Accent Detection Using Concatenated Prosodic and Spectral Features
Find Innovative SolutionsGenerate Solutions
Solution Overview
Problem
Current accent detection methods in oral English learning face challenges in accurately determining the pronunciation of syllables and stress positions due to the lack of integrity in feature calculation, especially under conditions of large reverberation and background noise, and the complexity of nonlinear speech features like MFCCs.
Innovation Solution
The proposed accent detection method combines prosodic features and speech spectrum features by obtaining a prosodic feature vector and a speech spectrum feature matrix, performing concatenate and redundancy removal operations, and using a classifier to improve accuracy and efficiency.
Engineering Contradictions & Design Principles
Engineering Contradiction Analysis
1Measurement precision
If nonlinear speech features like MFCCs are used for accent detection, then the detection capability is enhanced, but the calculation complexity increases
Solution Approach 1:
The patent segments the feature extraction process into two distinct parts: prosodic features (temporal and intensity characteristics) and speech spectrum features (frequency-domain characteristics from MFCCs). This segmentation allows each feature type to be processed independently and then combined, reducing the overall calculation complexity while maintaining detection accuracy
Solution Approach 2:
The patent merges prosodic features and speech spectrum features into a unified feature vector for accent detection. By combining these two complementary feature types, the system leverages both temporal-intensity information and frequency-spectral information, achieving enhanced detection accuracy without requiring complex nonlinear feature processing alone
2Measurement precision
If comprehensive feature calculation is performed for accent detection, then the detection accuracy is improved, but the processing speed decreases
Solution Approach 1:
The patent performs preliminary feature extraction by pre-computing prosodic features (such as syllable duration, intensity contours) and speech spectrum features (MFCCs) separately before the actual accent detection. This preliminary action organizes the data in advance, allowing the final classification stage to operate more efficiently on pre-processed feature vectors
Solution Approach 2:
The patent segments the feature calculation into parallel independent streams: one for prosodic features and another for speech spectrum features. These can be computed concurrently rather than sequentially, improving processing speed while maintaining comprehensive feature coverage for accurate accent detection
3Device complexity
If traditional accent detection methods are used, then the system simplicity is maintained, but the accuracy under reverberation and noise conditions deteriorates
Solution Approach 1:
The patent merges prosodic features (which capture temporal and intensity patterns robust to noise) with speech spectrum features (which capture frequency characteristics resilient to reverberation). This combination creates a more reliable detection system that maintains accuracy under adverse acoustic conditions while adding moderate complexity
Solution Approach 2:
The patent transforms the audio signal into different parameter domains: time-domain prosodic parameters (duration, intensity) and frequency-domain spectral parameters (MFCCs). By analyzing accent characteristics across multiple parameter spaces rather than a single domain, the system achieves improved reliability under reverberation and noise while maintaining a relatively simple overall architecture
Data Source
AI summary
Disclosed are an accent detection method, an accent detection device and a non-transitory storage medium. The accent detection method includes: obtaining audio data of a word; extracting a prosodic feature of the audio data to obtain a prosodic feature vector; generating a spectrogram based on the audio data to obtain a speech spectrum feature matrix; performing a concatenate operation on the prosodic feature vector and the speech spectrum feature matrix to obtain a first feature matrix, and performing a redundancy removal operation on the first feature matrix to obtain a second feature matrix; and classifying the second feature matrix by a classifier to obtain an accent detection result of the audio data.