One of the greatest issues facing documentary linguists is the transcription bottleneck. While the large quantity of audio and video data gener- ated as part of a documentary project serves as a long-lasting record of the language, without corresponding text transcriptions, it remains largely inaccessible for revitalization efforts and linguistic analysis. Automated Speech Recognition (ASR) is frequently proposed as the solution to this problem. However, two is- sues often prevent documentary linguists from making use of ASR models 1) the thought that the typical documentary project does not have sufficient data to develop an adequate ASR model and 2) that correcting the output of an ASR model would be more time-consuming for transcribers than simply creating a transcription from scratch. In this paper, we tackle both of these issues by developing an ASR model in the larger context of a documentation project for Nasal, a low-resource language of western Indonesia. Fine-tuning a larger pre-trained lan- guage model on 25 hours of transcribed Nasal speech, we produce a model that has a 44% word error rate. Despite this relatively high error rate, tests comparing speed of transcrib- ing from scratch and correcting ASR-generated transcripts show that the ASR model can sig- nificantly speed up the transcription process.

The performance of automatic speech recognition (ASR) systems has advanced substantially in recent years, particularly for languages for which a large amount of transcribed speech is available. Unfortunately, for low-resource languages, such as minority languages, regional languages or dialects, ASR performance generally remains much lower. In this study, we investigate whether data augmentation techniques could help improve low-resource ASR performance, focusing on four typologically diverse minority languages or language variants (West Germanic: Gronings, West-Frisian; Malayo-Polynesian: Besemah, Nasal). For all four languages, we examine the use of self-training, where an ASR system trained with the available human-transcribed data is used to generate transcriptions, which are then combined with the original data to train a new ASR system. For Gronings, for which there was a pre-existing text-to-speech (TTS) system available, we also examined the use of TTS to generate ASR training data from text-only sources. We find that using a self-training approach consistently yields improved performance (a relative WER reduction up to 20.5% compared to using an ASR system trained on 24 minutes of manually transcribed speech). The performance gain from TTS augmentation for Gronings was even stronger (up to 25.5% relative reduction in WER compared to a system based on 24 minutes of manually transcribed speech). In sum, our results show the benefit of using self-training or (if possible) TTS-generated data as an efficient solution to overcome the limitations of data availability for resource-scarce languages in order to improve ASR performance.