mwang-lifesize / speakeridentificationneuralnetworks

⇨ The Speaker Recognition System consists of two phases, Feature Extraction and Recognition. ⇨ In the Extraction phase, the Speaker's voice is recorded and typical number of features are extracted to form a model. ⇨ During the Recognition phase, a speech sample is compared against a previously created voice print stored in the database. ⇨ The highlight of the system is that it can identify the Speaker's voice in a Multi-Speaker Environment too. Multi-layer Perceptron (MLP) Neural Network based on error back propagation training algorithm was used to train and test the system. ⇨ The system response time was 74 µs with an average efficiency of 95%.

#Speaker Identification in Multispeaker Environment using Deep Neural Networks

Abstract

Human beings are capable of performing unfathomable tasks. A human being is able to focus on a single person’s voice in an environment of simultaneous conversations. We have tried to emulate this particular skill through an artificial intelligence system. Our system identifies an audio file as a single or multi-speaker file as the first step and then recognizes the speaker(s). Our approach towards the desired solution was to first conduct pre-processing of the audio (input) file where it is subjected to reduction and silence removal, framing, windowing and DCT calculation, all of which is used to extract its features. Mel Frequency Cepstral Coefficients (MFCC) technique was used for feature extraction. The extracted features are then used to train the system via neural networks using the Error Back Propagation Training Algorithm (EBPTA). One of the many applications of our model is in biometric systems such as telephone banking, authentication and surveillance.

Keywords: Speaker identification, neural network, Multi- Speaker, Mel Frequency Cepstral Coefficients (MFCC).

Research Paper published in Springer Journal.

For more details: download file ResearchPaper.pdf, projectreport