Speech Retrieval Using Phonetic and Word Recognition Segmentation
Find Innovative SolutionsGenerate Solutions
Solution Overview
Problem
Current speech retrieval methods face challenges in achieving high precision and recall simultaneously, particularly in large-scale call center speech data analysis, where re-executing speech recognition for each speaker is impractical due to time and cost constraints, and phoneme-based recognition lacks sufficient language information for accurate keyword retrieval.
Innovation Solution
A method that acquires a keyword as a character string or phoneme/syllable string, detects coinciding segments by comparing word speech recognition results with phoneme/syllable recognition results, and calculates an evaluation value using edit distance or N-best recognition ranks to output segments exceeding a predetermined threshold, thereby enhancing precision without re-executing speech recognition.
Engineering Contradictions & Design Principles
Engineering Contradiction Analysis
1Measurement precision
If word speech recognition is performed for all speech data to achieve high precision keyword retrieval, then retrieval precision is improved, but processing time and cost increase significantly
Solution Approach 1:
The patent segments the speech recognition process into two stages: first performing fast phoneme/syllable recognition on all speech data to create initial candidates, then performing more accurate word-level recognition only on segments that match phonetic patterns. This segmentation allows the system to maintain high precision while reducing overall processing time by avoiding complete re-recognition of all data.
Solution Approach 2:
The patent performs preliminary phoneme/syllable recognition on all speech data before keyword retrieval to establish a foundation of phonetic information. This preliminary action creates ready-to-use phonetic transcripts that can be quickly matched against keyword phonetic patterns, eliminating the need for time-consuming complete re-recognition during the retrieval process.
2Productivity
If phoneme-based speech recognition is used to reduce processing time, then productivity is improved, but retrieval precision decreases due to insufficient language information
Solution Approach 1:
The patent merges the advantages of phoneme-based and word-based recognition by combining phoneme/syllable recognition results with word-level linguistic information. The system uses phonetic patterns from phoneme recognition for fast initial matching, then incorporates word-level language models and dictionaries to refine results, achieving both speed and precision.
Solution Approach 2:
The patent changes the recognition parameter dynamically based on the retrieval stage: using phoneme/syllable-level granularity for initial fast retrieval, then switching to word-level granularity for precise keyword matching. This parameter change allows the system to optimize between processing speed and retrieval precision at different stages of the retrieval process.
3Measurement precision
If speech recognition is re-executed for each speaker to adjust recall and precision, then retrieval accuracy is improved, but time consumption increases significantly
Solution Approach 1:
The patent creates a reusable phonetic transcript copy of the speech data that can be quickly matched against keyword phonetic patterns without re-executing the original speech recognition. This copy serves as a ready-to-use reference that enables fast retrieval while maintaining accuracy, eliminating the need for time-consuming re-recognition for each speaker.
Solution Approach 2:
The patent performs the speech recognition action in advance to create phonetic transcripts before the retrieval process. These pre-generated transcripts serve as ready-to-use references that can be quickly matched against keywords, eliminating the need for time-consuming re-execution during retrieval operations.
4Reliability
If comprehensive speech recognition is performed on large-scale call center data, then retrieval completeness is improved, but cost and time requirements become impractical
Solution Approach 1:
The patent segments the large-scale speech data processing into two parts: first processing all data at the phoneme/syllable level to create comprehensive phonetic indices, then processing only relevant segments at the word level for keyword retrieval. This segmentation enables complete coverage of large datasets while maintaining manageable processing time and cost.
Solution Approach 2:
The patent performs comprehensive phoneme/syllable recognition as a preliminary action on all speech data to create a complete phonetic index. This preliminary processing of the entire dataset enables fast and complete keyword retrieval without requiring expensive and time-consuming re-recognition of all data for each query.
Data Source
AI summary
A method for speech retrieval includes acquiring a keyword designated by a character string, and a phoneme string or a syllable string, detecting one or more coinciding segments by comparing a character string that is a recognition result of word speech recognition with words as recognition units performed for speech data to be retrieved and the character string of the keyword, calculating an evaluation value of each of the one or more segments by using the phoneme string or the syllable string of the keyword to evaluate a phoneme string or a syllable string that is recognized in each of the detected one or more segments and that is a recognition result of phoneme speech recognition with phonemes or syllables as recognition units performed for the speech data, and outputting a segment in which the calculated evaluation value exceeds a predetermined threshold.