be more influenced by hair cell loss than by neuronal loss or strial atrophy. Spearman's correlation coefficient across frequencies was at most 0.7 and often below 0.5, with 1.0 indicating perfect correlation.Audiometric thresholds do not predict specific cellular damage in the human inner ear. Our study highlights the need for better non- or minimally-invasive tools, such as cochlear endoscopy, to establish cellular-level diagnosis and thereby guide therapy and monitor response to treatment.

Estimation of the length of the traveling wave envelope from response areas of neurons in the cochlear complex of the cat

Plots of discharge rate vs frequency at a number of intensities were determined for 163 neurons. Presumably the bandwidth of these response areas is determined by the shape of the traveling wave envelope on the basilar membrane. According to Greenwood, the distance from the peak to the apical foot of the wave is approximately constant [D. D. Greenwood, J. Acoust. Soc. Am. 34, 1364–1369 (1962)]. Using his frequency-distance function for the cat, we have calculated for each response area the distance along the basilar membrane between the locations for the best frequency and the upper cutoff frequency. When this “critical distance” (c.d.) is plotted versus best frequency for all units, c.d. tends to increase as best frequency declines below about 2 kHz. In order to obtain relatively constant c.d. throughout the entire frequency range, it is necessary to revise Greenwood's function such that the distances for the lower octaves deviate even more from a logarithmic representation than suggested by his original curve. [Supported by NIH grants NS 06225 ,and NS 12732.]

AUDITORY MASKING IN THE RAT.

Masked thresholds were measured with an operant tracking method at 4 frequencies between 1000 and 8000 cps and at 3–5 levels of wide-band noise. As in humans and cats, masked thresholds in the rat increased linearly with noise level. Critical ratios were calculated from the masked thresholds and found to be 10 or 11 dB greater than those in man and 5 to 6 dB greater than those in cats. Greenwood's function Δfcb = 10ax+b, relating critical bandwidths to position on the basilar membrane, was shown to fit reasonably well the data of this experiment. The constant relation shown by Fletcher between critical ratios and frequency DL's, previously found to apply to man and to cats, appears to apply to rats also.