Keywordsearch

This project is created for the ViFrU. The program can create transcriptions of wav and mp3 files by leveraging the latest Mozilla Deepspeech. Transcriptions can be searched for entered keywords. Abstracts and audio snippets of the found keywords can be extracted.

Installation

As a first step clone the repo by using the following command:

git clone https://github.com/Thomson2412/keywordsearch

Setup script

A setup script is provided to ease the installation process. However, this script is only tested on the Ubuntu Linux distro. Other Ubuntu based distributions might work but are not tested. Operating systems besides Linux are not supported by the setup script.

The setup script will install miniconda, create a new conda environment, download the required dependencies and the required deepspeech models and will create a desktop entry.

To run the setup script use the following command:

./install

The program can take advantage of gpu acceleration. Incase the system has a compatible NVIDIA gpu please pass the following argument to the install script:

./install --gpu

Manual

If, for whatever reason, the setup script cannot be used, manual installation can be performed with the following commands.

Miniconda installation:

wget https://repo.anaconda.com/miniconda/Miniconda3-latest-Linux-x86_64.sh -O install_files/miniconda.sh
chmod +x miniconda.sh
./miniconda.sh

Downloading deepspeech models (execute in the root folder of the project):

mkdir -p data/models
wget https://github.com/mozilla/DeepSpeech/releases/download/v0.9.3/deepspeech-0.9.3-models.pbmm -O data/models/deepspeech-0.9.3-models.pbmm
wget https://github.com/mozilla/DeepSpeech/releases/download/v0.9.3/deepspeech-0.9.3-models.scorer -O data/models/deepspeech-0.9.3-models.scorer

Creating conda environment:

conda env remove --name keywordsearch
conda create --name keywordsearch python=3.9
conda activate keywordsearch

Installing dependencies (execute in the root folder of the project with environment activated):

sudo apt install ffmpeg
sudo apt install gcc
sudo apt install portaudio19-dev

For usage without gpu acceleration use the following commands:

pip install .[cpu]

For usage with gpu acceleration use the following commands:

conda install cudatoolkit=10.1 cudnn=7.6
pip install .[gpu]

Usage

Keywordsearch can be used with a GUI or via the CLI.

GUI

To start the GUI use the desktop entry created by the setup script or use the following commands:

conda activate keywordsearch
python main.py

CLI

To use the CLI make sure the environment is activated with the following command:

conda activate keywordsearch

CLI usage:

python kwscli.py "FILES" --output "PATH TO OUTPUT FOLDER"

Example:

python kwscli.py input/test_file_1.mp3 input/test_file_2.wav --output output_folder/test_files/