 |
About AMP Lab Projects Downloads Publications People Links
Project - Speech-in-Silicon
Top of this page
| Motivation and Goal |
Until just recently, Sphinx 3 reached real time decoding on new P4 platforms, but for mobile devices such as PDAs processing capability is still fairly limited. In this project we developed a hardware reference model for a speech recognition engine, Lynx, written in C++. Based off of the large vocabulary continuous speech recognition software system Sphinx 3.0, the model assists in profiling and examining possible hardware optimizations for performance enhancements and efficiency. Using the same algorithms as Sphinx 3.0, the arithmetic units and memory structures of our hardware design differ greatly. By creating dedicated hardware architecture, we can take advantage of customized memory configurations and hardware parameters to best match the needs of the speech recognizer. Eventually a fully custom ASIC will be designed implementing the architectural decisions of this project.
Any suggestions or comments are welcome. Please send them to Edward Lin.