ASJ (Acoustical Society of Japan) 2015 Autumn Meeting
開催地
和文:
福島県会津若松市
英文:
Aizu-wakamatsu city, Fukushima Pref.
ファイル
アブストラクト
This paper presents our work to develop a DNN-based ASR system for the Indonesian language, a language that has limited training data. We evaluate the effectiveness of some common improvements to DNN acoustic models, such as speaker adaptive features and sequence-discriminative training. We also explore incorporating English in the training process as a potential way of reducing WER when the Indonesian training data is very limited.
各種検索
サポート
ヘルプ