|
|
--- |
|
|
license: mit |
|
|
language: |
|
|
- zh |
|
|
pipeline_tag: image-to-text |
|
|
--- |
|
|
|
|
|
# Target: Convert Scanned Images of IPA symbols to Pinyin |
|
|
Scanned images of IPA phonetic symbols for Chengdunese (成都话) in The Great Dictionary of Modern Chinese Dialects (現代漢語方言大詞典). |
|
|
|
|
|
# Training and Test Set |
|
|
* 2,553 images of IPA phonetic symbols generated from Pinyin pronunciations found in Sichuanese Dialect Dictionary (四川方言词典 教你一口地道的四川话) and the word list of the Shupin (蜀拼) input method. |
|
|
* 80/20 split on train/test |
|
|
|
|
|
# Results |
|
|
* Trained for 180 steps with a batch size of 32 |
|
|
* Final Character Error Rate of 0.795% on test set |
|
|
* TODO: label part of the scanned images to see if model generalizes on target task |
|
|
|
