Spectrum Envelope Contour Transformation for Natural Voice Synthesis
Find Innovative SolutionsGenerate Solutions
Solution Overview
Problem
Existing techniques for imparting sound expressions to voice signals, such as those described in Japanese Patent Application Laid-Open Publication No. 2014-2338, do not effectively generate natural audible sounds, particularly during attack and release portions of singing voices.
Innovation Solution
A sound processing method and apparatus that transform the spectrum envelope contours of a first sound signal by calculating differences with a reference sound signal, generating a synthesis spectrum envelope contour, and applying it to the first sound signal to produce a transformed sound signal with natural sound characteristics, specifically focusing on attack and release portions.
Engineering Contradictions & Design Principles
Engineering Contradiction Analysis
1Ease of manufacture
If existing techniques move harmonic components of voice signal to convert voice features, then voice distinctiveness is improved, but naturalness of audible sounds deteriorates
Solution Approach 1:
The patent introduces a spectrum envelope contour as an intermediary element to mediate between the original voice signal and the target sound characteristics. By operating on the spectrum envelope contour rather than directly moving harmonic components, the system achieves natural-sounding transformations while maintaining voice distinctiveness. The spectrum envelope contour serves as a mediator that preserves natural sound characteristics while enabling feature conversion.
Solution Approach 2:
The patent transforms voice features by changing parameters of the spectrum envelope contour, specifically by calculating differences between the spectrum envelope contour and reference spectrum envelope contours, and applying these differences to transform the contour. This parameter-based transformation approach maintains naturalness while achieving the desired voice feature conversion.
2Reliability
If spectrum envelope contour is transformed using difference calculation, then natural sound characteristics are improved, but processing complexity increases
Solution Approach 1:
The patent extracts the essential transformation information by calculating only the difference between the spectrum envelope contour and reference spectrum envelope contours, rather than processing the entire spectrum. This extraction of the differential component simplifies the processing while maintaining the natural sound characteristics, as only the relevant changes need to be computed and applied.
3Stability of the object's composition
If sound characteristics are aligned with reference signal, then continuity at boundaries is improved, but adaptability to different sounds decreases
Solution Approach 1:
The patent applies different transformation strategies to different portions of the sound signal. By calculating differences between the spectrum envelope contour and reference spectrum envelope contours at specific time points and applying these differences locally, the system maintains continuity at boundaries where reference signals are available while preserving the original sound's unique characteristics in regions where no reference exists. This local application of transformation ensures both continuity and adaptability.
Data Source
AI summary
A method obtains a first sound signal representative of a first sound, including a first spectrum envelope contour and a first reference spectrum envelope contour; obtains a second sound signal, representative of a second sound differing in sound characteristics from the first sound, including a second spectrum envelope contour and a second reference spectrum envelope contour; generates a synthesis spectrum envelope contour by transforming the first spectrum envelope contour based on a first difference between the first spectrum envelope contour and the first reference spectrum envelope contour at a first time point of the first sound signal, and a second difference between the second spectrum envelope contour and the second reference spectrum envelope contour at a second time point of the second sound signal; and generates a third sound signal representative of the first sound that has been transformed using the generated synthesis spectrum envelope contour.