Diphone Synthesis Confidence Scoring for Speech Quality

Resolve Bottlenecks,
Find Innovative SolutionsGenerate Solutions

Solution Overview

Problem

Diphone synthesis systems face challenges when they have insufficient or missing diphones, leading to unnatural speech transitions and reduced acoustic quality in voice synthesis.

Innovation Solution

A computer system that converts audio or text into natural-sounding speech by selecting the best matching diphones from a database, concatenating them to minimize abrupt transitions, and adjusting confidence scores to ensure smooth morphing, even when diphones are incomplete or missing.

Engineering Contradictions & Design Principles

VSEngineering Contradiction Analysis

1Adaptability or versatility

If diphone synthesis is performed with insufficient or missing diphones, then the synthesis system can operate with limited database resources, but the speech quality becomes unnatural with abrupt transitions

Engineering Contradiction:

Improveability to operate with limited diphone databaseVSAvoidspeech quality and naturalness

Core Design Contradiction:

Adaptability or versatilityVSReliability

Solution Approach 1:

The patent introduces a confidence scoring mechanism as an intermediary to evaluate the quality of diphone matches. When exact diphone matches are unavailable, the system uses confidence scores to identify the best available alternatives and applies cross-fading techniques to smoothly transition between them, thereby maintaining speech quality even with limited database resources

Inventive Principle:

Principle #24Intermediary (Mediator)

Solution Approach 2:

The system dynamically adjusts the cross-fade parameter (transition duration and intensity) based on the confidence score of available diphone matches. When confidence is low due to missing diphones, the cross-fade duration is increased to create smoother transitions, transforming the parameter to compensate for database limitations

Inventive Principle:

Principle #35Parameter changes

2Reliability

If diphone database is expanded to improve speech quality, then more complete phonetic coverage is achieved, but system complexity and resource requirements increase

Engineering Contradiction:

Improvespeech quality and phonetic coverageVSAvoiddiphone database size and management complexity

Core Design Contradiction:

ReliabilityVSDevice complexity

Solution Approach 1:

The system performs preliminary confidence scoring and diphone selection during the speech synthesis preparation phase. By pre-evaluating which diphones are available and their confidence scores, the system can plan synthesis paths in advance, reducing the need for exhaustive database searches and simplifying real-time processing

Inventive Principle:

Principle #10Preliminary action

Solution Approach 2:

When exact diphone matches are missing, the system creates synthetic copies by cross-fading between adjacent available diphones. This copying approach generates the missing phonetic content on-the-fly, reducing the immediate need to store every possible diphone variation in the database

Inventive Principle:

Principle #26Copying

3Reliability

If confidence scoring is used to select best matching diphones, then transition smoothness is improved, but processing time increases

Engineering Contradiction:

Improvetransition smoothness and matching accuracyVSAvoiddiphone selection and synthesis time

Core Design Contradiction:

ReliabilityVSLoss of time

Solution Approach 1:

The system implements a threshold-based confidence scoring mechanism where only diphone candidates above a certain confidence threshold are considered. This partial evaluation approach avoids exhaustive searching of all possible diphone combinations, reducing processing time while still ensuring adequate match quality for smooth transitions

Inventive Principle:

Principle #16Partial or excessive action

Data Source

PatentUS9905218B2Method and apparatus for exemplary diphone synthesizer

Publication Date: 2018.02.27 SPEECH MORPHING SYST

AI summary

Method and apparatus for diphone or concatenative synthesis to compensate for insufficient or missing diphones.