BriefGPT.xyz
May, 2022
使用预训练模型的自适应多语言语音识别
Adaptive multilingual speech recognition with pretrained models
HTML
PDF
Ngoc-Quan Pham, Alex Waibel, Jan Niehues
TL;DR
使用预训练的wav2vec 2.0和MBART50模型，结合自适应权重技术，显著提高公共数据集上多语言语音识别的准确性，比纯监督学习提高44%。同时我们还探索了如何通过微调结构以获得最佳模型。
Abstract
multilingual speech recognition
with
supervised learning
has achieved great results as reflected in recent research. With the development of pretraining methods on
→