Priyankoo Sarmah

Consonant-vowel unit recognition using dominant aperiodic and transition region detection

Abstract This work reports a method of Consonant-Vowel (CV) unit recognition by detecting the Dominant Aperiodic component Regions (DARs) and by predicting the Duration of Transition Regions (DTRs) in speech. DAR detection is performed using complementary information from source and vocal tract. While source information is extracted using sub-fundamental frequency filtering of speech, vocal tract information is extracted using a) Dominant Resonant Frequency (DRF) and b) High to Low Frequency component Ratio (HLFR), computed from Hilbert envelope of Numerator Group Delay (HNGD) spectrum of zero-time windowed signal. The DTR is predicted by using vocal tract constriction information. Subsequently, detected DARs and predicted DTRs are compared with manually marked regions and finally used for CV unit recognition of Indian languages. Conventionally, CV unit recognition is performed by anchoring the Vowel Onset Point (VOP) and assuming fixed durations for transition and consonant regions on either side of the VOP. However, in speech, the duration of transition and consonantal regions vary depending on the type of consonants and vowels. In the proposed method, the use of dynamic values for consonant duration and transition regions have resulted in better consonant recognition improving CV unit recognition.

Dual channel signal analysis of oral and nasal consonants

In this study we captured oral and nasal signals using a close-talk, head-worn condenser microphone and a contact microphone on the nose. Native speakers of Assamese with no history of any speech disorder were recorded reading three English passages containing phonetically balanced nasal and oral consonants (Rainbow passage), a second passage containing primarily nasal consonants (Nasal passage) and a third passage containing no nasal consonants at all (Zoo passage). Differences between the normalized intensity of signals obtained from the nasal channel and the oral channel (dn) were plotted. The plots indicated higher peaks at the location of the consonants. As expected, nasal consonants had higher peaks compared to non-nasal consonants. From these observations, an attempt was made to detect consonants in continuous speech. Apart from that, average dn values among the three types of passages were compared.

Text to speech synthesis system in Indian English

Grapheme to phoneme conversion modules are essential components in text-to-speech (TTS) systems. These modules operate before the phone sequence is fed into the synthesis routine. However, additional challenges emerge when such conversion modules are implemented with non-native varieties of languages, such as Indian English. As many of the the existing grapheme to phoneme dictionaries represent American English pronunciation, they are not suitable for use in Indian English TTS systems. Hence, in this work, an effort has been made to modify the existing English grapheme to phoneme dictionary by implementing specific rules for one particular variety of Indian English, namely Assamese English. The proposed method of dictionary modification is applied at the front end of the Indian English TTS, developed using unit selection synthesis and statistical parametric speech synthesis frameworks. In both frameworks, significant improvement is achieved in subjective evaluation when the dictionary is adapted to Assamese English pronunciation. The word error rate decreased from 46.67% to 7.69% after incorporating the variety specific modifications to the dictionary, indicating significant perceptual improvement.

Analysis of Glottal Stop in Assam Sora Language.

The objective of this work is to characterize the intervocalic glottal stops in Assam Sora. Assam Sora is a low resource language of the South Munda language family. Glottal stops are produced with gestures in the deep laryngeal level; hence, the estimated excitation source signal is used in this study to characterize the source dynamics during the production of Assam Sora glottal stops. From that, temporal domain voice source features, Quasi-Open Quotient (QOQ) and Normalized Amplitude Quotient (NAQ) are extracted along with spectral features such as H1-H2 ratio and Harmonic Richness Factor (HRF). One excitation source feature is extracted from the zero frequency filtered version of the speech signal to characterize the variations within the glottal cycles in glottal stop region. A recently proposed wavelet based voice source feature, Maxima Dispersion Quotient (MDQ) is also used to characterize the abrupt glottal closure during glottal stop production. From the analysis, it is observed that the features are salient enough to uniquely characterize glottal stops from the adjacent vowel sounds and may also be used in continuous speech. A Mann-Whitney U test confirmed the statistical significance of the differences between glottal stops and their adjacent vowels.