Speech Retrieval Using Confusion Networks to Reduce Index Size
Find Innovative SolutionsGenerate Solutions
Solution Overview
Problem
Conventional speech data retrieval methods face challenges in reducing the file size of inverted index tables and retrieving speech data accurately without being constrained by language models used in speech recognition, due to redundant arcs and limited retrieval performance when dealing with non-registered words.
Innovation Solution
The proposed solution involves creating a confusion network from the lattice output of speech recognition, which reduces the file size of the inverted index table and allows for high-accuracy retrieval by converting the lattice into a confusion network, enabling the inclusion of non-registered words and improving retrieval performance.
Engineering Contradictions & Design Principles
Engineering Contradiction Analysis
1Reliability
If an inverted index table is created using lattice data from speech recognition, then speech data can be retrieved, but the file size of the inverted index table becomes large due to redundant arcs
Solution Approach 1:
The patent extracts and removes redundant arcs from the lattice data to create a confusion network. Specifically, it identifies and eliminates arcs that represent the same phonetic sequence but with different word interpretations, thereby reducing the quantity of data in the inverted index table while preserving essential retrieval information.
Solution Approach 2:
The patent merges equivalent arcs in the lattice that represent the same phonetic content. By consolidating redundant representations of phonetic sequences into a single confusion network structure, it reduces the overall data volume while maintaining the ability to retrieve speech data accurately.
2Measurement precision
If speech recognition uses language models to recognize words, then recognition accuracy improves, but retrieval performance deteriorates for non-registered words
Solution Approach 1:
The patent changes the fundamental parameter of how speech data is indexed by using phonetic sequences rather than word-based indexing. This parameter change allows the system to recognize and retrieve data for phonetic patterns even when the corresponding words are not registered in the language model, thereby improving adaptability while maintaining recognition accuracy through phonetic matching.
Data Source
Figure 1~3
Figure 4~5
Figure 6
AI summary
A speech data retrieval apparatus (10) includes a speech database (1), a speech recognition unit (2), a confusion network creation unit (3), an inverted index table creation unit (4), a query input unit (6), a query conversion unit (7) and a label string check unit (8). The speech recognition unit (2) reads speech data from the speech database (1), carries out a speech recognition process with respect to the read speech data, and outputs a result of speech recognition process as a lattice in which a phoneme, a syllable, or a word is a base unit. The confusion network creation unit (3) creates a confusion network based on the output lattice and outputs the result of speech recognition process as the confusion network. The inverted index table creation unit (4) creates an inverted index table based on the output confusion network. The query input unit (6) receives a query input by a user, carries out a speech recognition process with respect to the received query, and outputs a result of speech recognition process as a character string. The query conversion unit (7) converts the output character string into a label string in which a phoneme, a syllable, or a word is a base unit. The label string check unit (8) checks the label string against the inverted index table and retrieves speech data which is included in both of the label string and the speech database (1).