Feature Vector Extraction Method by Normalizing Signal-to-Noise Ratio

摘要

When the speech recognition system is implemented in the automotive environment, the feature vector is extracted from the preprocessing stage of the speech recognition, taking into account the actual vehicle noise environment, and the word vector is used to create a word model. It is possible to perform the voice recognition, the process of obtaining the power spectrum through the fast Fourier transform (FFT) of the input voice and extracting the filter bank value by analysis using a triangular weighting function, and the filter bank extracted in the process Extracting a maximum value by taking a specific threshold value of the result value, and performing a normal cosine transform on the extracted value by performing triangular filtering to extract a normalized value of the signal-to-noise ratio; The use of complex adaptive algorithms is eliminated, minimizing the load on the recognition means, Different thresholds can be applied to different frequency bands according to sound characteristics, providing reliability in the analysis of the actual input speech with noise, and the variation that occurs when the input speech section is a non-voice section, i.e., a noise and silence section. Rejection minimizes the recognition impact of noise, providing reliability in speech recognition.