Spoken Language Processing Document Scoring via ASR Confidence

Solution Overview

Problem

Real-time Spoken Language Processing systems face challenges in obtaining accurate results before the end of speech, as existing methods struggle to effectively utilize initial words recognized by automatic speech recognition systems to identify relevant documents.

Innovation Solution

A computer-implemented method that calculates a score for each document in a document set using the first n words of output from an ASR system, updates the score with a prior distribution and a weight for interpolation based on the confidence score of the ASR output, and identifies a target document among the set.

Engineering Contradictions & Design Principles

VSEngineering Contradiction Analysis

1Measurement precision

If the system waits for complete speech recognition before identifying target documents, then the accuracy of document identification improves, but the response time increases and real-time processing capability deteriorates

Engineering Contradiction:

Improvedocument identification accuracyVSAvoidresponse time

Core Design Contradiction:

Measurement precisionVSLoss of time

Solution Approach 1:

The system performs preliminary document scoring using prior distributions before complete speech recognition is available. It calculates initial scores based on prior document importance distributions, then progressively updates these scores as speech recognition results become available, enabling early document identification while maintaining accuracy through iterative refinement.

Inventive Principle:

Principle #10Preliminary action

Solution Approach 2:

The system dynamically adjusts the weighting between prior distributions and speech recognition results over time. It uses a time-dependent interpolation factor that transitions from relying heavily on prior distributions at early stages to progressively incorporating speech recognition results as they become available, optimizing the balance between early response and final accuracy.

Inventive Principle:

Principle #15Dynamics

2Productivity

If the system uses only the first n words from ASR output for document scoring, then the processing speed improves and real-time results are achieved, but the measurement precision of document relevance deteriorates

Engineering Contradiction:

Improveprocessing speedVSAvoiddocument relevance accuracy

Core Design Contradiction:

ProductivityVSMeasurement precision

Solution Approach 1:

The system performs preliminary document scoring using prior distributions before complete speech recognition is available. It calculates initial scores based on prior document importance distributions, then progressively updates these scores as speech recognition results become available, enabling early document identification while maintaining accuracy through iterative refinement.

Inventive Principle:

Principle #10Preliminary action

Solution Approach 2:

The system implements a feedback mechanism where speech recognition confidence scores are used to adjust the weighting between prior distributions and current speech results. As speech recognition confidence increases over time, the system progressively reduces reliance on prior distributions and increases reliance on actual speech content, ensuring accurate document identification even when using limited initial words.

Inventive Principle:

Principle #23Feedback

3Stability of the object's composition

If the system relies heavily on prior document distributions for scoring, then the stability of document selection improves, but the adaptability to actual speech content deteriorates

Engineering Contradiction:

Improvedocument selection stabilityVSAvoidadaptability to speech content

Core Design Contradiction:

Stability of the object's compositionVSAdaptability or versatility

Solution Approach 1:

The system dynamically adjusts the weighting between prior distributions and speech recognition results over time. It uses a time-dependent interpolation factor that transitions from relying heavily on prior distributions at early stages to progressively incorporating speech recognition results as they become available, optimizing the balance between early response and final accuracy.

Inventive Principle:

Principle #15Dynamics

Solution Approach 2:

The system implements a feedback mechanism where speech recognition confidence scores are used to adjust the weighting between prior distributions and current speech results. As speech recognition confidence increases over time, the system progressively reduces reliance on prior distributions and increases reliance on actual speech content, ensuring accurate document identification even when using limited initial words.

Inventive Principle:

Principle #23Feedback

Data Source

PatentUS10152507B2Finding of a target document in a spoken language processing

Publication Date: 2018.12.11 INTERNATIONAL BUSINESS MACHINE CORPORATION

AI summary

Methods and systems are provided for finding a target document in spoken language processing. One of the methods includes calculating a score of each document in a document set, in response to a receipt of first n words of output of an automatic speech recognition (ASR) system, n being equal or greater than zero. The method further includes reading a prior distribution of each document in the document set from a memory device, and updating, for each document in the document set, the score, using the prior distribution, and a weight for interpolation, the weight for interpolation being set based on a confidence score of output of the ASR system. The method additionally includes finding a target document among the document set, based on the updated score of each document.