Audio Fingerprint Hash Generation via Spectrogram Peak Detection
Find Innovative SolutionsGenerate Solutions
Solution Overview
Problem
Current audio recognition techniques are not accurate, scalable, or cost-efficient enough to track royalties for music played on radios or sampled in new recordings on a global scale, as they rely on common fingerprint hashes and similar frequency intervals that can lead to incorrect identifications.
Innovation Solution
An apparatus and method that process audio streams by transforming them into spectrograms, detecting significant peaks, generating fingerprint hashes based on frequency and time differences, and recursively matching these hashes against stored hashes, while accounting for varying playback speeds and audio quality.
Engineering Contradictions & Design Principles
Engineering Contradiction Analysis
1Reliability
If traditional audio fingerprinting techniques are used, then the process can be run on a global scale, but the accuracy deteriorates due to common fingerprint hashes and similar frequency intervals causing incorrect identifications
Solution Approach 1:
The audio signal is divided into multiple frequency sub-bands, and fingerprints are generated for each sub-band separately. This segmentation allows the system to capture unique characteristics from different frequency regions, improving identification accuracy while maintaining scalability across global audio streams.
Solution Approach 2:
The patent introduces temporal dimension by generating multiple fingerprints at different time offsets within each audio segment. This multi-dimensional approach (frequency sub-bands × time offsets) creates a more robust fingerprint set that distinguishes between similar tracks and handles playback speed variations effectively.
2Reliability
If multiple fingerprint hashes are generated to improve accuracy, then identification reliability improves, but computational cost and processing time increase
Solution Approach 1:
Fingerprints are pre-generated for multiple time offsets and frequency sub-bands during the indexing phase and stored in the database. When an audio stream needs to be identified, the system retrieves and compares these pre-computed fingerprints rather than generating them in real-time, significantly improving processing efficiency while maintaining high identification accuracy.
3Adaptability or versatility
If the system accounts for varying playback speeds and audio quality, then adaptability improves, but system complexity increases
Solution Approach 1:
The system dynamically adjusts the time offset values used for fingerprint generation based on expected playback speed variations. By incorporating multiple time offsets that account for potential speed deviations, the system adapts to different playback conditions without requiring complex real-time speed detection and adjustment algorithms.
Data Source
AI summary
Apparatus, method, and computer program code for processing audio stream. The method includes: obtaining first peaks of an audio stream, wherein the first peak comprises a first peak amplitude at a first frequency and at a first time offset from a beginning of the audio stream; for each first peak, detecting a second peak in a window with a predetermined offset from the first peak, wherein the second peak comprises a second peak amplitude at a second frequency and at a second time offset from the beginning of the audio stream; and for each first peak, generating a fingerprint hash based on the first frequency, a time difference between the first time offset and the second time offset, a frequency difference between the first frequency and the second frequency, and an amplitude difference between the first amplitude and the second amplitude.