This is a neural machine translation project with goal of translating Chinese audio to English text for content moderation. The complete portion of the project uses paired bilingual sentences as data (cmn.txt) and a seq2seq model with and LSTM encoding layer and an Attention decoding layer to translate Chinese to English text. The highest accuracy achieved so far is 69.82% using the BLEU scale for unigram strings.
The most recent notebook is AudioModeratorV3clean_Jul7.ipynb. Slides and a report are also available at Capstone2_June30_slides.pptx and Capstone2_MilestoneReport_jun30.pdf, respectively.