“Speech banana” is a banana-shaped plot of speech power distribution, where the abscissa and ordinate represent frequency and intensity. By superimposing speech banana over an audiogram, tested with pure tones, degrees of gain or loss of individual speech sound could be predicted. Speech banana has been constructed for English (Northern and Downs, 1984) and Swedish (Liden and Fant, 1954); however, none has been proposed for tonal languages, such as Thai. This work presents a construction of speech banana for Thai, a language with 21 consonants and 5 lexical tones. Specifically, intensity of each phoneme in the speech banana was calculated by differences of sound pressure level between the local maxima of power spectral density and equal loudness contour at 0 dB. Distribution of the 21 consonants is around 170-5700 Hz and 25-65 dB. Predictions of gain or loss of the phonemes from the constructed speech banana and audiograms were evaluated based on perception test results from seven Thai sensori-neural hearing loss patients, where they identified what they heard from a pair of rhyming words (210 stimuli) differing in initial phonemes, equally distributed across phonemes. Interestingly, the results showed high prediction rates of 71.4-85.7% for phonemes predominantly emphasized on frequency below 2000 Hz.
Sexual violence is a severe and chronic occurrence around the world that has not been resolved. The stigmatized nature of sexual violence has forced victims and survivors to accept prejudiced accusations cultivated from discriminatory norms when they are never at fault nor responsible for such violations against their sexuality. LAW-U is an Artificial Intelligence (AI) chatbot that gives legal guidance to survivors of sexual violence by recommending the most relevant Supreme Court decisions to the survivors' situations. In Thai, "LAW-U" − pronounced similarly to "รออยู ่ " − means "I will wait for you", which signifies the chatbot's unconditional support to the user. 182 Thai Supreme Court cases of sexual violence, relating to Sections 276, 277, 278, and 279 of the Criminal Code, were used to develop Natural Language Processing (NLP) pipelines for LAW-U. Legal experts then generated mock-up dialogs from Supreme Court decisions which became the conversations used to train LAW-U. The computation of the similarity scores and the calculation of combined percentages of common keywords and keywords' synonyms were completed to increase the model's accuracy. When applying the model to the hold-out testing dataset, the accuracy was 88.89% for an exact match between the user's input and the Supreme Court case − this confirmed that LAW-U was ready for real-life application. LAW-U's unique design hopes to act as a precedent for other works at home and abroad to perpetuate awareness of sexual violence and eliminate any tolerance against these crimes by empowering sexual violence victims and survivors to reaffirm their inherent rights.
scite is a Brooklyn-based organization that helps researchers better discover and understand research articles through Smart Citations–citations that display the context of the citation and describe whether the article provides supporting or contrasting evidence. scite is used by students and researchers from around the world and is funded in part by the National Science Foundation and the National Institute on Drug Abuse of the National Institutes of Health.