| Nome: | Descrição: | Tamanho: | Formato: | |
|---|---|---|---|---|
| 2.29 MB | Adobe PDF |
Autores
Orientador(es)
Resumo(s)
A pronĂșncia Ă© um dos componentes mais crĂticos da competĂȘncia oral, podendo condicionar a inteligibilidade da fala mesmo quando o vocabulĂĄrio e a gramĂĄtica sĂŁo adequados. No caso do PortuguĂȘs Europeu, a oferta de ferramentas especĂficas para treino de pronĂșncia com feedback automatizado e registo sistemĂĄtico do desempenho Ă© ainda limitada. Este relatĂłrio descreve o desenvolvimento e a avaliação de um protĂłtipo de serious game para treino de pronĂșncia em PortuguĂȘs Europeu, concebido para funcionar em modo offline e apoiar tanto a prĂĄtica autĂłnoma como a integração em contextos educativos.
O trabalho segue uma metodologia de natureza aplicada, combinando engenharia de software e prototipagem iterativa. O protĂłtipo integra uma interface grĂĄfica desenvolvida em Python com Pygame, um mĂłdulo de reconhecimento automĂĄtico de fala baseado no modelo Whisper, um conjunto de frases organizadas em trĂȘs nĂveis de dificuldade e um sistema de registo de dados suportado em ficheiros JSON e na geração de relatĂłrios PDF. Em cada tentativa, o utilizador ouve uma frase em PortuguĂȘs Europeu, repete-a e obtĂ©m uma transcrição automĂĄtica, uma percentagem de acerto calculada por palavra e a identificação de determinados padrĂ”es de erro fonĂ©tico.
A avaliação do protĂłtipo, realizada em contexto de prova de conceito, mostra que o sistema funciona de forma estĂĄvel, gera mĂ©tricas de desempenho interpretĂĄveis e produz relatĂłrios estruturados que permitem acompanhar a evolução ao longo do tempo. SĂŁo igualmente discutidas limitaçÔes importantes, relacionadas com a escala e diversidade reduzidas dos dados, a dependĂȘncia de um modelo ASR genĂ©rico e a cobertura parcial dos fenĂłmenos fonĂ©ticos, apontando-se linhas de trabalho futuro para aprofundar a anĂĄlise fonĂ©tica, especializar o reconhecimento de fala e alargar o estudo a populaçÔes de utilizadores mais variadas.
Pronunciation is a critical component of oral proficiency and can strongly affect speech intelligibility, even when vocabulary and grammar are appropriate. In the case of European Portuguese, there is still a limited number of dedicated tools for pronunciation training that provide automatic feedback and systematic tracking of learner performance. This report presents the development and evaluation of a prototype serious game for pronunciation training in European Portuguese, designed to run offline and to support both individual practice and use in educational contexts. The work follows an applied research methodology, combining software engineering with iterative prototyping. The prototype integrates a graphical interface developed in Python with Pygame, an automatic speech recognition module based on the Whisper model, a set of sentences organised into three difficulty levels, and a data logging subsystem based on JSON files and PDF report generation. In each attempt, the user listens to a reference sentence in European Portuguese, repeats it, and receives an automatic transcription, a word-level accuracy score and the identification of specific phonetic error patterns. The proof-of-concept evaluation, based on a small exploratory dataset, indicates that the system operates stably, produces interpretable performance metrics and generates structured reports that enable monitoring of progress over time. Important limitations are also discussed, namely the limited scale and diversity of the data, the dependence on a generic ASR model and the partial coverage of phonetic phenomena. Future work directions include refining the phonetic analysis, specialising the speech recognition component and extending the empirical study to more diverse groups of users.
Pronunciation is a critical component of oral proficiency and can strongly affect speech intelligibility, even when vocabulary and grammar are appropriate. In the case of European Portuguese, there is still a limited number of dedicated tools for pronunciation training that provide automatic feedback and systematic tracking of learner performance. This report presents the development and evaluation of a prototype serious game for pronunciation training in European Portuguese, designed to run offline and to support both individual practice and use in educational contexts. The work follows an applied research methodology, combining software engineering with iterative prototyping. The prototype integrates a graphical interface developed in Python with Pygame, an automatic speech recognition module based on the Whisper model, a set of sentences organised into three difficulty levels, and a data logging subsystem based on JSON files and PDF report generation. In each attempt, the user listens to a reference sentence in European Portuguese, repeats it, and receives an automatic transcription, a word-level accuracy score and the identification of specific phonetic error patterns. The proof-of-concept evaluation, based on a small exploratory dataset, indicates that the system operates stably, produces interpretable performance metrics and generates structured reports that enable monitoring of progress over time. Important limitations are also discussed, namely the limited scale and diversity of the data, the dependence on a generic ASR model and the partial coverage of phonetic phenomena. Future work directions include refining the phonetic analysis, specialising the speech recognition component and extending the empirical study to more diverse groups of users.
Descrição
Palavras-chave
serious games treino de pronĂșncia PortuguĂȘs Europeu reconhecimento automĂĄtico de fala Whisper Computer-Assisted Pronunciation Training. pronunciation training European Portuguese automatic speech recognition
Contexto Educativo
Citação
Editora
Licença CC
Sem licença CC
