Voice Synthesis Spectral Alteration for Expressive Temporal Control
Find Innovative SolutionsGenerate Solutions
Solution Overview
Problem
Current voice synthesis technologies, such as those for singing voices, face challenges in imparting sufficient expressivity and naturalness, particularly in maintaining temporal changes in spectral features, leading to unnatural and less expressive synthesized voices.
Innovation Solution
A voice synthesis method that alters synthesis spectra based on amplitude spectrum envelope contours to impart a richer variety of voice expression, using a processor and memory to alter and synthesize voice samples, incorporating expressive samples with temporal changes in spectral features to enhance expressivity.
Engineering Contradictions & Design Principles
Engineering Contradiction Analysis
1Manufacturing precision
If conventional voice synthesis adjusts harmonic components to match target voice quality, then voice quality transformation is achieved, but temporal changes in spectral features are not accurately reflected resulting in insufficient expressivity
Solution Approach 1:
The patent segments the spectral features into multiple independent components: amplitude spectrum envelope, spectral centroid, and spectral roll-off. Each component is extracted and modified separately using amplitude spectrum envelope contours that capture temporal expressivity changes. This segmentation allows precise control over voice quality transformation while independently preserving or enhancing expressive temporal variations.
Solution Approach 2:
The patent applies dynamic amplitude spectrum envelope contours that vary over time to modify the spectral features. Instead of static adjustments, the system uses time-varying contours extracted from expressive reference voices to dynamically reshape the synthesis spectra, enabling accurate representation of temporal expressivity changes throughout the phoneme duration.
2Manufacturing precision
If voice synthesis focuses on matching harmonic components for quality transformation, then target voice quality is achieved, but temporal expressivity variations are lost
Solution Approach 1:
The patent performs preliminary extraction of amplitude spectrum envelope contours from expressive reference voices before applying them to the target synthesis. These contours pre-encode the temporal expressivity patterns that need to be preserved. By preparing these contour profiles in advance, the system ensures that temporal spectral feature information is captured and ready for application during the quality transformation process.
Solution Approach 2:
The amplitude spectrum envelope contour serves as an intermediary that bridges the reference expressive voice and the target synthesized voice. It carries the temporal expressivity information from the reference and applies it to modify the target's spectral features, thereby preventing loss of temporal information during the quality transformation process.
3Ease of manufacture
If conventional methods use static harmonic adjustment, then voice quality change is achieved, but natural temporal variations in expression are not captured
Solution Approach 1:
The patent changes multiple spectral parameters simultaneously: amplitude spectrum envelope shape, spectral centroid position, and spectral roll-off characteristics. These parameters are modified according to time-varying amplitude spectrum envelope contours, enabling precise control over both voice quality and temporal expressivity. This multi-parameter approach captures natural temporal variations that single-parameter methods miss.
Data Source
AI summary
A voice synthesis method according to an embodiment includes altering a series of synthesis spectra in a partial period of a synthesis voice based on a series of amplitude spectrum envelope contours of a voice expression to obtain a series of altered spectra to which the voice expression has been imparted, and synthesizing a series of voice samples to which the voice expression has been imparted, based on the series of altered spectra.