<p>This paper presents a method for extracting English vowel features based on an interpolatory wavelet pyramid algorithm. In the process of wavelet decomposition of English vowels, it was found that the parameters of the wavelet space can uniquely determine and describe vowel signals. However, when classical wavelet theory is applied to audio analysis, discrete integral formulas are often used to substitute for continuous integrals to obtain wavelet coefficients. The discrete integral is an approximate expression of the continuous integral, which often leads to significant numerical errors in calculations. This can result in substantial deviations in the extracted signal features. To address this issue, an algorithm for extracting speech signal features is proposed, that integrates the Mallat pyramid algorithm with interpolatory conjugate filters. The main information of audio signals can be captured at high resolutions through our algorithm, since it can obtain wavelet coefficients directly from samples without calculation errors due to interpolatory wavelets. Thus, the approximation errors from discrete integration can be avoided in our algorithm. In experiments, this algorithm was applied to datasets, and feature extraction was performed on audio signals; the extracted information was verified, so as to demonstrate the effectiveness and feasibility of this method. The experiment results show that our algorithm can well determine wavelet coefficients at very high resolutions for vowel signals, due to its ability to avoid discrete integration. Hence, the audio energies in wavelet spaces can be well characterized.</p>