Peter V De Souza

Title: Innovations of Peter V De Souza

Introduction

Peter V De Souza is a notable inventor based in Yorktown Heights, NY (US). He holds a total of 14 patents that contribute significantly to the fields of speech recognition and language modeling. His work has been influential in enhancing the accuracy and efficiency of various acoustic systems.

Latest Patents

Among his latest patents, one focuses on determining and using acoustic confusability, acoustic perplexity, and synthetic acoustic word error rate. This patent discloses two statistics for assessing the quality of language models, namely acoustic perplexity and the synthetic acoustic word error rate (SAWER). These statistics rely on methods for computing the acoustic confusability of words. The invention allows for the substitution of models of acoustic data in place of real acoustic data to determine acoustic confusability. An evaluation model and a synthesizer model are created, both of which are hidden Markov models. Once a matrix is established from these models, confusability calculations can be performed. The patent also discusses methods for normalizing and smoothing confusability, as well as techniques to enhance the speed of matrix inversion and confusability calculations. Additionally, a method for caching and reusing computations for similar words is introduced.

Another significant patent involves enhanced likelihood computation using regression in a speech recognition system. This innovation aims to achieve low error rates in speech recognition systems, particularly those employing rank-based decoding. The methodology discriminates between the most confusable incorrect leaves and the correct leaf by adjusting their ranks. By increasing the likelihood of the correct leaf while decreasing that of the confusable leaves, the system utilizes auxiliary information from neighboring frames to improve likelihood computation. This new approach incorporates prediction error likelihoods to enhance the rank position of the correct leaf.

Career Highlights

Peter has worked with prominent companies such as IBM and Apple Inc. His contributions to these organizations have been pivotal in advancing their technologies in speech recognition and language processing.

Collaborations

Throughout his career, Peter has collaborated with esteemed colleagues, including Michael Alan Picheny and Ponani S Gopalakrishnan. These partnerships have fostered innovation and development in their respective fields.

Conclusion

Peter V De Souza's contributions to the field of speech recognition and language modeling are noteworthy. His patents reflect a deep understanding of acoustic systems and a commitment to improving technology. His work