Phoneme Accent Determination Using Stochastic Probability Models
Find Innovative SolutionsGenerate Solutions
Solution Overview
Problem
Conventional speech synthesis technologies face challenges in accurately determining phonemes and accents, particularly in languages like Japanese and Chinese where word boundaries are not explicitly indicated, leading to unnatural-sounding synthetic speech due to insufficient context consideration and inefficient processing.
Innovation Solution
A system that stores a corpus of spellings, phonemes, and accents for individual word segmentations, allowing for the selection of phoneme and accent combinations based on their probability of occurrence in context, using a first corpus for common words and a second corpus for rare words, to generate more natural-sounding synthetic speech.
Engineering Contradictions & Design Principles
Engineering Contradiction Analysis
1Productivity
If conventional speech synthesis systems use frequency information to determine word segmentations and phonemes, then processing speed is improved, but accuracy of phoneme and accent determination deteriorates
Solution Approach 1:
The patent introduces a stochastic model as an intermediary between the speech synthesis system and the corpus data. This model calculates probabilities of phoneme and accent combinations based on contextual information from the corpus, rather than directly using frequency information. The stochastic model acts as a mediator that transforms raw frequency data into probabilistic predictions, improving accuracy while maintaining processing efficiency through mathematical modeling rather than exhaustive searching.
Solution Approach 2:
The patent changes the parameter used for determination from simple frequency counts to probability values derived from a stochastic model. By transforming the raw frequency information into probability distributions that consider contextual relationships, the system achieves more accurate phoneme and accent determination. The parameter change from frequency to probability allows the system to capture contextual nuances that simple frequency counting misses.
2Measurement precision
If the system considers context and probability of occurrence for phoneme and accent selection, then accuracy and naturalness of synthetic speech is improved, but processing complexity increases
Solution Approach 1:
The patent performs preliminary actions by pre-processing the corpus to extract and store contextual information and probability distributions for phoneme and accent combinations. During actual speech synthesis, the system only needs to query these pre-computed probabilities rather than performing complex contextual analysis in real-time. This preliminary preparation of probability data reduces the computational burden during the actual synthesis process while maintaining high accuracy.
Solution Approach 2:
The patent replaces complex mechanical processing of contextual information with a mathematical stochastic model. Instead of manually analyzing contextual relationships and making judgment-based selections, the system uses probability calculations to automatically determine the most appropriate phoneme and accent combinations. This substitution of mathematical modeling for mechanical processing simplifies the system architecture while improving accuracy.
3Adaptability or versatility
If multiple candidate word segmentations are evaluated based on frequency information, then versatility of word segmentation is improved, but processing time increases
Solution Approach 1:
The patent applies partial action by evaluating only the most probable word segmentations based on stochastic model predictions, rather than exhaustively evaluating all possible segmentations. The system generates a limited set of candidate segmentations with the highest probabilities and selects from those, achieving good versatility without the time cost of complete enumeration. This partial evaluation approach maintains adaptability for common cases while reducing processing time.
Data Source
AI summary
A system that outputs phonemes and accents of texts. The system has a storage section storing a first corpus in which spellings, phonemes, and accents of a text input beforehand are recorded separately for individual segmentations of the words that are contained in the text. A text for which phonemes and accents are to be output is acquired and the first corpus is searched to retrieve at least one set of spellings that match the spellings in the text from among sets of contiguous spellings. Then, the combination of a phoneme and an accent that has a higher probability of occurrence in the first corpus than a predetermined reference probability is selected as the phonemes and accent of the text.