README.md

Contents

STT Error Data • Twitter Sentiment Data • Joint Comp/Inc

Dependencies

Python 3.7.2, requests, numpy, nltk

1. STT Error Dataset

• Original dataset: Chatbot Corpus

• Make STT Error Dataset: Text-to-Speech -> audio -> Speech-to-Text

cd make_stterror_data
python main.py --data_dir data/intent_chatbot/

• Output:

  • TTS audios, STT recovered texts, BLEU scores
  • The stterror_data/chatbot/ directory was organized in such way to separate train and test in each TTS-STT combination

• Examples of sentences with STT error

  Corpus	TTS	iBLEU	Original	With STT error
  Chatbot	gtts	0.4376	"how can i get from garching to milbertshofen?"	"how can i get from garching to melbourne open."
  Chatbot	macsay	0.5042	"how can i get from garching to milbertshofen?"	"how can i get from garching to meal prep."

  STT: Wit.ai

  iBLEU = 1 - BLEU

2. Twitter Sentiment Dataset

• Original dataset: Kaggle's Sentiment140

• Tweets have natural human error (noise)

  

• Correct sentences are obtained with Amazon MTurk

• Examples of sentences with Natural Human error

  Original	With Error
  "goonite sweet dreamz"	"Good night, sweet dreams."
  "well i dunno..i didnt give him an ans yet"	"Well I don't know, I didn't give him an answer yet."
  "u kno who am i talkin bout??"	"Do you know who I am talking about?"

3. Joint Complete/Incomplete Data

• In order for the model to be robust to missing data it also needs to be trained on sentences with missing words.
• After making the incomplete dataset, there are two options
  1. Make dataset with Complete and Incomplete Data

     python make_joint_comp_inc_data.py
     

  2. Make dataset with Incomplete Data (add target sentence to tsv file to train autoencoder)

     python add_target_to_inc_data.py