Speech Intent Extraction Without ASR Lexicons
Find Innovative SolutionsGenerate Solutions
Solution Overview
Problem
Conventional speech recognition systems require building an automated speech recognition (ASR) system for each language and dialect, which is a tedious and expensive process due to the need for manual supervision and pronunciation lexicon generation.
Innovation Solution
The system processes raw sound signals by deconstructing them into small acoustic units (graphemes) and using machine learning models to determine user intent and characteristics without relying on ASR engines, employing pre-trained connectionist temporal classification models to map these units to candidate representations and labels.
Engineering Contradictions & Design Principles
Engineering Contradiction Analysis
1Measurement precision
If conventional ASR systems are used to understand user speech, then speech-to-text conversion can be achieved, but the system requires building separate ASR systems for each language and dialect which is tedious and expensive
Solution Approach 1:
The patent segments the speech processing task into two independent components: acoustic feature extraction (sound signal analysis) and semantic interpretation (intent detection). By separating these functions, the system eliminates the need for language-specific ASR systems while maintaining accurate speech understanding across multiple languages and dialects through a single universal processor.
Solution Approach 2:
The patent creates a universal speech processing system that can handle multiple languages and dialects simultaneously. The processor is designed to extract acoustic features and determine user intent without requiring separate ASR systems for each language, making the system multi-functional and language-agnostic.
2Measurement precision
If manual supervision and pronunciation lexicon generation are performed for each language, then ASR system accuracy is improved, but the development time and cost increase significantly
Solution Approach 1:
The patent enables the system to process and interpret speech from multiple languages without requiring manual supervision for each language. The processor automatically extracts acoustic features and determines intent, making the system self-sufficient and eliminating the need for tedious manual lexicon generation and supervision for each language and dialect.
3Adaptability or versatility
If pronunciation lexicons are generated for multiple dialects and accents, then speech coverage is improved, but the process becomes more tedious and expensive
Solution Approach 1:
Instead of adapting the system to each specific language and dialect by creating pronunciation lexicons, the patent inverts the approach: it extracts language-agnostic acoustic features from the sound signal and determines intent directly, bypassing the need for language-specific processing. This makes the system inherently adaptable to any language or dialect without requiring separate lexicon generation for each.
Data Source
AI summary
Systems and methods are disclosed herein for discerning aspects of user speech to determine user intent and/or other acoustic features of a sound input without the use of an ASR engine. To this end, a processor may receive a sound signal comprising raw acoustic data from a client device, and divides the data into acoustic units. The processor feeds the acoustic units through a first machine learning model to obtain a first output and determines a first mapping, using the first output, of each respective acoustic unit to a plurality of candidate representations of the respective acoustic unit. The processor feeds each candidate representation of the plurality through a second machine learning model to obtain a second output, determines a second mapping, using the second output, of each candidate representation to a known condition, and determines a label for the sound signal based on the second mapping.