1 |
Conversational telephone speech recognition for Lithuanian
|
|
|
|
In: ISSN: 0885-2308 ; EISSN: 1095-8363 ; Computer Speech and Language ; https://hal.archives-ouvertes.fr/hal-01837147 ; Computer Speech and Language, Elsevier, 2018, 49, pp.71-82 (2018)
|
|
BASE
|
|
Show details
|
|
2 |
Effective keyword search for low-resourced conversational speech
|
|
|
|
In: icassp 2017 ; https://hal.archives-ouvertes.fr/hal-01744176 ; icassp 2017, IEEE, Mar 2017, La Nouvelle Orléans, United States (2017)
|
|
BASE
|
|
Show details
|
|
3 |
An investigation into language model data augmentation for low-resourced STT and KWS
|
|
|
|
In: IEEE International Conference on Acoustics, Speech, and Signal Processing ; https://hal.archives-ouvertes.fr/hal-01837171 ; IEEE International Conference on Acoustics, Speech, and Signal Processing, IEEE, Mar 2017, New Orleans, United States (2017)
|
|
BASE
|
|
Show details
|
|
4 |
Language Model Data Augmentation for Keyword Spotting
|
|
|
|
In: Annual Conference of the International Speech Communication Association ; https://hal.archives-ouvertes.fr/hal-01837186 ; Annual Conference of the International Speech Communication Association , Jan 2016, San Francisco, United States (2016)
|
|
Abstract:
International audience ; This research extends our earlier work on using machinetranslation (MT) and word-based recurrent neural networks toaugment language model training data for keyword search inconversational Cantonese speech. MT-based data augmenta-tion is applied to two language pairs: English-Lithuanian andEnglish-Amharic. Using filtered N-best MT hypotheses for lan-guage modeling is found to perform better than just using the 1-best translation. Target language texts collected from the Weband filtered to select conversational-like data are used in severalmanners. In addition to using Web data for training the languagemodel of the speech recognizer, we further investigate using thisdata to improve the language model and phrase table of the MTsystem to get better translations of the English data. Finally,generating text data with a character-based recurrent neural net-work is investigated. This approach allows new word forms tobe produced, providing a way to reduce the out-of-vocabularyrate and thereby improve keyword spotting performance. Westudy how these different methods of language model data aug-mentation impact speech-to-text and keyword spotting perfor-mance for the Lithuanian and Amharic languages. The best re-sults are obtained by combining all of the explored methods.
|
|
Keyword:
[INFO.INFO-CL]Computer Science [cs]/Computation and Language [cs.CL]; [INFO]Computer Science [cs]; language modeling; low-resourced languages; machine translation; speech recognition; text augmentation
|
|
URL: https://hal.archives-ouvertes.fr/hal-01837186
|
|
BASE
|
|
Hide details
|
|
|
|