Badenhorst, JDe Waal, ADe Wet, Febe2012-06-292012-06-292012-05Badenhorst, J, De Waal, A and De Wet, F. Quality measurements for mobile data collection in the developing world. Third International Workshop on Spoken Languages Technologies for Under-resourced Languages (SLTU'12), Monkey Valley, Cape Town, South Africa, 7-9 May 2012http://hdl.handle.net/10204/5954Third International Workshop on Spoken Languages Technologies for Under-resourced Languages (SLTU'12), Monkey Valley, Cape Town, South Africa, 7-9 May 2012The collection of speech data suitable for speech technology development is a challenge for under-resourced languages. Factors such as cost, availability of mother-tongue speakers and vast geographic distances call for techniques to optimise the data collection process in order to reduce re-collection of data. The use of mobile devices facilitate remote speech data collection. Although mobile (and remote) data collection addresses the challenging factors mentioned above, the environment is still less controlled than in the case of laboratory or studio-based recordings. In this paper we firstly revisit semi-realtime, basic quality control checks as implemented on available mobile-based speech data collection software (Woefzela). In addition, we introduce a quality control technique that uses speech duration estimation to validate the acoustic quality of the speech samples. We compare both techniques with manual verifications.enSpeech data collectionResource-scarce environmentUnder-resourced languagesAutomatic speech recognitionMobile data collectionAndroidWoefzelaSpeech technology developmentConference PresentationBadenhorst, J., De Waal, A., & De Wet, F. (2012). Quality measurements for mobile data collection in the developing world. http://hdl.handle.net/10204/5954Badenhorst, J, A De Waal, and Febe De Wet. "Quality measurements for mobile data collection in the developing world." (2012): http://hdl.handle.net/10204/5954Badenhorst J, De Waal A, De Wet F, Quality measurements for mobile data collection in the developing world; 2012. http://hdl.handle.net/10204/5954 .TY - Conference Presentation AU - Badenhorst, J AU - De Waal, A AU - De Wet, Febe AB - The collection of speech data suitable for speech technology development is a challenge for under-resourced languages. Factors such as cost, availability of mother-tongue speakers and vast geographic distances call for techniques to optimise the data collection process in order to reduce re-collection of data. The use of mobile devices facilitate remote speech data collection. Although mobile (and remote) data collection addresses the challenging factors mentioned above, the environment is still less controlled than in the case of laboratory or studio-based recordings. In this paper we firstly revisit semi-realtime, basic quality control checks as implemented on available mobile-based speech data collection software (Woefzela). In addition, we introduce a quality control technique that uses speech duration estimation to validate the acoustic quality of the speech samples. We compare both techniques with manual verifications. DA - 2012-05 DB - ResearchSpace DP - CSIR KW - Speech data collection KW - Resource-scarce environment KW - Under-resourced languages KW - Automatic speech recognition KW - Mobile data collection KW - Android KW - Woefzela KW - Speech technology development LK - https://researchspace.csir.co.za PY - 2012 T1 - Quality measurements for mobile data collection in the developing world TI - Quality measurements for mobile data collection in the developing world UR - http://hdl.handle.net/10204/5954 ER -