Faniriniaina/Audio-data-collection

This data collection project was part of the Speech Recognition course at AMMI Rwanda 2019-2020. This data contains the audio recording of about 04h20mn in Malagasy language, the official language of Madagascar. It is divided into 23 parts (sessions) which contains each around 50 up to 200 sentences. Two native Malagasy contributed to achieve this collection task and the read text were provided from Wikipedia and the Catechism of the Catholic church in Madagascar Katesizy. The devices used were two Android phones (HUAWEI-DUB-LX2, TCL-5058I). The sensitivity of the microphones of the devices reduced the quality of the audio recordings. Another problem encountered was that it was not possible to immediately listen to the recording after each sentence. We have to finish the session before any possible listening. It was hard to detect wrong intonation when reading the data for this reason. This data collection was made possible thanks to the application Lig-Aikuma, a recording application built by PhD students in France (version 3.1.0). This application automatically creates aligned text and audio data which will be used for training. NOTE: This data cannot be used in any circumstance except for the purpose of this project AMMI-SR 2019-2020 or upon request.