NvidiaPatent applicationAudio & voiceInterest 7/10

Nvidia patents method to fix spoken words using AI voice cloning

Patent drawing
Patent drawing from the official filing.

Nvidia describes a system that uses machine learning to correct specific words in an audio recording. It compares two transcripts to find errors, generates a replacement sound for the wrong words, and adjusts that sound to match the speaker's original voice.

Why it mattersThis could make video editing or live broadcasting easier by allowing creators to fix verbal mistakes without re-recording the entire clip.

In the filing’s own words

Disclosed are systems and techniques for correcting speech audio using machine learning models. The techniques include comparing a first transcript of an input audio to a second transcript of the input audio to identify a set of modified words, wherein the second transcript is a modified version of the first transcript. The techniques include generating a synthetic audio using a text-to-speech machine learning model based on the set of modified words and generating a modified synthetic audio using a voice match machine learning model based on the synthetic audio and at least a portion of the input audio. The techniques include generating an output audio by combining the input audio and the modified synthetic audio.

What exactly they claim (claim 1)

1. A method comprising: comparing an original transcript of an input audio to a modified transcript corresponding to the input audio to identify a set of modified words; generating synthetic audio using a text-to-speech machine learning model based at least on the set of modified words; generating modified synthetic audio using a voice match machine learning model based at least on the synthetic audio and at least a portion of the input audio; and generating output audio by combining the input audio and the modified synthetic audio.

Facts

Company
Nvidia (NVIDIA Corporation)
Official title
SPEECH AUDIO CORRECTION USING MACHINE LEARNING MODELS
Publication number
US20260360380A1
Application number
18586992
Filed
Feb 26, 2024
Published
Nov 12, 2026
Inventors
5
Claims
20

Classified under

G10L 21/0216 Physics & computingG10L 13/08 Physics & computingG10L 15/26 Physics & computing

More from Nvidia in Audio & voice

Summaries are simplified from the filing’s own abstract and first claim. A patent filing is not a product announcement.
Patent data and drawings: U.S. Patent and Trademark Office (USPTO). Nebula is not affiliated with or endorsed by the USPTO.