Voice Characteristic Dependent Weighting in Acoustic Models
Find Innovative SolutionsGenerate Solutions
Solution Overview
Problem
Current text-to-speech systems fail to convincingly mimic human voice characteristics, such as speaker voices, styles, emotions, and accents, limiting their naturalness and versatility in applications like electronic games, e-books, and navigation systems.
Innovation Solution
A text-to-speech method that simulates various voice characteristics by inputting text, dividing it into acoustic units, selecting voice characteristics, and converting these units into speech vectors using an acoustic model with probability distributions, where parameters are weighted based on voice-dependent clusters and sub-clusters, allowing for adaptation to new voice characteristics through recalculating weights to match audio inputs.
Engineering Contradictions & Design Principles
Engineering Contradiction Analysis
1Reliability
If traditional text-to-speech systems are used, then speech output can be generated, but the voice characteristics do not sound like a human voice
Solution Approach 1:
The acoustic model parameters are segmented into multiple clusters, where each cluster represents a specific voice characteristic (e.g., different speakers, emotions, accents). This segmentation allows the system to selectively apply clusters based on the desired voice characteristics, improving naturalness while managing complexity through modular organization of model parameters.
Solution Approach 2:
Different clusters of acoustic model parameters are applied to different segments or portions of the speech output depending on the required voice characteristics. Each cluster contains parameters optimized for specific local voice qualities (e.g., pitch, timbre, rhythm), allowing the system to apply appropriate local quality adjustments without requiring complete reconfiguration of the entire acoustic model.
2Adaptability or versatility
If voice characteristic dependent weighting is implemented, then multiple voice characteristics can be simulated, but the system complexity increases
Solution Approach 1:
Multiple clusters of acoustic model parameters are pre-computed and stored before runtime, with each cluster representing a specific voice characteristic. During speech generation, the system only needs to select and weight these pre-computed clusters rather than calculating all parameters from scratch, thereby achieving high adaptability while reducing real-time computational complexity.
Solution Approach 2:
The system changes parameters by adjusting the weighting factors applied to pre-computed clusters rather than fundamentally altering the cluster structures themselves. This parameter change approach allows flexible simulation of different voice characteristics through simple weight adjustments, maintaining system versatility while avoiding the complexity of dynamic cluster reconfiguration.
3Measurement precision
If clusters and sub-clusters with decision trees are used, then voice characteristics can be precisely controlled, but the processing time increases
Solution Approach 1:
Decision trees for selecting and weighting clusters are constructed and optimized during the training phase before deployment. This preliminary action allows the system to use efficient pre-computed decision rules during runtime, achieving precise voice characteristic control through fast tree-based selections rather than complex real-time calculations, thereby reducing processing time while maintaining precision.
Data Source
AI summary
A text-to-speech method for simulating a plurality of different voice characteristics includes dividing inputted text into a sequence of acoustic units; selecting voice characteristics for the inputted text; converting the sequence of acoustic units to a sequence of speech vectors using an acoustic model having a plurality of model parameters provided in clusters each having at least one sub-cluster and describing probability distributions which relate an acoustic unit to a speech vector; and outputting the sequence of speech vectors as audio with the selected voice characteristics. A parameter of a predetermined type of each probability distribution is expressed as a weighted sum of parameters of the same type using voice characteristic dependent weighting. In converting the sequence of acoustic units to a sequence of speech vectors, the voice characteristic dependent weights for the selected voice characteristics are retrieved for each cluster such that there is one weight per sub-cluster.