Probabilistic Word Substitution Model for Speech Text Structuring

Resolve Bottlenecks,
Find Innovative SolutionsGenerate Solutions

Solution Overview

Problem

Automatic speech-to-text systems produce unstructured text that often requires significant manual editing to conform to standard document formats and styles, particularly in applications like medical dictation, where discrepancies in punctuation, formatting, and terminology necessitate substantial transcriptionist intervention.

Innovation Solution

A probabilistic word substitution model transforms unstructured speech recognition text into structured document text by establishing association probabilities between source speech recognition text and target document text, using dynamic programming and forward-backward search to generate the most likely structured text sequence, trained on archived dictations and corresponding documents.

Engineering Contradictions & Design Principles

VSEngineering Contradiction Analysis

1Measurement precision

If speech recognition systems produce verbatim transcription, then accuracy of literal speech is improved, but the text requires significant manual editing to conform to standard document formats

Engineering Contradiction:

Improveaccuracy of speech transcriptionVSAvoidmanual editing time

Core Design Contradiction:

Measurement precisionVSLoss of time

Solution Approach 1:

The patent introduces transformation modeling as an intermediary process between speech recognition and final document production. This mediator automatically transforms verbatim transcripts into structured documents by applying domain-specific rules for medical dictation, eliminating the need for manual editing while preserving transcription accuracy.

Inventive Principle:

Principle #24Intermediary (Mediator)

Solution Approach 2:

The system enables self-service by allowing the speech recognition output to automatically transform into structured documents through predefined transformation models. The system serves itself by detecting disfluencies, removing filler words, and reformatting content without human intervention, reducing manual editing time to near zero.

Inventive Principle:

Principle #25Self-service

2Manufacturing precision

If transcriptionists manually edit speech recognition output, then document quality and format compliance are improved, but productivity is reduced due to significant editing requirements

Engineering Contradiction:

Improvedocument format complianceVSAvoidtranscription processing speed

Core Design Contradiction:

Manufacturing precisionVSProductivity

Solution Approach 1:

The transformation modeling system performs self-service by automatically converting unstructured speech recognition text into structured documents compliant with medical standards. The system handles formatting, punctuation, and terminology transformation autonomously, achieving both high document quality and improved productivity without manual editing.

Inventive Principle:

Principle #25Self-service

Solution Approach 2:

The system applies parameter changes by transforming the structural parameters of the text from unstructured verbatim format to structured document format. This includes changing punctuation patterns, sentence structures, and organizational hierarchy to match medical document standards, achieving format compliance automatically.

Inventive Principle:

Principle #35Parameter changes

3Loss of information

If speech recognition transcribes all spoken content including disfluencies, then completeness of transcription is improved, but the resulting text contains false starts and repetitions that require editing

Engineering Contradiction:

Improvecompleteness of speech contentVSAvoidtext readability

Core Design Contradiction:

Loss of informationVSEase of operation

Solution Approach 1:

The transformation model extracts and removes unwanted elements from the speech transcript, including false starts, repetitions, and filler words. By selectively extracting only the meaningful content while discarding disfluencies, the system maintains completeness of information while dramatically improving text readability and reducing editing requirements.

Inventive Principle:

Principle #2Taking out (Extraction)

Solution Approach 2:

The system converts the harm of disfluencies into benefit by using the presence of false starts and repetitions as indicators for transformation. The transformation model identifies these problematic patterns and automatically corrects them, turning the originally harmful verbatim transcription into a benefit by producing clean, readable text that maintains all essential information.

Inventive Principle:

Principle #22Blessing in disguise (Convert harm into benefit)

Data Source

PatentUS7813929B2Automatic editing using probabilistic word substitution models

Publication Date: 2010.10.12 MICROSOFT TECHNOLOGY LICENSING LLC

AI summary

An input sequence of unstructured speech recognition text is transformed into output structured document text. A probabilistic word substitution model is provided which establishes association probabilities indicative of target structured document text correlating with source unstructured speech recognition text. The input sequence of unstructured speech recognition text is looked up in the word substitution model to determine likelihoods of the represented structured document text corresponding to the text in the input sequence. Then, a most likely sequence of structured document text is generated as an output.