We studied Japanese large-vocabulary continuous-speech recognition (LV CSR) for a Japanese business newspaper. To enable word N-grams to be used, sentences were first segmented into words (morphemes) using a morphological analyzer. Newspaper articles for about five years were used to train N-gram language models. To evaluate our recognition system, we recorded speech data for sentences from another set of articles. Using the speech corpus, LV CSR experiments were conducted. For 7k vocabulary, the word error rate was 82.8% when no grammar and context-independent acoustic models were used. This improved to 20.0% when both bigram language models and context-dependent acoustic models were used.
@inproceedings{matsuoka96_icslp, title = {Japanese large-vocabulary continuous-speech recognition using a business-newspaper corpus}, author = {Tatsuo Matsuoka and Katsutoshi Ohtsuki and Takeshi Mori and Sadaoki Furui and Katsuhiko Shirai}, year = {1996}, booktitle = {4th International Conference on Spoken Language Processing (ICSLP 1996)}, pages = {22--25}, doi = {10.21437/ICSLP.1996-6}, issn = {2958-1796},}Cite as:Matsuoka, T., Ohtsuki, K., Mori, T., Furui, S., Shirai, K. (1996) Japanese large-vocabulary continuous-speech recognition using a business-newspaper corpus. Proc. 4th International Conference on Spoken Language Processing (ICSLP 1996), 22-25, doi: 10.21437/ICSLP.1996-6