Advance research is going on in the field of text analytics and NLP. Deep learning approach like Seq2Seq models and BERT have been established for tasks like language translation, abstractive text summarization and image captioning etc.
To determine how models are performing, we need performance metrics like rouge score, WER, blue score. WER or word error rate provides us a good vision by providing in depth analysis like words substituted, inserted or deleted.
The original text is the reference text or the gold standard text, machine, or model generated text is the hypothesis test.
WER is formalized as (number of words present in machine-generated text but not in original text + number of words present in the original text but not in machine-generated text + number of words substituted) / (number of words in original text)
MER is formalized as (number of words present in machine-generated but not in original text + number of words present in the original text but not in machine-generated text) / (number of the word in machine-generated text)
HTML template :
Python FLASK Code:
How to run the code:
Make sure you have anaconda or python installed in your system with libraries flask, jiwer, numpy and re. Open anaconda prompt or terminal, navigate to WER folder and execute ‘python wer.py’. By default it will run on http://localhost:8018/ but you can change the port in wer.py file.