Creation of Standard Audio Speech Database in Malayalam for the Development of Voice Interactive Machine

Although audio-visual speech based application using resourced languages has achieved its acceleration towards robust outcomes, the realm of speech based application using under resourced languages has not gathered enough momentum in the research domain. One of the in-depth reason cited by the research community is the scarcity of application oriented speech database. The goal of this work is to present a new multi-application oriented speech database in Malayalam language and describing it under the background of available audio-visual speech database. This will be the first standard audio-visual speech database in Malayalam recorded in different condition to address specific problem. This database was recorded in 3 phases with unique speakers in each phase which makes it a speaker independent database. The first phase of recording creates audio-only speech database which contain 50 isolated phonemes captured in entirely two different condition one in isolated and noiseless environment and other in acoustically realistic environment. The second phase is utilized to capture audio and visual signal from 5 female speakers uttering isolated phonemes in acoustically and visually realistic environment. The third phase of recording contain audio-visual speech database recorded from 25 female and 5 male speakers uttering 50 Malayalam isolated phonemes and 207 connected words comprising of all allophonic variations in controlled environment. Each isolated phonemes and word in audio and video domain are properly segmented and labeled. The present database was the result of efforts taken in the past two years and it is expected to grow its dimension and content in recent years too.

Download