Speech-Text Alignment

Scripts to align speech audio with their text transcriptions in time.

READMEs for individual scripts and modules are found at directory level.

Setup

Clone Repository

git clone [email protected]:anilkeshwani/speech-text-alignment.git &&
    cd speech-text-alignment &&
    git submodule update --init --recursive --progress

Set Up Environment

Ensure the necessary binary requirements are installed:

apt install sox ffmpeg

Install the package and with it all dependencies including useful dependencies for development; specified via "dev" option to pip install.

conda create -n sardalign python=3.10.6 -y &&
    conda activate sardalign &&
    pip install pip==24.0 &&
    pip install -e .["dev"] &&
    pre-commit install --install-hooks

Note: We do not install the dataclasses library as per the fairseq MMS README it ships out of the box with Python 3.10.6.

Note: When running on Artemis / Poseidon, ensure support for CUDA is provided.

At the time of writing, NVIDIA / CUDA drivers were:

• NVIDIA-SMI: 525.89.02
• Driver Version: 525.89.02
• CUDA Version: 12.0

Data Processing and Performing Tasks

Documentation for performing data processing steps or tasks is found in scripts/README.md.