An architecture for a speech recognition system

Abstract

A RECOGNITION SYSTEM composed (in addition to memory chips) of afiont-end chip for feature extraction, a pattern matching chip and a microprocessor for the final (low data rate) decision and control will be discussed. A TTL prototype of this sytem is presently in use to provide voice-input t o an IC layout system. interface between the microphone and the linear A/D, short-time spectrum analysis using a bank of bandpass filters, logarithmic compression and endpoint detection. These functions could either be realized by analog processing blocks', or using digital techniques'. Typically, the output of the front-end chip is processed in frames of 20ms. During the training phase of the system, these frames form the reference patterns or templates, and are stored in the template memory. During the recognition phase, the frames form a test pattern which is transferred to the pattern matching circuitry. Each of the reference patterns is compared to the test pattern, and a distance between them is computed to determine the degree of similarity.

Smartphone-Based Online and Offline Speech Recognition System for ROS-Based Robots

“…A speech recognition system (e.g., systems presented in [22,23]) takes voice input, recognizes it, and generates an output in the form of text. Fig.…”

Instruction set extensions for Dynamic Time Warping

“…Early processors introduced in the 1980s featuring custom DTW accelerators for speech recognition [10,19,20,22]. More recently, low-power DTW accelerators were developed for bodyarea sensor networks to accelerate activity recognition [17,18].…”

“…Although some time-series data can be collected and stored onto a server, other data must be mined in real-time, as the value of the data rapidly degrades otherwise. Real-time time-series data mining applications include speech recognition [10,19,20,22], activity recognition [17,18], robotics, financial markets, user interfaces, and many others [23].…”