Text-to-Speech Template Conversion for Unsupported Characters

Resolve Bottlenecks,
Find Innovative Solutions
Generate Solutions

Solution Overview

Problem

Current text-to-speech systems generate unnatural speech when encountering characters that cannot be handled, such as Chinese characters in a TTS engine supporting American English, leading to blank portions and grammatical errors in speech output.

Innovation Solution

The system converts characters that cannot be read aloud into spaces and modifies the template text to remove corresponding modifiers and prepositions, ensuring grammatical correctness and natural speech output by generating a new text that can be processed by the TTS engine.

Engineering Contradictions & Design Principles

VSEngineering Contradiction Analysis

1Adaptability or versatility

If the TTS engine processes characters that cannot be read aloud (e.g., Chinese characters in an English-supported engine), then the text is converted with blank portions, but the speech output becomes unnatural and contains grammatical errors

Engineering Contradiction:
Improvecharacter encoding compatibilityVSAvoidspeech output quality
Core Design Contradiction:
Adaptability or versatilityVSReliability

Solution Approach 1:

The system performs preliminary detection of unsupported characters and converts them to spaces before text generation. It also preemptively modifies the template text to remove or adjust modifiers and prepositions associated with these characters, preventing grammatical errors before they occur in the final speech output

Inventive Principle:
Principle #10Preliminary action

Solution Approach 2:

The system introduces an intermediary processing step between text input and TTS conversion. This intermediary layer detects unsupported characters, converts them to spaces, and adjusts the template structure, acting as a mediator that bridges the gap between diverse character encodings and the TTS engine's language-specific requirements

Inventive Principle:
Principle #24Intermediary (Mediator)

2Loss of information

If the template text includes all information fields (tune name, artist name, album name), then complete information is provided, but grammatical errors occur when some fields contain unsupported characters

Engineering Contradiction:
Improveinformation completenessVSAvoidgrammatical correctness
Core Design Contradiction:
Loss of informationVSReliability

Solution Approach 1:

The system applies local quality modification by selectively adjusting only the portions of the template text associated with unsupported characters. Instead of removing or modifying the entire template, it locally modifies modifiers and prepositions adjacent to problematic fields while preserving the rest of the sentence structure, maintaining both information completeness and grammatical correctness

Inventive Principle:
Principle #3Local quality

Solution Approach 2:

The system changes the parameter state of template text elements dynamically. When unsupported characters are detected in specific fields, it modifies the presence or form of associated modifiers and prepositions, transforming the template structure based on the actual data quality and TTS engine capabilities

Inventive Principle:
Principle #35Parameter changes

Data Source

PatentEP2207165B1Information processing apparatus and text-to-speech method
Publication Date: 2018.10.03 ALPINE ELECTRONICS INC
  • EP2207165B1 patent drawingFigure 1
  • EP2207165B1 patent drawingFigure 2A~2B
  • EP2207165B1 patent drawingFigure 3A~3B

AI summary

An information processing apparatus playing back tunes includes an oral reading unit, a storage unit storing template texts for responses to questions from a user and template text conversion rules, an input unit for inputting a question from a user, and a control unit retrieving tune data and items of tune information associated with the tune data from an apparatus storing tune data. The control unit analyzes a question about a tune from a user to select a template text for a response to the question, detects characters in items of tune information of the tune, the characters replacing replacement symbols included in the template text, to perform conversion of the template text in a manner that depends on whether the characters can be read aloud, generates a text to be read aloud, using the converted template text, and causes the oral reading unit to read the text aloud.