Phoneme-Sequence Index for Keyword Spotting
Find Innovative SolutionsGenerate Solutions
Solution Overview
Problem
Conventional keyword spotting methods in audio corpora are inefficient due to their reliance on lattice-based systems that impose segment length and interaction constraints, limiting the scalability and speed of query searches.
Innovation Solution
The use of a phoneme-sequence index allows for independent searches of subword units in audio files, generating a result index that maps subword units to their locations, enabling efficient keyword spotting by processing query data into multiple sequences and searching these indices for candidate locations using speech recognition algorithms.
Engineering Contradictions & Design Principles
Engineering Contradiction Analysis
1Reliability
If lattice-based systems are used for keyword spotting, then segment length and interaction constraints are imposed, but scalability and search speed are limited
Solution Approach 1:
The patent divides the audio stream into non-overlapping time frames and processes phoneme sequences independently within each frame. This segmentation allows the system to search for multiple phoneme sequences simultaneously without being constrained by lattice-based segment length and interaction rules, thereby improving search speed while maintaining reliable phoneme sequence matching.
Solution Approach 2:
The patent implements a dynamic search process where the system can adaptively adjust the search scope based on the audio content and query requirements. The phoneme sequence search is performed independently in each time frame, allowing flexible handling of variable-length phoneme sequences and enabling faster search operations compared to static lattice-based approaches.
2Reliability
If lattice-based systems are used for keyword spotting, then segment length and interaction constraints are imposed, but scalability is limited
Solution Approach 1:
By segmenting the audio into time frames and processing phoneme sequences independently, the system achieves better scalability. The independent processing allows parallel computation and efficient handling of large audio corpora, making the system adaptable to various search scenarios without being constrained by lattice-based segment length and interaction rules.
Solution Approach 2:
The patent creates a universal search mechanism that can handle multiple types of phoneme sequence queries simultaneously. The system can search for exact phoneme matches, partial matches, and variations in the same framework, providing versatile functionality that scales well across different application scenarios without requiring separate processing paths.
3Productivity
If phoneme sequences are searched independently in each time frame, then search speed is improved, but computational complexity increases
Solution Approach 1:
The patent performs preliminary processing by extracting phoneme sequences from the audio stream in advance and organizing them into time frames. This preliminary action allows the system to prepare the data structure before the actual search operation, reducing the computational complexity during the search phase while maintaining high search speed through independent processing of pre-organized phoneme sequences.
Data Source
AI summary
In some aspects, a wordspotter is used to locate occurrences in an audio corpus of each of a set of predetermined subword units, which may be phoneme sequences. To locate a query (e.g., a keyword or phrase) in the audio corpus, constituent subword units in the query are indentified and then locations of those subwords are determined based on the locations of those subword units determined earlier by the wordspotter, for example, using a pre-built inverted index that maps subword units to their locations.