Detection of Sentence Modality on French Automatic Speech-to-text Transcriptions

Abstract This article analyzes the detection of sentence modality in French when it is applied on automatic speech-to-text transcriptions. Two sentence modalities are evaluated (questions and statements) using prosodic and linguistic information. The linguistic features consider the presence of discriminative interrogative patterns and two log-likelihood ratios of the sentence being a question rather than a statement: one based on words and the other one based on part-of-speech tags. The prosodic features are based on duration, energy and pitch features estimated over the last prosodic group of the sentence. The classifiers based on linguistic features outperform the classifiers based on prosodic features. The combination of linguistic and prosodic features gives a slight improvement on automatic speech transcriptions, where the correct classification performance reaches 72%. A detailed analysis shows that small errors in the determination of the segment boundaries are not critical.