A Small Vocabulary Automatic Filipino Speech Profanity Suppression System Using Hybrid Hidden Markov Model/Artificial Neural Network (HMM/ANN) Keyword Spotting Framework

This paper describes an implementation of speech recognition that recognizes and suppresses ten (10) defined profane and vulgar Filipino words. The adapted speech recognition architecture was that of the Oregon Graduate Institute's (OGI) Center for Spoken Language and Learning (CSLU). It utilizes a hybrid Hidden Markov Model / Artificial Neural Network (HMM/ANN) keyword spotting framework. The feature extraction method used was Mel-Frequency Cepstral Coefficients (MFCC). The ANN is a 3-layer feed-forward neural network using Multi-Layer Perceptron (MLP). In recognizing the words, an HMM decoder was used which implemented the Viterbi Beam Search Algorithm. Whenever a profane word was recognized, it would be replaced with a constant frequency tone. The training and testing data (recordings) were gathered from 30 random (15 male and 15 female) Filipino speakers.