Pitch Estimation Using Spectral Basis Vectors and Temporal Constraints
Find Innovative SolutionsGenerate Solutions
Solution Overview
Problem
Current computerized systems lack the ability to effectively detect and transcribe musical sources within a sound mixture, as they are not as developed as human ear training in associating sounds with specific instruments and notes.
Innovation Solution
A method for estimating features of a sound mixture using a model with a dictionary of spectral basis vectors, where the model is based on isolated training data and includes feature tagging, allowing for pitch estimation constrained by temporal data, such as a semantic continuity constraint.
Engineering Contradictions & Design Principles
Engineering Contradiction Analysis
1Measurement precision
If computerized systems use basic spectral analysis to detect sound sources, then the system complexity remains low, but the measurement precision of pitch and volume estimation is insufficient
Solution Approach 1:
The system performs preliminary training to pre-compute spectral basis vectors and construct dictionaries for each sound source type before actual pitch estimation. This pre-processing creates reusable models that improve measurement precision without increasing real-time computational complexity
Solution Approach 2:
The pitch estimation process is segmented into distinct phases: spectral decomposition using pre-computed basis vectors, dictionary matching for source identification, and temporal constraint application. This segmentation allows complex operations to be performed offline while keeping online processing manageable
2Reliability
If the system estimates features for each time frame independently, then the processing speed is fast, but the reliability of pitch tracking is poor due to temporal inconsistencies
Solution Approach 1:
The system applies temporal constraints that use feedback from previous time frame estimates to constrain current estimates. The semantic continuity constraint ensures that pitch estimates evolve smoothly over time, preventing unrealistic jumps and improving tracking reliability
Solution Approach 2:
The pitch estimation maintains continuity by applying temporal constraints that enforce smooth transitions between successive time frames. This ensures that the pitch tracking remains reliable while processing occurs efficiently at each time step
3Adaptability or versatility
If the system uses isolated training data to build source models, then the adaptability to different instruments is improved, but the loss of information occurs when sources are mixed together
Solution Approach 1:
The system transforms the sound mixture analysis into a different dimensional space using spectral decomposition. By representing sounds in terms of spectral basis vectors and matching them against dictionaries of normalized spectra, the system recovers information that would be lost in the time domain mixture
Solution Approach 2:
The spectral basis vectors serve as intermediaries between the mixed sound signal and the source identification process. These basis vectors capture the essential spectral characteristics of each source type, enabling the system to separate and identify individual sources even when they are mixed together
Data Source
AI summary
A sound mixture may be received that includes a plurality of sources. A model may be received for one of the source that includes a dictionary of spectral basis vectors corresponding to that one source. At least one feature of the one source in the sound mixture may be estimated based on the model. In some examples, the estimation may be constrained according to temporal data.