A Robin Hood approach to forced alignment: English-trained algorithms and their use on Australian languages

Abstract

Forced alignment automatically aligns audio recordings of spoken language with transcripts at the segment level, greatly reducing the time required to prepare data for phonetic analysis. However, existing algorithms are mostly trained on a few welldocumented languages. We test the performance of three algorithms against manually aligned data. For at least some tasks, unsupervised alignment (either based on English or trained from a small corpus) is sufficiently reliable for it to be used on legacy data for low-resource languages. Descriptive phonetic work on vowel inventories and prosody can be accurately captured by automatic alignment with minimal training data. Consonants provided significantly more challenges for forced alignment.

The Pacific Expansion: Optimizing Phonetic Transcription of Archival Corpora

“…Many software tools emerging from these innovations are based on speech and language models trained on huge corpora for major, wellresourced languages such as English. There is significant interest in adapting or tailoring forced alignment tools and techniques for use with data for low-resource languages, particularly material originating from archives and language documentation projects [13,14,15]. Phonetic annotation of such material allows greater inclusion of understudied languages in research on crosslinguistic phonetic patterns [16,17] and explorations of language-internal patterns of variation and change [18].…”

Computational sociophonetics using automatic speech recognition

“…An example of this would be using English ARPABET to transcribe words from Mixtec, and then treating the Mixtec words as new English words, which would be handled by a pre‐existing English acoustic model (DiCanio et al., 2013). This technique has been used to align data from languages like Swedish (Young & McGarrah, 2021); Tongan (Johnson et al., 2018), Chatino (Ćavar et al., 2016) and Triqui (Hatcher & DiCanio, 2019) from Mexico, Bribri (Coto‐Solano & Flores‐Solórzano, 2017), Malecu and Cabécar from Costa Rica (Coto‐Solano & Flores‐Solórzano, 2016); Nikyob from Nigeria (Kempton, 2017), Matukar Panau from Papua New Guinea (Barth et al., 2020); Yidiny from Australia (Babinsky et al., 2019); and North Australian Kriol (Jones et al., 2017).…”