面向伊卡语的自动语音识别
Automatic Speech Recognition for the Ika Language
AI总结：
本文利用《新约》伊卡语语音数据微调 wav2vec 2.0 多语言模型，仅用1小时训练数据即实现低资源 ASR，并验证大模型优势及过拟合问题。
摘要：
我们提出了一种高成本效益的方法，用于为伊卡语等低资源语言开发自动语音识别（ASR）模型。我们在一个高质量语音数据集上对预训练的 wav2vec 2.0 大规模多语言语音模型进行微调，该数据集由伊卡语的《新约》圣经译本汇编而成。结果表明，仅使用略多于1小时的训练数据，微调多语言预训练模型即取得0.5377的词错误率（WER）和0.2651的字符错误率（CER）。参数量为10亿的较大模型优于参数量为3亿的较小模型，原因在于其更高的复杂度以及存储更丰富语音表示的能力。然而，我们观察到模型对小型训练数据集存在过拟合，降低了泛化能力。研究结果证明了利用多语言预训练模型服务低资源语言的潜力。未来工作应聚焦于扩充数据集，并探索缓解过拟合的技术。
英文摘要：
We present a cost-effective approach for developing Automatic Speech Recognition (ASR) models for low-resource languages like Ika. We fine-tune the pretrained wav2vec 2.0 Massively Multilingual Speech Models on a high-quality speech dataset compiled from New Testament Bible translations in Ika. Our results show that fine-tuning multilingual pretrained models achieves a Word Error Rate (WER) of 0.5377 and Character Error Rate (CER) of 0.2651 with just over 1 hour of training data. The larger 1 billion parameter model outperforms the smaller 300 million parameter model due to its greater complexity and ability to store richer speech representations. However, we observe overfitting to the small training dataset, reducing generalizability. Our findings demonstrate the potential of leveraging multilingual pretrained models for low-resource languages. Future work should focus on expanding the dataset and exploring techniques to mitigate overfitting.
机构、总结、中文摘要均由AI分析而来，可能存在错误，请以论文正文为准。
