Intermediate Audio Rendering for Stable Multichannel Object Separation
Find Innovative SolutionsGenerate Solutions
Solution Overview
Problem
Existing methods for extracting audio objects from multichannel audio signals, such as Dolby® Atmos™, face challenges like lack of convergence, time continuity, and problematic separation, especially for close objects, and require improved deep learning techniques for effective source separation.
Innovation Solution
A method using a machine learning algorithm, specifically a deep neural network, iteratively trained to generate intermediate audio signals with audio objects and position metadata by comparing rendered multichannel audio signals, employing loss functions and penalties to optimize the extraction process.
Engineering Contradictions & Design Principles
Engineering Contradiction Analysis
1Manufacturing precision
If traditional source separation methods are used to extract audio objects from multichannel audio, then the process is computationally simpler, but the separation quality is poor especially for close objects and lacks convergence
Solution Approach 1:
The patent introduces an intermediate audio signal as a mediator between the input multichannel audio and the output audio objects. This intermediate representation serves as a bridge that facilitates better separation by decomposing the complex separation task into manageable stages, improving separation quality while managing model complexity through structured processing
Solution Approach 2:
The patent segments the audio separation process into distinct components: extracting audio objects, generating position metadata, and creating an intermediate audio signal. This segmentation allows each component to be optimized independently, with the neural network focusing on object extraction while separate modules handle positioning and residual audio processing
2Measurement precision
If deep neural networks are used to improve audio object extraction, then separation accuracy improves, but training convergence becomes problematic
Solution Approach 1:
The patent implements feedback mechanisms through loss functions that compare the rendered output audio against the target audio objects and intermediate signal. This feedback guides the neural network training process, enabling convergence by continuously adjusting weights to minimize separation errors and improve extraction accuracy iteratively
Solution Approach 2:
The patent performs preliminary actions by pre-processing the input multichannel audio to generate the intermediate audio signal before final object extraction. This preliminary processing prepares the data in a more suitable format for the neural network, facilitating faster and more reliable convergence during training
3Adaptability or versatility
If audio objects are extracted from multichannel audio, then Dolby Atmos format compatibility is achieved, but time continuity is lost
Solution Approach 1:
The patent maintains time continuity by ensuring the neural network processes audio in a temporally coherent manner, generating position metadata that varies smoothly over time. The intermediate audio signal preserves temporal relationships, allowing continuous tracking of audio objects across time frames while maintaining Dolby Atmos compatibility
Data Source
AI summary
Described herein is a method for training a machine learning algorithm. The method may comprise receiving a first input multichannel audio signal. The method may comprise generating, using the machine learning algorithm, an intermediate audio signal based on the first input multichannel audio signal. The method may comprise rendering the intermediate audio signal into a first output multichannel audio signal. Further, the method may comprise improving the machine learning algorithm based on a difference between the first input multichannel audio signal and the first output multichannel audio signal. Described herein are further an apparatus for generating an intermediate audio format from an input multichannel audio signal as well as a respective computer program product comprising a computer-readable storage medium with instructions adapted to carry out said method when executed by a device having processing capability.