Evaluation of MFCC for emotion identification in Hindi speech

Abstract

Emotion is an important cue to express the feeling of one's expression. It is also an important factor in expressive speech synthesis. If one wants to understand the meaning of an utterance, it must be in proper emotion otherwise semantically utterance will go to the wrong direction and give wrong result. It is also important to identify the emotion of the user to understand the feeling of an utterance. Keeping the facts in view, an algorithm is discussed to identify the emotional style of an utterance. Identification process adopts the concept of Feature Extraction based on Mel - Frequency cepstrum coefficients (MFCC). For performing experiments and verification of results, an emotion rich speech database is prepared with the help of 20 native speakers. Each speaker has been given 25 sentences and asked to generate utterances in five expressive states "neutral", "sadness", "anger", "surprise" and "happiness". These sentences are recorded in Normal laboratory conditions with 44.1 KHz sampling rate and 16-bit precision with mono channel. Auto spectral subtraction and Multiband noise gating technique are used for noise reduction.