Analysis of cochlear implant vocoder simulation including the current spread effect in the presence of background noise Kauˆ e Werner 1 , Rafael Chiea 2 , J´ ulio A. Cordioli 3 , Stephan Paul 4 1 Laborat´ orio de Vibra¸c˜ oes e Ac´ ustica, UFSC, Florian´opolis, Brasil, Email: kaue.werner@lva.ufsc.br 2 Laborat´ orio de Processamento Digital de Sinais, UFSC, Florian´opolis, Brasil, Email: rafaelchiea@gmail.com 3 Laborat´ orio de Vibra¸c˜ oes e Ac´ ustica, UFSC, Florian´opolis, Brasil, Email: julio.cordioli@ufsc.br 4 Laborat´ orio de Vibra¸c˜ oes e Ac´ ustica, UFSC, Florian´opolis, Brasil, Email: stephan.paul@ufsc.br Introduction A cochlear implant (CI) is an implantable device that re- stores auditory recognition of patients with severe to pro- found hearing loss. It extracts band-pass-filtered acoustic envelope information and modulates current pulse trains to stimulate the auditory nerve. Nowadays, the ability to recognize speech by CI recipients in quiet environments has achieved parity with normal hearing subjects (NH), but the presence of background noise greatly reduces the speech recognition performance in IC users. A common simulation method used to acoustically rep- resent speech recognition by CI users, that has been used extensively to study aspects of speech understanding, is the vocoder. It represents envelope information by mod- ulating an acoustic carrier, discarding the acoustic tem- poral fine structure from the original waveform. These vocoder simulations have been used in listening tests with NH and have been shown by many to provide results con- sistent with the outcome of cochlear implants. However, this method fails to consider important parameters such as pulse shape, stimulation rate, channel selection and current spread, that may be strongly related to speech recognition. The present work is therefore focused on the effect of cur- rent spread on speech recognition. The envelope infor- mation given by the vocoder was analysed using vocoder methods of simulation taking into account current spread in the cochlea and objectve metrics for speech recogni- tion, in quiet and noisy speech conditions. Current spread and vocoder The current spread in CIs is the spatial spread of elec- trical field generated by current pulse stimulation along neuron populations. It is known that the number of neu- rons excited by an electrical stimulus increases consid- erably as the stimulus level increases. The effect also depends on the electrode configuration of the device. In a monopolar configuration (MP), current pulses are de- livered to individual intra-cochlear electrodes with ref- erence to a far-field ground. For bipolar configuration (BP), the stimulation occurs when a potential difference is created between neighbouring electrodes leading to a current flow between them. Through experimental measurements, some authors have proposed models for the spatial current spread along the auditory nerve using exponential functions [1, 2, 3], given that the electric current at a position x is given by I (x)= I 0 e - |x 0 -x| λ , (1) where I 0 is the current pulse level, x 0 is the electrode position and λ is the length constant, that depends on the CI electrode configuration. The model demonstrates that as the stimulus current amplitude increases, both the width and the peak of the spatial current spread pro- file increase. Several studies reported measurements of current spread in bipolar and monopolar configurations. The length constant for bipolar configuration is 2 - 4mm and for monopolar 8 - 11mm [4]. A novel approach presented by Boghdady et al.[5], named neural-based vocoder, considers the CI pulse trains as the input that electrically stimulate neuron populations along the auditory nerve. Figure 1 presents a schematics of the proposed method. For each channel of the 22- electrode array, they used a local neuronal population of 47 to 48 neurons. Figure 1: Neural-based vocoder simulation blocks [5]: (1) current pulse sequence as input; (2) implantable electrode array; (3) integrate and fire model applied in neural popula- tions; (4) spatiotemporal integration; (5) average population activity for each channel; (6) effective loudness as temporal envelopes for each channel. Current spread is modelled as a Gaussian distribution of the stimulus amplitude along neighbouring neuronal populations nearby each electrode. The leaky integrate and fire model was implemented to simulate spiking ac- tivity through a finite difference approach of the model’s equation. After modelling the spiking activity for each DAGA 2016 Aachen 36