Audio Signal Processing Loudness Normalization for Sound Source Separation
Find Innovative SolutionsGenerate Solutions
Solution Overview
Problem
Current sound source separation technologies face challenges in accurately processing audio signals with varying loudness levels, which affects the performance of machine learning models used for sound source separation, particularly in distinguishing between different audio components like voice and background instruments.
Innovation Solution
The proposed method involves normalizing the loudness of audio signals based on a pre-designated target loudness using a machine learning model, which includes gate logic, to enhance sound source separation by classifying frequency bin-specific scores and applying mask warping to reduce interference, and then de-normalizing the signals to maintain the original loudness ratio.
Engineering Contradictions & Design Principles
Engineering Contradiction Analysis
1Measurement precision
If loudness normalization is applied to audio signals before machine learning processing, then sound source separation performance is improved, but the complexity of the processing system increases
Solution Approach 1:
The patent applies loudness normalization as a preliminary processing step before the audio signal is fed into the machine learning model for sound source separation. By pre-normalizing the loudness of the input audio signal to match a target loudness level, the system prepares the data in an optimal state for the ML model, improving separation performance without requiring changes to the core separation algorithm
Solution Approach 2:
The patent introduces loudness normalization as an intermediary processing stage between the raw audio input and the machine learning model. This intermediary step acts as a mediator that transforms the input signal into a standardized form, enabling the ML model to process signals with consistent loudness characteristics and thereby improving overall system performance
2Measurement precision
If loudness normalization and de-normalization processes are added, then accuracy of audio component separation is improved, but processing time increases
Solution Approach 1:
The loudness normalization is performed as a preliminary step before the computationally intensive machine learning inference, preparing the signal in advance to avoid the need for repeated adjustments during processing. The de-normalization is applied as a final step after separation, allowing the ML model to work with normalized signals for optimal accuracy while minimizing overall processing time
Data Source
AI summary
Disclosed is an operation method of an audio signal processing device configured to process an audio signal including a first audio signal component and a second audio signal component. The operation method includes: receiving the audio signal; normalizing loudness of the audio signal, based on a pre-designated target loudness; acquiring the first audio signal component from the audio signal having the normalized loudness, by using a machine learning model;and de-normalizing loudness of the first audio signal component, based on the pre-designated target loudness.