IITD-speech-vone

Github repository for the open source library "IITD-speech-vone", https://pypi.org/project/IITD-speech-vone/

This package is a useful library for automating content moderation activities by using Machine learning based classifiers.

The package has been updated to python3 and is no longer available for python2

System dependencies required

1. python3
2. ffmpeg: Install it by: sudo apt-get install ffmpeg
3. gcloud: Install it by: curl https://sdk.cloud.google.com | bash and follow the onscreen instructions
4. sudo apt-get install libpq-dev python-dev
5. sudo apt-get install python-numpy libicu-dev

Running instructions

• We recommend creating a virtual environment

git clone https://github.com/abhiaj/IITD-speech-vone.git
cd IITD-speech-vone/
pip install -r requirements.txt
cd IITD-speech-vone/Name/libsvm-3.23
make
cd python
make
cd ../../../
polyglot download embeddings2.hi
polyglot download ner2.hi

You are now ready to run the classifiers and entity extractors using the library functions (the details of the functions is followed).

• Once you have installed then just write up a tester script, for eg. check out test.py

python test.py

Utility functions

utils.remove_silence(input_audio)

Given an input audio in WAV format, this function will remove silence periods from it and make an audio file named input_audio_rmsilence.wav

utils.remove_silence_url(input_url)

Given an input audio web url, this function will remove silence periods from it and make an audio file named temp_rmsilence.wav

utils.get_audio_from_url(input_url, output_file)

Given an input url having mp3 audio file and a output filename, this function generates output audio file

utils.get_transcript(bucket_name, bucket_folder, key_json, audio_folder)

bucket_name -> name of google cloud bucket bucket_folder -> name of the folder in the bucket where we want to upload audio files key_json -> json file with credentials audio_folder -> the folder containing audio files and we need transcripts of these audio files this code outputs a folder named as 'audio_folder'_transcripts which has transcripts in .txt format with same name as audio file

utils.get_gender(input_audio)

Given an input_audio of a file in mp3, this function will try to predict its gender It will return a string denoting male or female

utils.get_gender_url(input_url,download_permanently = False)

Given an input_audio web url of a file in mp3, this function will try to predict its gender It will return a string denoting male or female. If download permanently is set to true then the audio data will be permanently saved.

utils.get_themes(input_file)

This function takes transcpript of audio file in text format as input and returns list of themes relevant to this file

utils.get_quality(input_audio)

Given an input_audio, this function will try to predict its quality It will return 'reject' if audio is rejected(poor quality) and 'accept' if audio is accepted(good quality).

utils.get_quality_url(input_url,download_permanently = False)

Given an input_audio's web url, this function will try to predict its quality It will return 'reject' if audio is rejected(poor quality) and 'accept' if audio is accepted(good quality). If download permanently is set to true then the audio data will be permanently saved.

utils.get_loc(input_transcript)

Given an input_transcript, this function will try to extract embedded location in it It will return a list of dictionaries, each dictionary having a labelled State, District, SubDistrict and Confidence value for it.

utils.get_name(input_transcript)

Given an input_transcript, this function will try to extract embedded name in it It will return a list of strings, each string being a viable name in i

utils.get_dob(sentence)

Given an input hindi sentence , this function will try to extract embedded date in it It will return a dictionary in form of Json Object with keys as 'Date', 'Month' and 'Year'. Value for month could be hindi month or english month whatever the input sentence contains.

NOTE: The version of sklearn library must be "0.22.2" otherwise the models wouldn't give correct results For using automatic transcript utility(function), you need to have google cloud access.

1. Create account on google cloud.
2. Enable google cloud speech api: Go to Navigation menu and select APIs and Services->Library->Speech API(search) and enable it
3. Create a project and a service account for it by navigating to IAM and admin(on Navigation drop down menu) and then go to Service accounts.
4. Generate service account key for the service account created and download the key.
5. This key's path is to be provided in the utility function.
6. Also create a bucket on the Storage menu of Navigation(Drop down menu) and change its settings to public.
7. You also need to set the bucket's permissions for accessing it, i.e. you need to associate your service account with it by linking your account with Storage Object Admin section of Bucket permissions.