Cochlear Implant Speech Processing Research Articles

Speech recognition as a function of the number of electrodes used in the SPEAK cochlear implant speech processor.

Speech recognition was measured in listeners with the Nucleus-22 SPEAK speech processing strategy as a function of the number of electrodes. Speech stimuli were analyzed into 20 frequency bands and processed according to the usual SPEAK processing strategy. In the normal clinical processor each electrode is assigned to represent the output of one filter. To create reduced-electrode processors the output of several adjacent filters were directed to a single electrode, resulting in processors with 1, 2, 4, 7, 10, and 20 electrodes. The overall spectral bandwidth was preserved, but the number of active electrodes was progressively reduced. After a 2-day period of adjustment to each processor, speech recognition performance was measured on medial consonants, vowels, monosyllabic words, and sentences. Performance with a single electrode processor was poor in all listeners, and average performance increased dramatically on all test materials as the number of electrodes was increased from 1 to 4. No differences in average performance were observed on any test in the 7-, 10-, and 20-electrode conditions. On sentence and consonant tests there was no difference between average performance with the 4-electrode and 20-electrode processors. This pattern of results suggests that cochlear implant listeners are not able to make full use of the spectral information on all 20 electrodes. Further research is necessary to understand the reasons for this limitation and to understand how to increase the amount of spectral information in speech received by implanted listeners.

Electrical stimulation of the auditory nerve: the coding of frequency, the perception of pitch and the development of cochlear implant speech processing strategies for profoundly deaf people.

1. The development of speech processing strategies for multiple-channel cochlear implants has depended on encoding sound frequencies and intensities as temporal and spatial patterns of electrical stimulation of the auditory nerve fibres so that speech information of most importance of intelligibility could be transmitted. 2. Initial physiological studies showed that rate encoding of electrical stimulation above 200 pulses/s could not reproduce the normal response patterns in auditory neurons for acoustic stimulation in the speech frequency range above 200 Hz and suggested that place coding was appropriate for the higher frequencies. 3. Rate difference limens in the experimental animal were only similar to those for sound up to 200 Hz. 4. Rate difference limens in implant patients were similar to those obtained in the experimental animal. 5. Satisfactory rate discrimination could be made for durations of 50 and 100 ms, but not 25 ms. This made rate suitable for encoding longer duration suprasegmental speech information, but not segmental information, such as consonants. The rate of stimulation could also be perceived as pitch, discriminated at different electrode sites along the cochlea and discriminated for stimuli across electrodes. 6. Place pitch could be scaled according to the site of stimulation in the cochlea so that a frequency scale was preserved and it also had a different quality from rate pitch and was described as tonality. Place pitch could also be discriminated for the shorter durations (25 ms) required for identifying consonants. 7. The inaugural speech processing strategy encoded the second formant frequencies (concentrations of frequency energy in the mid frequency range of most importance for speech intelligibility) as place of stimulation, the voicing frequency as rate of stimulation and the intensity as current level. Our further speech processing strategies have extracted additional frequency information and coded this as place of stimulation. The most recent development, however, presents temporal frequency information as amplitude variations at a constant rate of stimulation. 8. As additional speech frequencies have been encoded as place of stimulation, the mean speech perception scores have continued to increase and are now better than the average scores that severely-profoundly deaf adults and children with some residual hearing obtain with a hearing aid.

A new concept for Cochlear implant speech processing.

A new speech processing concept and its implementation cochlear implant neural net, simulation and stimulation (CINSTIM), for cochlear implant systems is currently under investigation. The main objectives of this new strategy are to improve speech perception within the early rehabilitation period for prelingually deaf cochlear implant users and to improve speech perception in noise for all implant users. The advantage of this approach is the capability to assess and compare entirely different speech processing methods within the existing implant hardware. The processing concept is based on discrete sequences of distinguishable stimulus patterns, robust speech processing, and a classifier. The embedded Kohonen classifier employs unsupervised neural net algorithms. Initial experimental results have validated the general processing scheme as well as the development of this complex software. In addition, a limited alphabet of distinguishable stimulus patterns was achieved and assessed with experienced users of a conventional cochlear implant system. Test subjects were able to learn and recognize up to 60% of these artificial patterns after a few training sessions. Currently this speech processing strategy is being evaluated through the Cochlear MSP system; however, it is compatible with the other computer driven digital cochlear implant systems.

Within‐subject comparisons of analog and pulsatile speech processors for cochlear implants

Analog and pulsatile speech processors for cochlear implants represent widely different strategies for coding perceptual information in speech. Between‐subject comparisons of their efficacy, however, have been confounded by large and unquantified subject differences. Two patients implanted with multiple electrode arrays have been tested both with four‐channel analog and pulsatile coding strategies. The analog processor performed bandpass filtering with syllabic compression/limiting and stimulated four bipolar electrode pairs. The pulsatile processor delivered volleys of biphasic pulses at the rate of F0 during voiced intervals of speech to four of six possible bipolar electrode pairs selected on the basis of spectral peak frequencies. Tests of auditory capabilities showed that the two coding strategies provided similar overall benefit. Analyses of consonant and vowel confusions revealed that the two strategies were not equally effective for some types of speech information. The results of these analyses can be used to identify the strengths and weaknesses of analog and pulsatile speech coding strategies for cochlear implants.

Speech processing studies using an acoustic model of a multiple-channel cochlear implant.

The speech perception of two multiple-channel cochlear implant patients was compared with that of three normally hearing listeners using an acoustic model of the implant for 22 different speech tests. The tests used included a minimal auditory capabilities battery, both closed-set and open-set word and sentence tests, speech tracking and a 12-consonant confusion study using nonsense syllables. The acoustic model represented electrical current pulses by bursts of noise and the effects of different electrodes were represented by using bandpass filters with different center frequencies. All subjects used a speech processor that coded the fundamental voicing frequency of speech as a pulse rate and the second formant frequency of speech as the electrode position in the cochlea, or the center frequency of the bandpass filter. Very good agreement was found for the two groups of subjects, indicating that the acoustic model is a useful tool for the development and evaluation of alternative cochlear implant speech processing strategies.