Speech Synthesis Cost Function Segmentation
Find Innovative SolutionsGenerate Solutions
Solution Overview
Problem
Concatenative text-to-speech systems face challenges in determining the best combination of speech units due to the vast number of possible combinations, leading to suboptimal prosody and intonation in synthesized speech.
Innovation Solution
A method that selects speech units using a cost function comprising unit costs, concatenation costs, and event type costs, along with a linguistic model to determine prosody, ensuring optimal matching of phonetic elements and event types.
Engineering Contradictions & Design Principles
Engineering Contradiction Analysis
1Measurement precision
If dynamic programming is used to search the database efficiently, then the system can determine the best fit speech units, but the computational complexity becomes extremely high due to the enormous number of possible combinations
Solution Approach 1:
The patent segments the cost function into three distinct components: segmental costs (matching phonetic elements), concatenation costs (smoothness between adjacent speech units), and event type costs (prosodic features like stress and intonation). This segmentation allows the system to evaluate speech units through multiple independent criteria rather than treating all combinations equally, thereby reducing the effective search space while maintaining comprehensive evaluation.
Solution Approach 2:
The patent introduces event type costs as an additional parameter that captures supra-segmental features such as stress, intonation, and rhythm. By adding this parameter dimension, the system can differentiate between speech units that may have identical segmental and concatenation costs but differ in prosodic quality, thereby improving matching accuracy without requiring exhaustive search of all possible combinations.
2Device complexity
If the cost function only considers segmental criteria and concatenation costs, then the computation is more manageable, but the prosody and intonation of the synthesized speech are suboptimal
Solution Approach 1:
The patent adds a third dimension to the cost function by incorporating event type costs that evaluate prosodic features. While segmental costs and concatenation costs operate at the phoneme level, event type costs operate at the supra-segmental level, considering stress patterns, intonation contours, and rhythmic structure. This dimensional expansion allows the system to optimize for natural speech qualities without proportionally increasing computational complexity.
3Measurement precision
If all possible speech unit combinations are evaluated, then the best matched combination can be determined, but it becomes impossible to process even in modern and fast computer systems
Solution Approach 1:
The patent applies partial action by focusing computational resources on the most significant cost components. Rather than evaluating all possible combinations with equal weight, the segmented cost function allows the system to prioritize segmental matching (phonetic accuracy) and concatenation smoothness, while using event type costs to provide guidance for prosodic quality. This selective evaluation approach achieves near-optimal results with dramatically reduced computational effort.
Data Source
AI summary
Speech is synthesized for a given text by determining a sequence of phonetic components based on the text, determining a sequence of target phonetic elements associated phonetic components, determining a sequence of target event types associated with the phonetic components and determining a sequence of speech units from a plurality of stored speech unit candidates by use of a cost function. The cost function comprises a unit cost, a concatenation cost, and an event type cost for each speech unit in the sequence of speech units. The unit cost of a speech unit is determined with respect to the corresponding target phonetic element, while the concatenation cost of a speech unit is determined with respect to adjacent speech units and the event type cost of each speech unit is determined with respect to the corresponding target event type.