Razvoj glasovnega asistenta

CÖR, URBAN (author), Bajec, Marko (mentor), Lebar Bajec, Iztok (co-mentor)

Glasovni asistenti postajajo vse bolj priljubljeni in uporabni, vendar nobeden ne podpira slovenščine. Prav tako nismo odkrili pomočnikov, ki bi podpirali celoten cevovod glasovnega asistenta za slovenščino. V diplomskem delu smo razvili glasovnega asistenta za slovenski jezik za androidno napravo. Prepoznava ključne besede poteka lokalno na napravi, brez odvisnosti od interneta, s prilagojenim akustičnim modelom. Prepoznava govora poteka na zunanji storitvi z večjim akustičnim modelom. Za logiko glasovnega asistenta smo uporabili storitev Rasa, ki iz transkripcije uporabnikovega govora prepozna namen in kontekst. Odgovor te storitve se uporabi za sintezo odgovora, ki se nato predvaja uporabniku. Logiko na storitvi Rasa smo razširili, tako da ta storitev podpira tudi dva nova scenarija. Aplikacijo smo testirali v tihem okolju in v okolju s šumom za oba spola. Testirali pa smo tudi, delovanje aplikacije z uporabo druge storitve za transkripcijo govora. Aplikacija je najbolje delovala v tihem okolju za moškega govorca, saj v tem primeru ni prišlo do izpada iz toka delovanja aplikacije in smo čez celoten cevovod prišli v 100% testiranj. Primerljivo dobro pa je delovala tudi z uporabo druge storitve za transkripcijo govora, pri kateri smo čez celoten cevovod prišli v 94% testiranj.

• Language: Slovenian
• Keywords: glasovno upravljanje, virtualni pomočnik, glasovni pomočnik
• Work type: Bachelor thesis/paper
• Typology: 2.11 - Undergraduate Thesis
• Organization: FRI - Faculty of Computer and Information Science ; FMF - Faculty of Mathematics and Physics
• Year: 2022
• PID: 20.500.12556/RUL-135578
• COBISS.SI-ID: 102689283
• Publication date in RUL: 21.03.2022

Secondary language

• Language: English
• Title: Development of the voice assistant

Abstract

Voice assistants are becoming more and more popular and useful. However, none of the best-known voice assistants support the Slovene language. We also have not found any assistants that would support the entire pipeline of the voice assistant for Slovene language. We have developed a voice assistant for Slovene language and for an Android device. Keyword detection is done locally on the device, without needing the Internet, and with a modified acoustic model. For speech recognition, we used an outside service with a bigger language model. For the logic of the voice assistant, we used Rasa. Rasa can detect the intent and context of the transcribed speech. The response from this service is used for speech synthesis and is then played back to the user. We tested the application in two different environments, silent and noisy, and for the male and female voice. We also tested if the application functions properly if we use another service for speech recognition. The application worked best in the silent environment for a male speaker. In that case we finished the pipeline in 100% of tests. We achieved comparable results when we used another service for speech recognition. We finished the pipeline in 94% of tests.

• Keywords: voice control, virtual assistant, voice assistant