@inproceedings{yuan-etal-2022-data,
    title = "Data Augmentation for the Post-Stroke Speech Transcription ({PSST}) Challenge: Sometimes Less Is More",
    author = "Yuan, Jiahong  and
      Cai, Xingyu  and
      Church, Kenneth",
    editor = "Kokkinakis, Dimitrios  and
      Themistocleous, Charalambos K.  and
      Fors, Kristina Lundholm  and
      Tsanas, Athanasios  and
      Fraser, Kathleen C.",
    booktitle = "Proceedings of the RaPID Workshop - Resources and ProcessIng of linguistic, para-linguistic and extra-linguistic Data from people with various forms of cognitive/psychiatric/developmental impairments - within the 13th Language Resources and Evaluation Conference",
    month = jun,
    year = "2022",
    address = "Marseille, France",
    publisher = "European Language Resources Association",
    url = "https://preview.aclanthology.org/add-emnlp-2024-awards/2022.rapid-1.9/",
    pages = "71--79",
    abstract = "We employ the method of fine-tuning wav2vec2.0 for recognition of phonemes in aphasic speech. Our effort focuses on data augmentation, by supplementing data from both in-domain and out-of-domain datasets for training. We found that although a modest amount of out-of-domain data may be helpful, the performance of the model degrades significantly when the amount of out-of-domain data is much larger than in-domain data. Our hypothesis is that fine-tuning wav2vec2.0 with a CTC loss not only learns bottom-up acoustic properties but also top-down constraints. Therefore, out-of-domain data augmentation is likely to degrade performance if there is a language model mismatch between {\textquotedblleft}in{\textquotedblright} and {\textquotedblleft}out{\textquotedblright} domains. For in-domain audio without ground truth labels, we found that it is beneficial to exclude samples with less confident pseudo labels. Our final model achieves 16.7{\%} PER (phoneme error rate) on the validation set, without using a language model for decoding. The result represents a relative error reduction of 14{\%} over the baseline model trained without data augmentation. Finally, we found that {\textquotedblleft}canonicalized{\textquotedblright} phonemes are much easier to recognize than manually transcribed phonemes."
}