Audio Similarity Evaluator, Audio Encoder, Methods and Computer Program

Simple SummaryContent extracted from patent full text and abstract with AI.

This patent introduces an advanced audio similarity evaluator that assesses the perceptual similarity between an input audio signal and a reference audio signal. Instead of comparing raw waveforms, the method analyzes the signals' temporal envelopes across multiple frequency bands and their modulations, producing a similarity measure that aligns with how human listeners perceive audio. The invention integrates this evaluator into audio encoders, enabling more intelligent and perceptually relevant decision-making in encoding, especially for parametric or semi-parametric audio coding (e.g., bandwidth extension like MPEG-H IGF). This approach can also be implemented or learned by neural networks for efficiency.

Use CasesContent extracted from patent full text and abstract with AI.

• Automated parameter selection in audio encoders for music streaming and broadcasting to improve audio quality at low bitrates.
• Perceptual quality assessment tools for audio codec development and benchmarking, providing metrics closer to human hearing.
• Optimization of bandwidth extension and intelligent gap filling in modern audio codecs (e.g., MPEG-H, HE-AAC).
• Training data generation for machine learning models that need to predict perceptual audio similarity or set audio codec parameters.
• Adaptive audio coding for mobile communications and other bandwidth-constrained environments, enhancing user experience.
• Quality assurance in content production workflows, to detect or prevent perceptual degradations in post-processing.

BenefitsContent extracted from patent full text and abstract with AI.

• Provides a perceptual (listener-centric) similarity measure, resulting in better alignment with human audio quality judgments compared to traditional waveform metrics.
• Enables more efficient and higher quality audio encoding, especially for advanced (parametric) codecs, by selecting parameters that yield the best perceived quality.
• Improves automation in audio encoding workflows, reducing the need for expert manual tuning or subjective listening tests.
• Can be learned via neural networks, significantly lowering computational complexity for real-time or large-scale applications after initial training.
• Reduces audible artefacts such as switching noise, especially when frequency or parameter changes occur during playback, by using a context-aware stability criterion.
• Flexible integration: The similarity evaluation can be used as a standalone tool, part of an encoder, or as a training/ground truth generator for machine learning systems.

Technical Classifications (CPCs)

Inventors & Applicants

Patent Abstract

An audio similarity evaluator obtains envelope signals for a plurality of frequency ranges on the basis of an input audio signal. The audio similarity evaluator is configured to obtain a modulation information associated with the envelope signals for a plurality of modulation frequency ranges, wherein the modulation information describes the modulation of the envelope signals. The audio similarity evaluator is configured to compare the obtained modulation information with a reference modulation information associated with a reference audio signal, in order to obtain an information about a similarity between the input audio signal and the reference audio signal. An audio encoder uses such an audio similarity evaluator. Another audio similarity evaluator uses a neural net trained using the audio similarity evaluator.