Audio Key Estimation Using Learned Models
Find Innovative SolutionsGenerate Solutions
Solution Overview
Problem
Existing techniques face challenges in accurately estimating the key of musical pieces, especially when the frequency of the key note is low or when the power of each pitch name is not correlated with the tonality type, leading to difficulties in identifying the key for various types of musical pieces.
Innovation Solution
An audio analysis method that uses a learned model to generate key information by inputting a time series of feature amounts from audio signals, which has learned the relationship between keys and time series of feature amounts, allowing for accurate key estimation.
Engineering Contradictions & Design Principles
Engineering Contradiction Analysis
1Measurement precision
If rule-based methods are used to estimate key from audio signals, then the system complexity is low, but the key estimation accuracy deteriorates for various types of musical pieces
Solution Approach 1:
The patent transforms the key estimation problem from a rule-based approach to a parameter-based statistical model approach. By changing the fundamental parameter from simple frequency counting to statistical distribution analysis of pitch names, the system achieves higher accuracy across diverse musical pieces while managing complexity through probabilistic modeling.
Solution Approach 2:
The patent replaces the mechanical rule-based system with a statistical estimation model. Instead of applying fixed rules for key determination, the system uses statistical analysis of pitch name distributions and their relationships with tonality types, substituting deterministic mechanics with probabilistic reasoning to improve accuracy.
2Measurement precision
If the frequency of the key note is used to determine the key, then the method is simple, but the key estimation accuracy deteriorates when the key note frequency is low
Solution Approach 1:
The patent creates a universal statistical model that handles multiple scenarios simultaneously - it works whether the key note frequency is high or low, and adapts to different tonality types. The model's multi-functionality allows it to process various musical piece types without requiring separate methods for different frequency conditions, maintaining both accuracy and reasonable simplicity.
3Measurement precision
If the power of each pitch name is used to identify tonality type, then the method is straightforward, but the key estimation accuracy deteriorates when power and tonality type are not correlated
Solution Approach 1:
The patent introduces an intermediary statistical relationship between pitch names and tonality types. Instead of directly using pitch power to determine tonality, the system uses statistical models that capture the probabilistic relationships between pitch name distributions and tonality types, serving as an intermediary layer that handles cases where direct correlation is weak or absent.
Data Source
AI summary
An audio analysis method is realized by a computer and includes generating key information which represents a key, by inputting a time series of a feature amount of an audio signal into a learned model that has learned a relationship between keys and time series of feature amounts of audio signals.