ACOUSTICALLY CHARACTERIZING NASAL SOUNDS OF KAMBAATA LANGUAGE

Abstract—this  paper  analyzes  acoustic  characteristics  of nasal  sounds  of  Kambaata  language  at  their  place  of
articulation  by  using  autoregressive  moving  average(ARMA)  model.  The  ARMA  model  is  an  extension  of
Linear Predictive Coding (LPC) model, which incorporatethe  zeros  of  the  transfer  function  of  vocal  tract  coupled
with  nasal  cavity.  Since  nasal  sound  production  involves the coupling of nasal cavity, so it can be modeled by ARMA
process. Sounds of minimal or nearly minimal pair words, containing  singleton  or  geminated  nasal  phoneme  at  the
initial, medial or final positions were recorded while read by  five male and  five  female native  speakers  is collected.
Formant  and  ant-formant  frequencies  of  the  collected speech has been extracted and analyzed by using one-way
ANOVA  test  to  analyze  the  differences  of  nasal  sounds. The  overall  duration  measurement  is  also  used  to
characterize the acoustic nature of nasals. It was observed  that  Kambaata  geminated  nasals  are  found  to  have
statistically longer in overall duration than their singleton conjugates for both females and males in all word positions
(initial, medial  and  final).  Anti-formant  frequencies  are found  to  have  statistically  significant  difference  for  both
female and males at all positions of the target phoneme. In future, it can be extended for noisy and live environments.