Hybrid Predictive Model for Speech Prosody Expressiveness
Find Innovative SolutionsGenerate Solutions
Solution Overview
Problem
Fully parametric model-based systems suffer from low prosodic expressiveness due to statistical averaging, while exemplar-based models are less robust due to reliance on low-level features, limiting their ability to replicate the full range of expressiveness observed in natural speech data.
Innovation Solution
A hybrid parametric/exemplar-based predictive model that extracts high-level structures from input data using a parametric model and pairs them with exemplars from training data to enhance prosody prediction, combining the strengths of both approaches and alleviating their weaknesses.
Engineering Contradictions & Design Principles
Engineering Contradiction Analysis
1Device complexity
If fully parametric model-based systems use statistical averaging, then the system footprint is small, but the predicted prosody suffers from low prosodic expressiveness due to flat intonation
Solution Approach 1:
The patent merges parametric models and exemplar-based models into a hybrid architecture. The parametric model extracts high-level structures and generates initial predictions, while the exemplar-based model refines these predictions by comparing with stored training exemplars, thereby combining the compactness of parametric systems with the expressiveness of exemplar-based systems.
Solution Approach 2:
The hybrid model applies different processing strategies to different aspects of prosody prediction. The parametric model handles global structural patterns, while the exemplar-based model handles local expressive variations, allowing each component to optimize for its specific function.
2Adaptability or versatility
If exemplar-based models use low-level features for selection, then the model is more expressive, but it becomes less robust
Solution Approach 1:
The parametric model serves as an intermediary that processes input data through high-level feature extraction before passing it to the exemplar-based model. This intermediate processing layer filters and structures the data, making the exemplar selection more robust while preserving expressiveness.
Solution Approach 2:
The parametric model performs preliminary feature extraction and high-level structure identification before the exemplar-based model conducts detailed prosody prediction. This preliminary processing prepares the data in a more robust manner for subsequent exemplar matching.
Data Source
AI summary
Systems and methods for prosody prediction include extracting features from runtime data using a parametric model. The features from runtime data are compared with features from training data using an exemplar-based model to predict prosody of the runtime data. The features from the training data are paired with exemplars from the training data and stored on a computer readable storage medium.