Neural Network Audio Compression Using Pitch and Vector Quantization
Find Innovative SolutionsGenerate Solutions
Solution Overview
Problem
Existing audio compression solutions are resource-intensive and produce inadequate results, struggling to compress speech audio sequences at low bitrates without degrading audio quality.
Innovation Solution
An audio processing system trains neural networks to encode and decode speech audio sequences using pitch detection and vector quantization, separating pitch and symbolic representations to enable efficient compression at low bitrates while maintaining high quality.
Engineering Contradictions & Design Principles
Engineering Contradiction Analysis
1Quantity of substance
If traditional audio codecs are used to compress speech audio sequences, then storage capacity and bandwidth are reduced, but audio quality degrades at low bitrates
Solution Approach 1:
The patent segments the audio signal into pitch data and symbolic representations (phonemes), encoding them separately. This segmentation allows each component to be optimized independently, enabling high-quality reconstruction at low bitrates by efficiently representing the essential characteristics of speech without requiring high bitrate transmission of the entire audio signal.
Solution Approach 2:
The patent extracts and encodes only the essential features of speech (pitch and phonemes) rather than compressing the entire audio signal. By taking out and separately encoding these critical components, the system achieves efficient compression while maintaining audio quality, as these extracted features contain the most important information for speech reconstruction.
2Productivity
If neural networks are trained to encode and decode speech audio sequences with separation of pitch and symbolic representations, then compression efficiency improves, but system complexity increases
Solution Approach 1:
The neural network system is segmented into distinct components: pitch detection module, phoneme recognition module, encoding module, and decoding module. Each module handles a specific aspect of the processing, which improves compression efficiency through specialized optimization while managing complexity by organizing functions into separate, manageable units rather than a monolithic system.
Solution Approach 2:
The system uses self-service through automated training of neural networks on speech data, where the model learns optimal encoding and decoding strategies without manual intervention. The pitch detection and phoneme recognition components automatically adapt to different speech patterns, reducing the need for manual system configuration and simplifying deployment despite the underlying complexity.
Data Source
AI summary
Embodiments are disclosed for training an audio processing system to perform high-quality speech audio encoding and decoding using neural networks. In particular, in one or more embodiments, the disclosed systems and methods comprise receiving an audio sequence, the audio sequence including speech audio, generating pitch data representing detected pitch within the audio sequence, passing the audio sequence through an audio encoder to generate a vector representation of the audio sequence, generating, by a vector quantizer, an encoded vector representation of the audio sequence using the vector representation of the audio sequence and a codebook of discrete vectors, and reconstructing, by an audio decoder, the audio sequence using the pitch data and the encoded vector representation of the audio sequence.