US 11715457

U.S. Patent 11715457, titled "Real time correction of accent in speech audio signals," was issued to Sanas Ai Inc. (current assignee) on August 1, 2023. The inventors are Andrei Golman and Dmitrii Sadykov. The patent was filed on December 19, 2022.

Abstract:
The patent describes systems and methods for real-time correction of accent in speech audio signals. The method involves dividing a speech audio signal into input chunks, extracting acoustic features and accent-reduced linguistic features from these chunks and their associated context, and receiving a speaker embedding for a human speaker. These components (speaker embedding, acoustic features, and linguistic features) are then provided to a synthesis module to generate a melspectrogram with a reduced accent. Finally, a vocoder uses the melspectrogram to generate an output chunk of an output audio signal.

Plain-language overview of independent claims:

• Independent Claim 1: This claim outlines a method for real-time accent correction in speech. It involves taking a speech audio signal and breaking it into small segments called "input chunks." For each chunk, the system extracts two types of information: "acoustic features" (like pitch and energy) and "linguistic features" (which have had the accent reduced). This extraction uses the current chunk and some preceding audio data (context). The system also gets a "speaker embedding," which represents the unique voice characteristics of the speaker. Finally, it uses the speaker embedding, acoustic features, and linguistic features to create a new audio segment (an "output chunk") that has a corrected accent.

Regarding CAFC 2026 dockets for patent 11715457, the provided search results indicate active litigation related to this patent. Specifically, a PTAB case IPR2026-00275 was filed and is pending, and a US case was filed in the California Northern District Court (case 3:25-cv-05666). The petitioner for the IPR case is "Unified Patents PTAB Data."