Feng-Ju Chang

Title: Innovations of Feng-Ju Chang

Introduction

Feng-Ju Chang is a notable inventor based in Pittsburgh, PA, who has made significant contributions to the field of automatic speech recognition. With a total of two patents to his name, Chang's work focuses on enhancing the efficiency and accuracy of speech recognition technologies.

Latest Patents

Chang's latest patents include innovative techniques for performing automatic speech recognition (ASR). One of his patents describes an ASR component that integrates contextual information from user profile data into audio encoding data. This integration allows the ASR system to predict tokens corresponding to spoken input, utilizing personalized words such as contact names and device names. The ASR component is designed to apply attention to audio frames relevant to these personalized words, enhancing the recognition process.

Another patent involves a multi-channel transformer acoustic model that processes multiple audio signals from a microphone array. This model outputs probabilities for acoustic units of an utterance represented in the audio signals. It employs self-attention on embeddings of the audio signals and cross-channel attention to model relationships across channels and time frames. The multi-channel model includes a transducer for frame-by-frame processing, further improving the accuracy of speech recognition.

Career Highlights

Feng-Ju Chang is currently employed at Amazon Technologies, Inc., where he continues to develop cutting-edge technologies in speech recognition. His work is instrumental in advancing the capabilities of ASR systems, making them more responsive and personalized.

Collaborations

Chang has collaborated with notable colleagues such as Athanasios Mouchtaris and Siegfried Kunzmann, contributing to the innovative projects at Amazon Technologies.

Conclusion

Feng-Ju Chang's contributions to automatic speech recognition demonstrate his commitment to innovation in technology. His patents reflect a deep understanding of user needs and the potential for personalized speech recognition systems.