Neuromorphic detection of speech dynamics

Abstract

Speech and voice technologies are experiencing a profound review as new paradigms are sought to overcome some specific problems which cannot be completely solved by classical approaches. Neuromorphic Speech Processing is an emerging area in which research is turning the face to understand the natural neural processing of speech by the Human Auditory System in order to capture the basic mechanisms solving difficult tasks in an efficient way. In the present paper a further step ahead is presented in the approach to mimic basic neural speech processing by simple neuromorphic units standing on previous work to show how formant dynamics – and henceforth consonantal features – can be detected by using a general neuromorphic unit which can mimic the functionality of certain neurons found in the upper auditory pathways. Using these simple building blocks a General Speech Processing Architecture can be synthesized as a layered structure. Results from different simulation stages are provided as well as a discussion on implementation details. Conclusions and future work are oriented to describe the functionality to be covered in the next research steps.

Introduction

Neuromorphic Speech Processing is an emerging field which has attracted the attention of many researchers looking for new paradigms helping to better understand the underlying brain processes involved in speech perception, comprehension and production [10], [21]. This study can also be extended to cognitive audio (voice and sound processing by humans in general) when aspects as emotion or speaker recognition are concerned or in scene analysis [20], [21], [30]. The present paper is aimed to extend previous work on Neuromorphic Speech Processing [8] using a layered architecture of artificial Neuron-like Units derived from the functionality of the main types of neurons [14] found in the auditory pathways from the cochea to the primary and secondary auditory cortex [9]. In these early stages the typology a General Neuromorphic Computing Unit (GNCU) was defined using well-known paradigms from mask Image Processing [13]. It was also shown in the referred previous work [9] how one of these Mask Units can be adapted to model different processes as Lateral Inhibition to enhance Formant Detection. It was also shown how using different masks the GNCU could be configured to detect formant dynamics (ascending or descending resonance patterns appearing in certain speech sounds). The present work is intended to show how based on this GNCU a general layered architecture can be defined for the labelling of phonemes from formant positions and dynamics, advancing one step in the definition of a fully Bio-inspired Speech Processing Architecture. The paper is organized as follows: A brief description of formants and formant dynamics is given in Section 2. In Section 3 the different units found in the Auditory Pathway are defined accordingly to their functionality. The structure of the GNCU is shown to mimic the different units of interest for Speech Processing, and a Neuromorphic Speech Processing Architecture based on these units is presented. The purpose of Section 4 is to introduce plausible neural circuits to implement specific functions and comment results from simulations. Conclusions and future work are presented in Section 5.