For Anshuman Chadha, this research experience has definitely influenced his future. After finishing his research project and completing a degree in Information & Computer Science, he plans to attend graduate school with the hopes of working further with speech and linguistics technologies. In addition to the proposal and grant writing skills he has developed, this project also gave him the opportunity to improve his technical writing skills. To supplement to his academic work, Anshuman interns at Raytheon Systems as a software engineer. In his spare time, he enjoys reading Tom Clancy and James Clavell novels. Anshuman is also one of two UCI undergraduates selected to present his research at 2004 UC-Day in Sacramento.

Speech coding is digital representation of the speech sound. Traditional speech-coding schemes handle acceptable quality speech at bit rates that exceed 2,400 bits per second (bps). This project is based on an extremely low bit-rate speech-coding scheme, on the order of 40 bps. In order to get such a low compression rate, all recognized speech is coded at the lexeme (word level), with low-level elements such as tone and frequency completely disregarded. The lexemes are coded using a probability-based compression mechanism. This coded data is then decoded and decompressed using text-to-speech. Results from this experiment showed that while there were errors introduced by the speech recognition engine, listeners were often able to recover from such errors by inferring what was trying to be expressed. Results also showed that errors made by the listeners in recognizing the synthesized samples were highly dependent on the content of the samples, especially with regard to the familiarity the listeners had with the topic. Applications of this method include speech storage and communications over low-capacity channels.

Cristina Videira Lopes School of Information & Computer Science

The manipulation of information in its digital form is one of computer science’s fundamental gifts to all other sciences. Things get interesting and fun when the data is related to what human beings can do as in speech communication. It turns out that we, humans, have a relatively low bit rate when communicating through speech. Anshuman’s project confirmed the rate of 40 bits per second that has been suggested before as the rate of verbal communication. We all know that there’s a lot more to speech than the words, but this project focused only on the words. Compared to the high bandwidth computers out there, we don’t say much!