Amplitude independent feature extraction for effective speech retrieval

Since storing and accessing a digital data is becoming cheaper, the amount of data in multimedia format is continuously increasing. Indexing and organizing of this data is quite important. There is a blended amount of data exists in the form of audio, such as teleconferences, broadcasted news, reports, interviews, radio broadcasting, etc. To deal with this critical data, users need an effective system for spoken word retrieval. Based on the needs, this paper introduces a novel approach of amplitude independent content-based retrieval system for audio format through speech data. In this study, the input is audio query which is getting through real time or existing data. According to the previous studies, speech retrieval is done with the help of transcribed units, which causes lack of system accuracy. This research work is proposed to increase the efficiency of the retrieval system. Features are used to match the acoustic similarities between the queries with existing documents in the corpus. Furthermore, to fulfill the system efficiency and accuracy requirements, Mel Frequency Cepstral Coefficients and formant feature extraction will be used for feature extraction. The resultant average output of these features will be provided to support vector machine for classification purpose, by considering every average value as a single feature value.