Back to results
US 2026/0196232 A1Published application
Generative model for audio enhancement using reference vectors
AssigneeNVIDIA Corporation
InventorsMihir Nyayate, Swagat Ranjan Mohapatra, Angshuman Ghosh, Ambrish Dantrey
Life of this patent
Estimating the term…
Abstract
Apparatuses, systems, and techniques for real-time audio enhancement using a generative model. In at least one embodiment, output audio produced by the generative model corresponds to a version of input audio that is recorded using equipment or in an environment having qualities consistent with a reference vector.
Claims21
A computer-implemented method for audio enhancement, comprising:
- encoding input audio in a frequency domain by a first encoder using one or more first learned parameters to produce first latent vectors;
- encoding the input audio in a time domain by a second encoder using one or more second learned parameters to produce second latent vectors;
- fusing the first latent vectors and the second latent vectors in the time domain to produce fused latent vectors;
- and
- processing the fused latent vectors and a reference vector in the time domain according to one or more third learned parameters to generate output audio, the reference vector corresponding to audio recorded using equipment or in an environment that is different from the input audio, wherein the output audio corresponds to a version of the input audio that is recorded using equipment or in an environment having qualities consistent with the reference vector.
2 more elements
A system for audio enhancement, comprising:
- a memory that stores input audio;
- and
- one or more processors coupled to the memory to perform operations including:
- encoding the input audio in a frequency domain by a first encoder using first learned parameters to produce first latent vectors;
- encoding the input audio in a time domain by a second encoder using second learned parameters to produce second latent vectors;
- fusing the first latent vectors and the second latent vectors in the time domain to produce fused latent vectors;
- and
- processing the fused latent vectors and a reference vector in the time domain according to third learned parameters to generate output audio, the reference vector corresponding to audio recorded using equipment or in an environment that is different from the input audio, wherein the output audio corresponds to a version of the input audio that is recorded using equipment or in an environment having qualities consistent with the reference vector.
5 more elements
A processor comprising:
- one or more processing units to generate output audio corresponding to a version of input audio that is recorded using equipment or in an environment having qualities consistent with a reference vector by processing the reference vector and fused latent vectors in a time domain, wherein the fused latent vectors comprise first latent vectors produced by encoding the input audio in a frequency domain by a first encoder using first learned parameters and second latent vectors produced by encoding the input audio in the time domain by a second encoder using second learned parameters.
Dependencies are inferred from the claim text.