Encoding by Reconstructing Phase Information Using a Structure Tensor on Audio Spectrograms

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

This patent describes an apparatus and method for reconstructing the phase information of audio signals from magnitude spectrograms using the structure tensor, a mathematical tool typically used in image processing. By analyzing the direction and rate of frequency changes for each time-frequency bin in a spectrogram, the invention allows for improved reconstruction of audio signals, especially for distinguishing between harmonic, percussive, and residual components. This method is notably effective for handling complex audio cases like frequency-modulated sounds (e.g., vibrato) that traditional methods have trouble classifying.

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

• Audio signal compression and coding, by transmitting only magnitude spectrograms and reconstructing phases at the decoder for efficient storage or transmission.
• Music information retrieval, including tasks like instrument separation, beat tracking, and vocal detection.
• Audio restoration or enhancement, particularly for remastering, cleaning, or isolating certain components from historical or noisy recordings.
• Real-time audio effects and remixing, enabling more precise manipulation of percussive and harmonic elements independently (e.g., dynamic upmixing or equalization).
• Automatic music transcription, chord detection, and rhythm analysis in digital audio workstations (DAWs) and research tools.

BenefitsContent extracted from patent full text and abstract with AI.

• Provides more accurate separation and reconstruction of audio components, especially for signals with frequency-modulation (e.g., vibrato), overcoming prior limitations.
• Enables lossy audio coding techniques with lower bitrates by only needing to transmit magnitude data and reconstructing phase efficiently at the receiver/decoder.
• Reduces computational complexity compared to prior state-of-the-art phase reconstruction methods, making it practical for real-time or embedded applications.
• Improves perceptible audio quality when reconstructing signals from spectrograms, due to better phase estimation.
• Versatile implementation: The method can be applied in software, hardware, or hybrid systems, and integrated into existing digital audio processing pipelines.

Technical Classifications (CPCs)

Inventors & Applicants

Patent Abstract

An apparatus for phase reconstruction from a magnitude spectrogram of an audio signal is provided. The apparatus comprises a frequency change determiner (110) being configured to determine a change of a frequency for each time-frequency bin of a plurality of time-frequency bins of the magnitude spectrogram of the audio signal depending on the magnitude spectrogram of the audio signal, and a phase reconstructor (140) being configured to generate phase values for the plurality of time-frequency bins depending on the changes of the frequencies determined for the plurality of the time-frequency bins.