Speech Evaluation Apparatus Pitch Frequency Detection Accuracy
Find Innovative SolutionsGenerate Solutions
Solution Overview
Problem
Existing pitch estimation techniques inaccurately determine the fundamental pitch frequency due to distortion in voice waveforms, often recognizing voices with low inflection as having high inflection, as high correlation is found at frequencies that are integral multiples of the fundamental pitch frequency.
Innovation Solution
A speech evaluation apparatus that includes a frequency analysis unit, spectrum transforming unit, correlation calculating unit, and control unit, which analyzes input voice spectra, transforms them based on a provisional change amount, calculates correlation values, and corrects the change amount to accurately determine the fundamental pitch frequency even with distortion, by comparing the input spectrum with a processed spectrum.
Engineering Contradictions & Design Principles
Engineering Contradiction Analysis
1Measurement precision
If pitch estimation is performed by detecting peak of voice spectrum using correlation between sections, then pitch frequency can be estimated, but distortion in voice waveform causes incorrect detection at integral multiple frequencies
Solution Approach 1:
The patent dynamically adjusts the search range for pitch frequency based on the change amount from the previous frame. Instead of searching the entire frequency range, the system limits the search to a dynamic range centered around the previously detected pitch frequency, adapted to the estimated change amount. This dynamic adjustment maintains detection accuracy while preventing false detection at integral multiple frequencies, even when voice waveform distortion occurs.
Solution Approach 2:
The system uses feedback from the previously detected pitch frequency and its change amount to constrain the current pitch detection. The change amount calculated from the previous frame is fed back to limit the search range of the current frame, creating a feedback loop that prevents drift and false detection. This feedback mechanism ensures that even under distortion, the pitch detection remains reliable by preventing jumps to integral multiple frequencies.
2Measurement precision
If the search range for pitch frequency is limited based on change amount from previous frame, then detection accuracy improves, but the system becomes more complex
Solution Approach 1:
The system performs preliminary estimation of the pitch frequency change amount using autocorrelation on the voice waveform before conducting the spectrum peak detection. This preliminary action provides a predicted change range that constrains the subsequent pitch detection process. By preparing the search range in advance based on the previous frame's data, the system reduces the computational complexity of the main detection process while maintaining high accuracy.
Data Source
AI summary
A speech evaluation apparatus includes a memory, and a processor coupled to the memory and configured to generate a first input spectrum obtained by frequency transforming a first signal that is a signal of a first period, generate a second input spectrum obtained by frequency transforming a second signal that is the signal of a second period earlier than the first period, generate a processed spectrum obtained by transforming frequency of the second input spectrum based on a change ratio set in advance, calculate a correlation value between the first input spectrum and the processed spectrum, and determine a change amount of pitch frequency from the first signal to the second signal based on the change ratio and the correlation value.