A novel method based on cross correlation maximization, for pattern matching by means of a single parameter. Application to the human voice

TL;DR

This paper introduces a new cross correlation maximization technique using a single parameter for pattern matching in time series, demonstrated on Spanish vowel voice recognition, showing effective differentiation with minimal data length.

Contribution

The work presents a novel statistical method for pattern matching that simplifies similarity measurement to a single parameter, applied successfully to voice recognition.

Findings

Method accurately distinguishes vowels with minimal data (~2 ms).

Effective pattern matching achieved using a single parameter.

Demonstrates potential for real-time voice recognition applications.

Abstract

This work develops a cross correlation maximization technique, based on statistical concepts, for pattern matching purposes in time series. The technique analytically quantifies the extent of similitude between a known signal within a group of data, by means of a single parameter. Specifically, the method was applied to voice recognition problem, by selecting samples from a given individual recordings of the 5 vowels, in Spanish. The frequency of acquisition of the data was 11.250 Hz. A certain distinctive interval was established from each vowel time series as a representative test function and it was compared both to itself and to the rest of the vowels by means of an algorithm, for a subsequent graphic illustration of the results. We conclude that for a minimum distinctive length, the method meets resemblance between every vowel with itself, and also an irrefutable difference with the rest of the vowels for an estimate length of 30 points (~2 10-3 s).