TTS Linguistic Analysis Component for Seamless Style Switching

Resolve Bottlenecks,
Find Innovative SolutionsGenerate Solutions

Solution Overview

Problem

Conventional text-to-speech (TTS) systems require significant computational resources and time to switch between different speech styles, as they need to unload and load style-specific linguistic analysis components and speech bases, making seamless style switching impossible.

Innovation Solution

A TTS system that includes a linguistic analysis component capable of dynamically processing style indications within the input text to generate phonetic transcriptions for multiple styles, allowing a single linguistic analysis component and speech base to produce speech in various styles without the need for component swapping, using techniques like unit selection and statistical modeling.

Engineering Contradictions & Design Principles

VSEngineering Contradiction Analysis

1Reliability

If conventional TTS systems use style-specific linguistic analysis components and speech bases for each speech style, then speech quality and style accuracy are improved, but system complexity and resource consumption increase significantly when switching between styles

Engineering Contradiction:

Improvespeech style accuracyVSAvoidsystem complexity

Core Design Contradiction:

ReliabilityVSDevice complexity

Solution Approach 1:

The patent implements a universal linguistic analysis component that can process multiple speech styles through style indication parameters, eliminating the need for separate style-specific components. The single component dynamically adapts to different styles (e.g., neutral, joyful, didactic) by receiving style indications and adjusting its processing accordingly, thereby reducing system complexity while maintaining style accuracy.

Inventive Principle:

Principle #6Universality (Multi-functionality)

Solution Approach 2:

The system employs dynamic style switching where the linguistic analysis component and speech base can be changed on-the-fly without complete reloading. Style parameters are adjusted dynamically during operation, allowing seamless transitions between speech styles while maintaining system stability and reducing resource expenditure.

Inventive Principle:

Principle #15Dynamics

2Adaptability or versatility

If conventional TTS systems load and unload style-specific components when switching speech styles, then speech style versatility is improved, but processing time and latency increase

Engineering Contradiction:

Improvespeech style versatilityVSAvoidstyle switching latency

Core Design Contradiction:

Adaptability or versatilityVSLoss of time

Solution Approach 1:

The patent pre-loads multiple speech bases into memory before they are needed, so that when style switching is required, the system can immediately activate the appropriate pre-loaded base without incurring loading delays. This preliminary preparation eliminates the time penalty associated with component loading during style transitions.

Inventive Principle:

Principle #10Preliminary action

Solution Approach 2:

The system introduces a style indication parameter as an intermediary that controls the linguistic analysis component's behavior without requiring component replacement. This mediator allows the same component to produce different speech styles by adjusting its processing based on the style indication, thereby avoiding the time-consuming process of unloading and loading different components.

Inventive Principle:

Principle #24Intermediary (Mediator)

3Adaptability or versatility

If conventional TTS systems employ multiple voice-specific components for different styles, then speech style diversity is improved, but computational resource expenditure increases

Engineering Contradiction:

Improvespeech style diversityVSAvoidcomputational resource expenditure

Core Design Contradiction:

Adaptability or versatilityVSUse of energy by moving object

Solution Approach 1:

The patent implements a universal linguistic analysis component that serves multiple speech styles through style indication parameters, eliminating the need to maintain and process multiple separate style-specific components. This single component dynamically adapts to different styles (neutral, joyful, didactic, etc.), significantly reducing computational resource expenditure while preserving full style diversity.

Inventive Principle:

Principle #6Universality (Multi-functionality)

Solution Approach 2:

The system merges multiple style-specific processing functions into a single linguistic analysis component that handles all styles through parameter adjustment. By combining what would traditionally require separate components into one unified processor, the system reduces the computational overhead associated with managing multiple independent components while maintaining the ability to generate diverse speech styles.

Inventive Principle:

Principle #5Merging (Combining)

Data Source

PatentUS10192541B2Systems and methods for generating speech of multiple styles from text

Publication Date: 2019.01.29 CERENCE OPERATING CO

AI summary

A text-to-speech (TTS) system includes components capable of supporting the generation of speech output in any of multiple styles, and may switch seamlessly from producing speech output in one style to producing speech output in another style. For example, a concatenative TTS system may include a speech base storing speech units associated with multiple speech styles, and a linguistic analysis component to generate a phonetic transcription specifying speech output in any of multiple styles. Text input may include a style indication associated with a particular segment of the input text. The linguistic analysis component may invoke encoded rules and/or components based upon the style indication, and generate a phonetic transcription specifying a speech style, which may be processed to generate output speech.