Speech Recognition Accuracy Prediction via Feature Extraction

Resolve Bottlenecks,
Find Innovative Solutions
Generate Solutions

Solution Overview

Problem

Current automatic speech recognition systems are inefficient in predicting recognition accuracy without processing large audio files, making it time-consuming and impractical for real-world deployment, especially in applications requiring high detection rates and low false alarms across multiple languages and channel conditions.

Innovation Solution

A method using machine learning to predict speech recognition performance by computing a Feature of Merit (FOM) score based on relevant features such as phonemes, syllables, and pronunciation, allowing for the creation of a mapping that can generalize to new keywords, enabling accurate prediction without extensive processing.

Engineering Contradictions & Design Principles

VSEngineering Contradiction Analysis

1Measurement precision

If speech recognition accuracy is measured by processing large audio files, then measurement precision is improved, but loss of time and productivity deteriorate

Engineering Contradiction:
Improverecognition accuracyVSAvoidprocessing time
Core Design Contradiction:
Measurement precisionVSLoss of time

Solution Approach 1:

The system performs preliminary analysis by extracting relevant features (phoneme counts, syllable structures, pronunciation patterns) from keywords before actual speech recognition deployment. This preliminary feature extraction creates a foundation for accurate prediction without requiring full audio processing, thus saving time while maintaining measurement precision.

Inventive Principle:
Principle #10Preliminary action

Solution Approach 2:

The invention extracts only the essential features needed for accuracy prediction (phoneme characteristics, syllable patterns, pronunciation features) from the complete audio processing pipeline. By taking out just these critical elements and using them to train a prediction model, the system achieves accurate recognition predictions without processing entire audio files, thereby reducing time loss.

Inventive Principle:
Principle #2Taking out (Extraction)

2Reliability

If speech recognition performance is evaluated through full audio file processing, then reliability is improved, but productivity and resource consumption worsen

Engineering Contradiction:
Improvedetection accuracyVSAvoidprocessing efficiency
Core Design Contradiction:
ReliabilityVSProductivity

Solution Approach 1:

Instead of processing actual audio files to evaluate speech recognition performance, the system creates a predictive model that copies the essential recognition patterns from training data. This model then predicts performance on new keywords without requiring actual audio processing, maintaining reliability through pattern matching while dramatically improving productivity by avoiding repetitive full-file processing.

Inventive Principle:
Principle #26Copying

Solution Approach 2:

The invention changes the evaluation parameters from processing complete audio files to analyzing extracted linguistic features (phoneme counts, syllable structures, pronunciation patterns). By transforming the input parameters from raw audio data to processed linguistic characteristics, the system achieves reliable performance evaluation with significantly improved processing efficiency.

Inventive Principle:
Principle #35Parameter changes

3Measurement precision

If recognition threshold filtering is applied to separate target words, then measurement precision is improved, but device complexity and processing requirements worsen

Engineering Contradiction:
Improveword identification accuracyVSAvoidsystem complexity
Core Design Contradiction:
Measurement precisionVSDevice complexity

Solution Approach 1:

The system performs preliminary feature extraction and training before deployment, creating a pre-computed prediction model that encodes the threshold filtering logic. During actual operation, the pre-trained model directly predicts recognition accuracy without requiring complex real-time threshold filtering operations, thus maintaining measurement precision while reducing system complexity during execution.

Inventive Principle:
Principle #10Preliminary action

Data Source

PatentEP2891147B1Method and system for predicting speech recognition performance using accuracy scores
Publication Date: 2020.08.12 INTERACTIVE INTELLIGENCE INC
  • EP2891147B1 patent drawingFigure 1
  • EP2891147B1 patent drawingFigure 2
  • EP2891147B1 patent drawingFigure 3

AI summary

A system and method are presented for predicting speech recognition performance using accuracy scores in speech recognition systems within the speech analytics field. A keyword set is selected. Figure of Merit (FOM) is computed for the keyword set. Relevant features that describe the word individually and in relation to other words in the language are computed. A mapping from these features to FOM is learned. This mapping can be generalized via a suitable machine learning algorithm and be used to predict FOM for a new keyword. In at least embodiment, the predicted FOM may be used to adjust internals of speech recognition engine to achieve a consistent behavior for all inputs for various settings of confidence values.