Shifeng Pan

Title: Shifeng Pan - Innovator in Speech Synthesis Technology

Introduction

Shifeng Pan is an accomplished inventor based in Redmond, WA (US). He is currently associated with Microsoft Technology Licensing, LLC. Although he has not yet been granted any patents, his work in the field of speech synthesis is noteworthy.

Latest Patent Applications

One of his significant patent applications is titled "Cross-Speaker Style Transfer Speech Synthesis." This disclosure provides methods and apparatuses for training an acoustic model aimed at implementing cross-speaker style transfer. The process involves a style encoder and utilizes training data that includes text, speaker ID, style ID, and acoustic features corresponding to a reference audio. A reference embedding vector is generated through the style encoder based on the acoustic features. Adversarial training is then performed on this reference embedding vector using the style ID and speaker ID to eliminate speaker information while retaining style information. Subsequently, a style embedding vector is generated based on the reference embedding vector after adversarial training. Finally, predicted acoustic features are produced based on a state sequence corresponding to the text, a speaker embedding vector linked to the speaker ID, and the style embedding vector.

Conclusion

Shifeng Pan's contributions to the field of speech synthesis, particularly through his innovative patent application, highlight his potential as an inventor in this evolving technology landscape. His work at Microsoft Technology Licensing, LLC positions him at the forefront of advancements in speech synthesis.