Flipkart Grid 2.0 : Software Development Challenge

Solving for Voice Interactions in Indian Houses & Neighborhoods

Contributors

Antreev Singh Brar

Gurbaaz Singh Nandra

Som Tambe

Relevant Links

Github Repository

Demonstration in Colab

Video Demonstration

Audio Clips

Transcripted JSON files

Deliverable PPT : Round 2

Usage

• Clone the repository to
  • Local System:

  git clone https://github.com/antreev-brar/FlipkartGrid.git

  • Google Colab:

  !git clone https://github.com/antreev-brar/FlipkartGrid.git

• Install the dependencies
  • For Windows:

  pip install -r requirements.txt

  • For Linux/Mac:

  pip3 install -r requirements.txt

  • For Google Colab:

  !pip install -r requirements.txt

• Execute the bash script for whole process:
  • Local System:

  chmod u+x run.sh
  ./run.sh input-folder-name

  • Google Colab:

  !chmod u+x run.sh
  !./run.sh input-folder-name

  NOTE: By default, we have made mp3 folder for input-folder-name. So write mp3 in the command line in that place to use default.

Post Execution Process

• ffmpeg converts .mp3 audios and populate the .wav audios in wav folder.
• DTLN performs noise suppression and stores processed wavs in output folder.
• Speaker-Diarization identifies main speaker and stores audio of main speaker in output folder.
• transcription_api.py processes each file and stores the transcriptions in transcription folder.

Dependencies

torch
tensorflow==2.2.0-alpha0
keras==2.3.1
soundfile
wavinfo
pydub
libasound2-dev
portaudio19-dev 
libportaudio2
libportaudiocpp0
ffmpeg
pyaudio
numpy
glob
os

Acknowledgments

• DTLN (Paper)
• Speaker-Diarization (Paper1) (Paper2)