Rodolfo Delmonte, Poetry and computer speech with SPARSAR |
|
Seminar delivered in English by Rodolfo Delmonte (Ca' Foscari University of Venice) on April 1, 2020 held ON LINE within the VeDPH seminar series Seminars in Digital and Public Humanities 2019/20.
Abstract: In this paper we present SPARSAR, a system for English poetry recital. The system is parasitic on TextToSpeech (TTS) systems available both online and on Macintosh computers. It creates prosodic parameters and phonetic transcriptions on any input text to be used by the TTS in order to normalize and improve current systems which are statistically based. In order to show TTS inability to produce semantically coherent and expressive readings Italian texts will be used at first and critical points indicated and discussed. Then SPARSAR architecture will be introduced and its three layers presented in detail. The ability of the system to generate appropriate prosodic parameters will be discussed in relation to a poem by Sylvia Plath, Edge. The peculiarity of this poem is its richness in enjambments, which are not captured at all by statistically based TTS. Eventually, latest work on Elisabethan poetry will be presented and a Sonnet by Shakespeare will be transcribed and annotated with prosodic parameters’ values by the system; in particular, it will be shown how the lack of a specific component to account for contractions and rhyming violations makes best commercial TTS systems even unable to pronounce words correctly. #VeDPH GitHub repo (with Del Monte code): https://github.com/vedph/ and for slides: https://github.com/vedph/event_materials |