Speech Recognition Phoneme Language Model Segmentation
Find Innovative SolutionsGenerate Solutions
Solution Overview
Problem
Existing human speech recognition methods face challenges in improving recognition performance and speed, particularly in accurately decoding phonemes and converting them into vocabulary due to limitations in phoneme language modeling and error correction.
Innovation Solution
The proposed solution involves a speech recognition method and apparatus that uses a phoneme language model with position-dependent phone n-grams to improve phoneme recognition accuracy and a vocabulary recognition unit that employs edit distance calculations to correct errors, dividing phoneme sequences into connection phonemes including vowels to reduce search space and enhance recognition speed.
Engineering Contradictions & Design Principles
Engineering Contradiction Analysis
1Measurement precision
If phoneme language model is used to recognize phoneme series from feature vector series, then speech recognition accuracy is improved, but recognition speed deteriorates due to complex probability calculations
Solution Approach 1:
The patent segments the phoneme series recognition process into distinct stages: feature vector extraction, phoneme probability calculation using language models, and vocabulary matching. By dividing the complex recognition process into manageable segments, the system can optimize each stage independently to balance accuracy and speed
Solution Approach 2:
The patent performs preliminary actions by pre-calculating and storing phoneme probabilities in language models before actual speech recognition. The phoneme language model is trained in advance with probability distributions for phoneme sequences, allowing the recognition system to quickly query pre-computed probabilities rather than calculating them in real-time
2Measurement precision
If edit distance calculation is performed on entire phoneme series to correct recognition errors, then vocabulary recognition accuracy is improved, but computational complexity and time increase
Solution Approach 1:
The patent applies segmentation by dividing the phoneme series into smaller units (phoneme frames or substrings) before performing edit distance calculations. Instead of calculating edit distance across the entire phoneme sequence, the system processes smaller segments independently, reducing the computational complexity from O(n²) to O(k×n²) where k is the number of segments
Solution Approach 2:
The patent implements local quality by applying different processing strategies to different parts of the phoneme series. High-probability phoneme regions may receive less intensive error correction, while low-probability or ambiguous regions undergo more rigorous edit distance analysis. This selective approach optimizes the balance between accuracy and computational resources
3Speed
If phoneme series is divided into connection phonemes including vowels for edit distance calculation, then search space is reduced improving speed, but recognition precision may deteriorate
Solution Approach 1:
The patent segments the phoneme series by identifying connection phonemes (particularly vowels) as anchor points that divide the sequence into manageable units. This segmentation reduces the search space for edit distance calculations while maintaining recognition precision by ensuring that each segment contains sufficient phonetic information through the inclusion of connection phonemes
Solution Approach 2:
The patent uses connection phonemes (vowels) as intermediary elements that facilitate the division of phoneme series. These connection phonemes serve as stable reference points that maintain the structural integrity of the segmented phoneme units, ensuring that the segmentation process does not lose critical phonetic information while reducing computational complexity
Data Source
AI summary
A speech recognition apparatus generates a feature vector series corresponding to a speech signal, and recognizes a phoneme series corresponding to the feature vector series using sounds corresponding to phonemes and a phoneme language model. In addition, the speech recognition apparatus recognizes vocabulary that corresponds to the recognized phoneme series. At this time, the phoneme language model represents connection relationships between the phonemes, and is modeled according to time-variant characteristics of the phonemes.