Disfluency Removal via Voice-Chat Parallel Corpus
Find Innovative SolutionsGenerate Solutions
Solution Overview
Problem
Existing machine learning techniques have not effectively addressed the issue of removing disfluencies from voice transcripts, which distort syntax and semantics, making it difficult to generate accurate transcripts from voice conversations.
Innovation Solution
A method involving the creation of a disfluency remover model trained using a parallel corpus of voice and chat transcripts, where voice transcripts with disfluencies are matched with disfluency-free chat transcripts to learn the relationship between the two, using neural network layers to encode sentences and generate vector pairs for training the model.
Engineering Contradictions & Design Principles
Engineering Contradiction Analysis
1Measurement precision
If machine learning techniques are used to improve voice transcript quality, then transcript accuracy can be improved, but disfluencies in voice transcripts cannot be effectively removed
Solution Approach 1:
The patent uses chat transcripts as an intermediary medium to train the disfluency remover model. Chat transcripts serve as a bridge between voice transcripts with disfluencies and the desired clean output, providing semantically equivalent but disfluency-free training examples that enable the model to learn effective disfluency removal patterns
Solution Approach 2:
The patent performs preliminary actions by creating a parallel corpus of voice and chat transcripts before training the model. This pre-processing step of assembling and aligning training data in advance allows the model to be trained on comprehensive examples of disfluency patterns, improving its effectiveness in removing disfluencies from new transcripts
2Ease of operation
If disfluencies are removed from voice transcripts, then transcript clarity is improved, but the process requires complex training data preparation and model development
Solution Approach 1:
The patent uses chat transcripts as copies or equivalents of voice transcripts. These chat transcripts replicate the semantic content and structure of voice transcripts but without disfluencies, providing ready-made training examples that simplify the model development process while maintaining transcript clarity
Data Source
AI summary
A method may including obtaining a voice transcript corpus and a chat transcript corpus, extracting voice transcript sentences from the voice transcript corpus and chat transcript sentences from the chat transcript corpus, encoding, by a series of neural network layers, the voice transcript sentences to generate voice sentence vectors, encoding, by the series of neural network layers, the chat transcript sentences to generate chat sentence vectors, determining, for each voice sentence vector, a matching chat sentence vector to obtain matching voice-chat vector pairs, and adding, to a parallel corpus, matching voice-chat sentence pairs using the matching voice-chat vector pairs. Each of the matching voice-chat sentence pairs may include a voice transcript sentence and a matching chat transcript sentence. The method may further include training a disfluency remover model using the parallel corpus.