Speech Recognition System Using Dynamic Path Scoring for Pronunciation Evaluation

Resolve Bottlenecks,
Find Innovative Solutions
Generate Solutions

Solution Overview

Problem

Existing speech recognition systems face challenges in accurately recognizing speech signals due to acoustic interference, variations in speaking styles and accents, and inefficiencies in processing, particularly with children's voices, leading to high computational costs and inability to recognize reading mistakes.

Innovation Solution

A speech recognition system employing a weighted finite state transducer-based algorithm that dynamically adapts to acoustic parameters and pronunciation, refining user-defined text by removing non-lexical punctuation and non-spoken content, and providing feedback on speech evaluation with lower latency and computational cost.

Engineering Contradictions & Design Principles

VSEngineering Contradiction Analysis

1Adaptability or versatility

If existing voice recognition systems use databases containing many versions of given utterances to accommodate variations in delivery, then the systems can handle variations in accents and speaking styles, but the computational cost becomes very high and the analysis time increases significantly

Engineering Contradiction:
Improvehandling variations in speech deliveryVSAvoidcomputational efficiency
Core Design Contradiction:
Adaptability or versatilityVSProductivity

Solution Approach 1:

The patent segments the speech recognition process into distinct modules: acoustic feature extraction, phoneme recognition, word recognition, and sentence-level evaluation. This segmentation allows each module to process specific aspects independently, reducing the computational burden of handling all variations simultaneously while maintaining adaptability through targeted processing at each stage.

Inventive Principle:
Principle #1Segmentation

Solution Approach 2:

The system dynamically adjusts its processing based on the input characteristics. It uses adaptive thresholding for phoneme detection and dynamically selects which database versions to compare based on the detected speech patterns. This dynamic approach allows the system to handle variations efficiently by activating only the necessary processing paths rather than exhaustively comparing all database versions.

Inventive Principle:
Principle #15Dynamics

2Measurement precision

If existing voice recognition systems access databases with multiple versions of utterances to improve recognition accuracy, then the systems can recognize different accents and styles, but the systems require significant time to perform analysis

Engineering Contradiction:
Improvespeech recognition accuracyVSAvoidanalysis time
Core Design Contradiction:
Measurement precisionVSLoss of time

Solution Approach 1:

The patent performs preliminary acoustic feature extraction and phoneme segmentation before full word and sentence recognition. By preprocessing the speech signal to identify phoneme boundaries and characteristics in advance, the system reduces the time required for subsequent matching against database versions, as the search space is already constrained by the preliminary analysis.

Inventive Principle:
Principle #10Preliminary action

Solution Approach 2:

The system implements early termination criteria in the recognition process. When a sufficient match is found between the processed speech and database versions, the system skips remaining comparisons and proceeds to evaluation. This allows the system to achieve acceptable accuracy without exhaustively analyzing all database versions, thereby reducing overall analysis time.

Inventive Principle:
Principle #21Skipping (Rushing through)

3Ease of operation

If existing voice recognition systems use conventional recognition methods, then the systems can process speech signals, but they are incapable of recognizing reading mistakes and pronunciation errors and providing feedback

Engineering Contradiction:
Improvebasic speech processing capabilityVSAvoidpronunciation error detection
Core Design Contradiction:
Ease of operationVSLoss of information

Solution Approach 1:

The patent implements a feedback mechanism where the system compares the recognized speech against the expected text, identifies deviations such as pronunciation errors and reading mistakes, and provides feedback to the user. This feedback loop enables the system to not only process speech but also to evaluate and correct user input, transforming a passive recognition system into an active learning and correction system.

Inventive Principle:
Principle #23Feedback

Solution Approach 2:

The patent adds an evaluation dimension to the traditional recognition process. Instead of only determining what was said, the system evaluates the quality and accuracy of the speech by comparing it against reference text and providing scores for different aspects such as pronunciation accuracy, fluency, and completeness. This additional dimension enables error detection and feedback without compromising the basic recognition capability.

Inventive Principle:
Principle #17Another dimension (Dimensionality change)

Data Source

PatentEP3718107B1Speech signal processing and evaluation
Publication Date: 2023.11.08 AURIS TECH LTD
  • EP3718107B1 patent drawingFigure 1
  • EP3718107B1 patent drawingFigure 2
  • EP3718107B1 patent drawingFigure 3

AI summary

There is disclosed a system that, when in operation, evaluates speech, for example evaluates a speech signal generated using a microphone to record an oral utterance. The system comprises: a user device that receives a user-defined text; and a processing unit that is operatively coupled to the user device, wherein the processing unit: refines the user- defined text to form a base text having a plurality of base text-units; defines at least one path traversing the plurality of base text-units, wherein the at least one path traverses at least one of the plurality of base text-units sequentially; receives an utterance for the user-defined text from the user device; processes the utterance to generate an interpretation text having a plurality of interpretation text-units; maps the plurality of interpretation text-units to the plurality of base text-units; assigns a score to the at least one path based on the mapped quality; identifies a traversed path from the at least one path; and calculates the assigned score associated with the traversed path to evaluate the speech.