Voice Processing Apparatus Real-Time Spectrum Normalization
Find Innovative SolutionsGenerate Solutions
Solution Overview
Problem
Current voice recognition systems take longer to complete due to the need to wait for the end of an utterance interval before calculating the average spectrum, which reduces the efficiency of voice recognition and can be affected by noise and changes in the acoustic environment.
Innovation Solution
A voice processing apparatus and method that acquires and normalizes audio signal spectra in real-time using an average spectrum calculated from previous utterances or stored spectra, allowing for immediate voice recognition without waiting for the end of an utterance and reducing noise interference by detecting changes in the acoustic environment.
Engineering Contradictions & Design Principles
Engineering Contradiction Analysis
1Measurement precision
If the average spectrum is calculated after waiting for the end of an utterance interval, then the voice recognition accuracy is improved, but the time from start of utterance to end of voice recognition becomes longer
Solution Approach 1:
The system pre-calculates and stores average spectra during non-utterance intervals before they are needed for recognition. When an utterance occurs, the pre-computed average spectrum is immediately available for normalization, eliminating the waiting time while maintaining accuracy. This is implemented by the average spectrum calculation unit that operates during silence periods and stores results for later use.
2Measurement precision
If the average spectrum is calculated using data from the current utterance, then the voice recognition accuracy is improved, but the time required for recognition increases
Solution Approach 1:
The system performs spectrum averaging during non-utterance intervals in advance, storing the computed average spectra. During actual utterances, these pre-computed averages are immediately applied for normalization without requiring real-time calculation, thus maintaining high recognition accuracy while enabling fast real-time processing.
Solution Approach 2:
The system dynamically switches between using pre-computed average spectra from non-utterance intervals and real-time spectra during utterances. The selection is adaptive based on the utterance state, allowing the system to optimize between accuracy and speed depending on the current operational context.
3Adaptability or versatility
If the voice recognition system processes acoustic signals with superimposed noises, then the system can operate in noisy environments, but the voice recognition rate is reduced
Solution Approach 1:
The system extracts and removes noise components from the acoustic signal by comparing the signal spectrum against the average spectrum (which represents primarily background noise). The normalization process subtracts the average spectrum from the signal spectrum, effectively extracting and eliminating noise components while preserving the voice signal.
Solution Approach 2:
The average spectrum acts as an intermediary that mediates between the noisy acoustic signal and the voice recognition process. By introducing this intermediate representation of background noise, the system can systematically remove noise influences before the signal is processed by the voice recognition algorithm.
4Measurement precision
If the system waits for the end of utterance to calculate average spectrum, then the noise influence is reduced, but the processing efficiency decreases
Solution Approach 1:
The system performs the computationally intensive average spectrum calculation during non-utterance intervals in advance, storing the results for immediate use during utterances. This preliminary action separates the heavy computation from the time-critical recognition process, achieving both noise reduction and high processing efficiency.
Solution Approach 2:
The system continuously updates and maintains the average spectrum during non-utterance intervals, ensuring that the noise characterization is always current and accurate. This continuous preparation of noise models during idle periods ensures that when utterances occur, the system can immediately apply up-to-date noise reduction without interruption.
Data Source
AI summary
A voice processing apparatus includes: a feature amount acquisition unit configured to acquire a spectrum of an audio signal for each frame; an utterance state determination unit configured to determine an utterance state for each frame on the basis of the audio signal; and a spectrum normalization unit configured to calculate a normalized spectrum in a current utterance by normalizing a spectrum for each frame in the current utterance using at least an average spectrum acquired until the present time.