A Survey on Single and Multi-Speaker Voice Recognition In Android

Computer Engineering

For communication voice is the most preferred natural form and the foremost goal of developing this system is to make interaction with machines as easy as possible. Voice recognition and speech recognition are considered same however, they have distinct meanings. Speech recognition detects only the words, while the voice recognition disregards the language or anything linguistic for that matter and detects the identity of the person speaking. In this given paper, we discuss two parts, In the first part of the paper, we discuss the means of processing the speech signal using the Mel Frequency Cepstral Coefficient and the basics of voice characteristics. At hardware level the problem of voice recognition had been already implemented, but here we propose a solution with higher level of accuracy on the software end. Moreover, both single speaker and multi-speaker single recognition application being developed on android is discussed in the second part. The multispeaker feature enables the system to identify more than one person speaking by splitting the input into smaller blocks and treating each block as a single speaker identification problem. For the case of single speaker recognition, the algorithm is able to correctly identify the individual 90.3% of the time. In the multispeaker recognition case, up to 3 individuals are correctly identified roughly 40% of the time. For a database consisting of 415 samples, the average time for identification is 220ms.

This work is licensed under a Creative Commons Attribution-ShareAlike 4.0 International License.