School of Languages and Linguistics - Research Publications

Permanent URI for this collection

Search Results

Now showing 1 - 3 of 3
  • Item
    Thumbnail Image
    Building Speech Recognition Systems for Language Documentation: The CoEDL Endangered Language Pipeline and Inference System (ELPIS)
    Foley, B ; Arnold, J ; Coto-Solano, R ; Durantin, G ; Ellison, TM ; van Esch, D ; Heath, S ; Kratochvíl, F ; Maxwell-Smith, Z ; Nash, D ; Olsson, O ; Richards, M ; San, N ; Stoakes, H ; Thieberger, N ; Wiles, J (ISCA, 2018)
    Machine learning has revolutionized speech technologies for major world languages, but these technologies have generally not been available for the roughly 4,000 languages with populations of fewer than 10,000 speakers. This paper describes the development of ELPIS, a pipeline which language documentation workers with minimal computational experience can use to build their own speech recognition models, resulting in models being built for 16 languages from the Asia-Pacific region. ELPIS puts machine learning speech technologies within reach of people working with languages with scarce data, in a scalable way. This is impactful since it enables language communities to cross the digital divide, and speeds up language documentation. Complete automation of the process is not feasible for languages with small quantities of data and potentially large vocabularies. Hence our goal is not full automation, but rather to make a practical and effective workflow that integrates machine learning technologies.
  • Item
    Thumbnail Image
    The Pacific Expansion: Optimizing phonetic transcription of archival corpora
    Billington, R ; Stoakes, H ; Thieberger, N (ISCA-INT SPEECH COMMUNICATION ASSOC, 2021)
    For most of the world’s languages, detailed phonetic analyses across different aspects of the sound system do not exist, due in part to limitations in available speech data and tools for efficiently processing such data for low-resource languages. Archival language documentation collections offer opportunities to extend the scope and scale of phonetic research on low-resource languages, and developments in methods for automatic recognition and alignment of speech facilitate the preparation of phonetic corpora based on these collections. We present a case study applying speech modelling and forced alignment methods to narrative data for Nafsan, an Oceanic language of central Vanuatu. We examine the accuracy of the forced-aligned phonetic labelling based on limited speech data used in the modelling process, and compare acoustic and durational measures of 17,851 vowel tokens for 11 speakers with previous experimental phonetic data for Nafsan. Results point to the suitability of archival data for large-scale studies of phonetic variation in low-resource languages, and also suggest that this approach can feasibly be used as a starting point in expanding to phonetic comparisons across closely-related Oceanic languages.
  • Item
    Thumbnail Image
    Building Speech Recognition Systems for Language Documentation: The CoEDL Endangered Language Pipeline and Inference System (ELPIS)
    Foley, B ; Arnold, J ; Coto-Solano, R ; Durantin, G ; Mark, E ; van Esch, D ; Heath, S ; Kratochvíl, F ; Maxwell-Smith, Z ; Nash, D ; Olsson, O ; Richards, M ; San, N ; Stoakes, H ; Thieberger, N ; Wiles, J (International Speech Communication Association, 2018-08-30)
    Machine learning has revolutionised speech technologies for major world languages, but these technologies have generally not been available for the roughly 4,000 languages with populations of fewer than 10,000 speakers. This paper describes the development of Elpis, a pipeline which language documentation workers with minimal computational experience can use to build their own speech recognition models, resulting in models being built for 16 languages from the Asia-Pacific region. Elpis puts machine learning speech technologies within reach of people working with languages with scarce data, in a scalable way. This is impactful since it enables language communities to cross the digital divide, and speeds up language documentation. Complete automation of the process is not feasible for languages with small quantities of data and potentially large vocabularies. Hence our goal is not full automation, but rather to make a practical and effective workflow that integrates machine learning technologies.