Waveform-Based Audio Classification for Selective Dialogue Enhancement

Resolve Bottlenecks,
Find Innovative Solutions
Generate Solutions

Solution Overview

Problem

Current media processing systems face challenges in enhancing audio streams without distorting non-speech segments, as existing methods apply enhancements uniformly across all audio content, potentially leading to poor user experience.

Innovation Solution

A computing system dynamically and selectively enhances audio streams by classifying segments as primarily speech or music using waveform attributes and applying dialogue enhancement only to speech segments, with crossfading to seamlessly transition between enhanced and unenhanced segments.

Engineering Contradictions & Design Principles

VSEngineering Contradiction Analysis

1Measurement precision

If dialogue enhancement is applied uniformly across the entire audio stream, then speech clarity is improved, but music and non-speech content quality deteriorates

Engineering Contradiction:
Improvespeech clarityVSAvoiddistortion of music
Core Design Contradiction:
Measurement precisionVSObject-affected harmful factors

Solution Approach 1:

The audio stream is divided into multiple segments that are individually classified as speech or non-speech content. This segmentation allows different processing to be applied to different portions of the audio stream, enabling speech enhancement without affecting music quality.

Inventive Principle:
Principle #1Segmentation

Solution Approach 2:

Dialogue enhancement is applied selectively only to segments classified as speech, while music and non-speech segments are left unenhanced. This local application of enhancement ensures that speech clarity is improved without introducing distortion to music content.

Inventive Principle:
Principle #3Local quality

2Object-affected harmful factors

If dialogue enhancement is applied selectively based on speech detection, then music quality is preserved, but speech clarity improvement is reduced

Engineering Contradiction:
Improvemusic quality preservationVSAvoidspeech clarity
Core Design Contradiction:
Object-affected harmful factorsVSMeasurement precision

Solution Approach 1:

The system continuously analyzes audio segments to detect speech content and dynamically adjusts enhancement application accordingly. This feedback mechanism ensures that enhancement is applied only when speech is detected, optimizing both music preservation and speech clarity improvement.

Inventive Principle:
Principle #23Feedback

Solution Approach 2:

The enhancement application is made dynamic rather than static, allowing the system to adaptively switch between enhanced and unenhanced states based on real-time speech detection. This dynamic approach optimizes the balance between speech clarity and music quality.

Inventive Principle:
Principle #15Dynamics

3Measurement precision

If speech segments are identified and enhanced individually, then speech clarity is maximized, but processing complexity increases

Engineering Contradiction:
Improvespeech clarityVSAvoidprocessing complexity
Core Design Contradiction:
Measurement precisionVSDevice complexity

Solution Approach 1:

The audio stream is divided into manageable segments that can be independently classified and processed. This segmentation reduces the overall processing complexity by breaking down the complex task of entire-stream analysis into simpler segment-level operations.

Inventive Principle:
Principle #1Segmentation

Solution Approach 2:

The system applies enhancement only to the extent necessary - only to segments containing speech - rather than processing the entire audio stream uniformly. This partial action approach reduces unnecessary processing complexity while maintaining speech clarity improvement.

Inventive Principle:
Principle #16Partial or excessive action

Data Source

PatentEP4243018A1Automatic classification of audio content as either primarily speech or primarily music, to facilitate dynamic application of dialogue enhancement
Publication Date: 2023.09.13 ROKU INC
  • EP4243018A1 patent drawingFigure 1
  • EP4243018A1 patent drawingFigure 2
  • EP4243018A1 patent drawingFigure 3

AI summary

A method for dynamically controlling enhancement of an audio stream is provided, where the audio stream defines a sequence of audio segments over time. Each audio segment defines a waveform having a plurality of waveform attributes. For each audio segment of the sequence of audio segments, the method includes: (i) determining a set of waveform-attribute values of the audio segment's waveform attributes, (ii) computing a first distance between the determined set of waveform-attribute values and a first predefined set of waveform-attribute values representative of speech, and computing a second distance between the determined set of waveform-attribute values and a second predefined set of waveform-attribute values representative of music, (iii) using the computed first and second distances as a basis to classify the audio segment as primarily speech or rather primarily music, and (iv) controlling, based on the classifying, whether or not to enhance the audio segment for output.