Embedded Architectures for Speech and Machine Learning: A Design Space Exploration Approach

In recent years, research and development in the field of "natural interface" technologies and machine learning techniques have gained paramount importance. Text to Speech (TTS) conversion and Automated Speech Recognition (ASR) form a key part of such interface technologies. On the other hand, the future generation of intelligent computing devices will obviously require applications like machine learning based classifiers to perform classification tasks in a variety of fields ranging from data mining to recognition tasks in image and video. However, the real-time performance of such applications in hand-held platforms have not been explored till date. The present work takes into consideration some representative applications from these domains, evaluates their performance on embedded processor architectures and performs modifications in the underlying hardware platform for real time performance improvement of the concerned applications. The book can be used as supplementary text for advanced courses in embedded systems for demonstrating practical examples of key techniques like workload characterization, hardware-software co-design, custom instruction synthesis etc.