End-to-End Automatic Speech Recognition for the Archive of Ainu Folklore
| Project/Area Number |
18K19814
|
|---|
| Research Category |
Grant-in-Aid for Challenging Research (Exploratory)
|
| Allocation Type | Multi-year Fund |
|---|
| Review Section |
Medium-sized Section 61:Human informatics and related fields
|
|---|
| Research Institution | Kyoto University |
|---|
Principal Investigator |
|
|---|
| Co-Investigator(Kenkyū-buntansha) |
奥田 統己 札幌学院大学, 人文学部, 教授 (60224151)
|
|---|
| Project Period (FY) |
2018-06-29 – 2021-03-31
|
| Project Status |
Completed (Fiscal Year 2020)
|
| Budget Amount *help |
¥6,240,000 (Direct Cost: ¥4,800,000、Indirect Cost: ¥1,440,000)
Fiscal Year 2019: ¥3,120,000 (Direct Cost: ¥2,400,000、Indirect Cost: ¥720,000)
Fiscal Year 2018: ¥3,120,000 (Direct Cost: ¥2,400,000、Indirect Cost: ¥720,000)
|
|---|
| Keywords | アイヌ語 / 音声認識 / 音声合成 / 消滅危機言語 / End-to-Endモデル |
|---|
| Outline of Final Research Achievements |
We have investigated automatic speech recognition (ASR) of Ainu folklores (Uwepeker). First, we constructed an Ainu speech corpus for the Saru dialect based on data provided by two museums which have constructed Ainu archive. Next, we built an ASR system based on the end-to-end model, and compared four recognition units of phones, syllables, word pieces, and words. With the syllable unit, we achieved phone recognition accuracy of 93.7% and word recognition accuracy of 78.3%. To address the problem of significant degradation in the speaker-open condition, an unsupervised speaker adaptation method using CycleGAN is proposed. Finally, we also implemented language identification in Japanese and Ainu mixed speech by combining phone and word recognition modules.
|
| Academic Significance and Societal Importance of the Research Achievements |
アイヌ文化の多くが口頭で継承されてきましたが、アイヌ語は2009年にUNESCOにより「極めて深刻な」消滅危機言語に認定される事態となっています。以前から口頭伝承を録音・記録する活動が様々に行われてきたが、その書き起こし・アーカイブ化には膨大な手間とアイヌ語の知識を必要とするため、多くが未整備のままでした。
本研究により、アイヌ語のアーカイブ構築の効率化への寄与が期待されます。実際に、アイヌ民族博物館において音声と書き起こし同期のための対応付けに活用され、1時間のデータに対して、人手で1日要する作業がほぼ完全に自動化できました。
|
Report
(4 results)
Research Products
(13 results)
