This project was developed just for fun in early 2019. I wanted to create an end-to-end machine learning pipeline and try something new as I was learning machine learning and deep learning.
This system takes your speech as input and converts the spoken sentence -> English text -> Hindi Text -> Hindi Speech. The final output will be a similar interface to that we get of Google Translate.
You can watch the demo in following links
- https://drive.google.com/file/d/1HJ0B0-8iz_8j6HwipfQ5kpbcBdG-ZFWK/view?usp=sharing
- https://drive.google.com/file/d/1H9jYESDGzuKvyxcT-0w8gSYul533VsRb/view?usp=sharing
I have only done the translation for a few short phrases. So do not expect it to work on all phrases.
Python libraries or Tools used :
- ffmpeg - to change audio sampling rate and encoding
- Mozilla Deepspeech - offline speech-to-text engine
- Tensorflow-nmt - English to Hindi Translation
You will need Ubuntu 16.04 or higher and Python 3.6 or higher
- Clone the repo
- Run
$ sh run_first.sh. This will install all pre-requisites - Run
$ python app.pyThis will start a flask server on localhost