Synthetic Utterance Training for Voice Framework Accuracy
Find Innovative SolutionsGenerate Solutions
Solution Overview
Problem
Existing utterance-based frameworks face challenges in acquiring sufficient training samples efficiently, particularly when dealing with diverse training sets and identifying causes of reduced accuracy, which leads to resource wastage and reduced functionality.
Innovation Solution
The use of synthetic samples generated through vector space representations and genetic algorithms to train and test utterance-based frameworks, reducing reliance on natural samples and improving training efficiency and robustness.
Engineering Contradictions & Design Principles
Engineering Contradiction Analysis
1Adaptability or versatility
If natural training samples are acquired from individuals, then training diversity is improved, but time consumption and acquisition difficulty increase
Solution Approach 1:
The patent applies synthetic sample generation to create artificial training data that copies the essential characteristics of natural speech samples. By generating synthetic utterances with diverse acoustic properties, the system achieves training diversity without the time-consuming process of collecting natural samples from multiple individuals.
Solution Approach 2:
The patent uses parameter changes by modifying acoustic parameters (pitch, timbre, speaking rate, accent) in the synthetic sample generation process. This allows the creation of diverse training samples by varying these parameters rather than collecting samples from different people, thereby reducing acquisition time while maintaining diversity.
2Measurement precision
If more training samples are collected, then framework accuracy is improved, but resource requirements and processing time increase
Solution Approach 1:
The patent generates synthetic training samples that replicate the statistical properties and acoustic characteristics of natural speech. This copying approach allows the framework to be trained on large volumes of diverse data without the linear increase in collection and processing time that would result from acquiring equivalent natural samples.
Solution Approach 2:
The patent performs preliminary synthesis of training samples before the actual training process. By pre-generating diverse synthetic samples with known ground truth annotations, the system eliminates the time-consuming processes of sample collection, verification, and annotation that would otherwise be required to achieve the same training volume.
3Productivity
If synthetic samples are used for training, then training efficiency is improved, but sample realism and naturalness may be reduced
Solution Approach 1:
The patent employs parameter changes by systematically varying acoustic parameters in the synthetic sample generation. This includes adjusting pitch contours, timbre characteristics, speaking rates, and accent parameters to create samples that closely mimic natural speech variations, thereby improving naturalness while maintaining synthesis efficiency.
Solution Approach 2:
The patent creates composite training samples by combining synthetic base utterances with recorded natural speech segments or noise characteristics. This composite approach allows the synthetic samples to retain generation efficiency while incorporating authentic natural speech properties, thereby improving realism without sacrificing training efficiency.
Data Source
AI summary
Systems, methods, and devices for training and testing utterance based frameworks are disclosed. The training and testing can be conducting using synthetic utterance samples in addition to natural utterance samples. The synthetic utterance samples can be generated based on a vector space representation of natural utterances. In one method, a synthetic weight vector associated with a vector space is generated. An average representation of the vector space is added to the synthetic weight vector to form a synthetic feature vector. The synthetic feature vector is used to generate a synthetic voice sample. The synthetic voice sample is provided to the utterance-based framework as at least one of a testing or training sample.