Speech Recognition Text Processing via Attention Interaction
Find Innovative SolutionsGenerate Solutions
Solution Overview
Problem
The readability of texts directly output by speech recognition is poor due to lack of punctuation, numerical formatting, and redundant words, leading to increased reading difficulty.
Innovation Solution
A speech recognition text processing method and apparatus that performs punctuation addition, numerical normalization, and word removal processing using an attention interaction layer to improve text quality, incorporating deep learning models and preset processing results to enhance data diversity and accuracy.
Engineering Contradictions & Design Principles
Engineering Contradiction Analysis
1Ease of operation
If speech recognition directly outputs text without processing, then processing speed is fast, but text readability is poor
Solution Approach 1:
The patent applies preliminary action by performing punctuation addition, numerical normalization, and word removal processing on speech recognition text before final output. The text processing module pre-processes the raw recognition results through multiple processing units that add punctuation marks, normalize numerical expressions, and remove redundant words in advance, so that the text is ready for direct display without requiring additional processing steps later.
2Ease of operation
If multiple processing steps are applied to speech recognition text, then text quality is improved, but system complexity increases
Solution Approach 1:
The patent merges multiple text processing functions into a single integrated text processing module. The module combines punctuation addition processing, numerical normalization processing, and word removal processing into one unified structure that receives speech recognition text and outputs processed text through coordinated operation of internal processing units, thereby reducing system complexity while maintaining comprehensive text quality improvement.
Solution Approach 2:
The text processing module exhibits multi-functionality by simultaneously performing multiple types of processing on the input text. A single module handles punctuation addition, numerical normalization, and word removal through its internal processing units, making the system more versatile and reducing the need for separate dedicated modules for each processing task.
3Measurement precision
If word removal processing is performed through attention mechanism, then processing accuracy is improved, but computational resources increase
Solution Approach 1:
The patent applies partial action by implementing word removal processing that focuses only on the most critical aspects of text refinement. The attention-based processing unit selectively identifies and removes redundant words and phrases while preserving important content, applying computational resources only where needed rather than processing every word uniformly, thereby achieving good accuracy with reduced computational overhead.
Data Source
AI summary
A speech recognition text processing method includes acquiring an original text obtained by speech recognition, and performing at least one of punctuation addition processing and numerical normalization processing on the original text to obtain an intermediate result. The method further includes performing word removal processing on the original text through an attention interaction layer based on the intermediate result to obtain a processed text of the original text.