Ahots sintetiko pertsonalizatuak: esperientzia baten deskribapena

Abstract

Ahotsa ezinbestekoa da giza komunikaziorako, eta haren galerak eragin handia du pertsonak gizartean integratzeko prozesuan. Testu-ahots bihurketak ahots sintetikoa eman diezaieke ahozko desgaitasuna duten pertsonei. Irtenbide arruntenek ahots estandarra izaten dute normalean, eta, horregatik, erabiltzaile batzuek zailtasunak dituzte beren burua ahots horrekin identifikatzeko. Horregatik, ahots sintetiko pertsonalizatuak sortu behar dira, eta ahozko desgaitasuna duten pertsonei ahots-katalogo bat eskaini behar zaie, beren beharretara egokitzen den ahots bat aukeratu ahal izan dezaten. ZureTTS proiektuaren helburua ahots pertsonalizatu horiek ematea da, bai gaztelaniaz, bai euskaraz. Ahotsa galduko duten pertsonek edo ahotsik ez dutenei ahotsa eman nahi dieten pertsona altruistek 100 esaldi grabatzen dituzte, AhoMyTTS web-atariaren bidez. Esaldi horiekin, egokitze-prozesu bat egiten da, grabaketako ahotsaren antzeko ahots sintetiko bat sortzeko. Erabiltzaileari sintesi-motor bat ematen zaio ahots pertsonalizatu horrekin batera, ahozko mezuak sortzea eskaintzen duten aplikazioetan erabiltzeko. Gainera, ahots-katalogo bat ere badago, grabaketarik egin ezin duen pertsona batek ahots horien artean gustukoena aukeratu dezan. 1.200 pertsonak baino gehiagok erabili dute sistema hori ahots pertsonalizatu bat lortzeko, eta haietatik 58 hautatu ditugu katalogoan sartzeko. Erabiltzaileei egindako inkestek erakusten dute gustura daudela ahots sintetikoaren hainbat alderdirekin: gehienen ustez, ahots sintetikoa jatorrizkoaren antzekoa da, atsegina eta argia, baina robotiko samarra. Lan honek garapen jasangarrirako 10. helburuari laguntzen dio, herrialde bakoitzaren barneko eta herrialdeen arteko desberdintasunak murriztuz. Era berean, garapen jasangarrirako 4. helburuari ere laguntzen dio, guztiontzako kalitatezko hezkuntza inklusiboa nahiz bidezkoa bermatzea errazten duten tresnak eskainiz.; The voice is so essential for human communication that its loss drastically affects the integration of people in society. Text-to-speech can provide a synthetic voice for people with oral disabilities. The most common solutions usually provide a standard voice, and users have difficulties to identify themselves with it. For this reason, we need to create personalized synthetic voices and offer a catalogue of voices to people with oral disabilities so that they can choose one that suits their needs. The objective of the ZureTTS project is to provide these personalized voices, both in Spanish and in Basque. Through the AhoMyTTS web portal, people who are going to lose their voice or altruistic people who want to provide voices to those who do not have it, record 100 carefully se-lected sentences. A synthetic voice with similar characteristics to the voice of the recording is generated by applying an adaptation process. The user is provided with a synthesis engine along with that personalized voice, so that they can use it in applications that require oral message generation. In addition, we offer a catalogue of voices to choose from if one is no longer able to record. More than 1,200 people have used the system to obtain a personalized voice and 58 of them have been selected to be included in the cata-logue. User surveys show user satisfaction with various aspects of the synthetic voice: most think that the synthetic voice is similar to the original, pleasant and clear, although a bit robotic. This work contributes mainly to goal 10 for sustainable development by re-ducing inequality within and among countries. It also contributes to goal 4 for sustainable development, providing tools that facilitate access for all to an inclusive, equitable and quality education.