Ajay - (AI-DJ)

Generates a custom radio show with realistic speech transitions between songs.

Questions? Comments? Job opportunities? Contact me: https://linkedin.callumstuart.com

Currently Proof of Concept, and very VERY early in development

Has only been tested on a Debian system so far, may or may not work on Windows.

Example output (Video)

Installation

Requirements (Local and docker install)

NVIDIA drivers for your system

A HuggingFace account with an API key

Install (Local)

Download the repo and cd into it

git clone https://github.com/cstuart1310/AI-DJ.git && cd AI-DJ

Install python requirements

pip install -r requirements.txt

Login to huggingface

huggingface-cli login --token yourTokenHere

Run the script

python ai_dj_main.py --music /path/to/music/

Install (Docker)

Download the CUDA Toolkit to allow CUDA to interface with docker

Download the repo and cd into it

git clone https://github.com/cstuart1310/AI-DJ.git && cd AI-DJ

Build the docker image, passing your huggingface api key generated from HuggingFace.co

docker build . --build-arg HUGGINGFACE_KEY="yourkeyinquotes"

Run the docker image, passing into it the path for models to download to/currently exist. Default downloads for huggingface go to /root/.cache/huggingface/hub docker run -v /path/to/music:/Music -v /path/to/models/:/root/.cache/huggingface/hub --gpus all ai-dj

Command-Line Arguments

• --music /path/to/music The directory containing MP3s to transition between (Default: /Music)

• --intermissions 1/2/3/... The number of songs to play before an intermission. (Default: 1)

• --shuffle Shuffles the order of the files from the music folder

• --output /path/to/output The directory to output the combined MP3 of transitions and songs to

• --length 1/2/3/... The number of songs to process from the given directory

• --subdirs Searches subdirs of the music dir for mp3's, as opposed to just the given music dir

Example run with arguments

Local run:

Full Command

python ai_dj_main.py --music /home/music --intermissions 5 --shuffle --output /home/dj --length 50 --subdirs

Breakdown

• --music /home/music Reads music from /home/music

• --intermissions 5 Plays 5 songs and then a transition

• --shuffle Shuffles the order of the found songs

• --output /home/dj Outputs the combined MP3 file to /home/dj

• --length 50 Only processes 50 songs found in /home/music (Or the maximum if less than 50 are found)

• --subdirs Looks through subdirectories of /home/music for songs

Docker Run:

Full Command

docker run -v /home/music:/inContainer/music -v /root/.cache/huggingface/hub:/root/.cache/huggingface/hub --gpus all ai-dj --music /inContainer/music --intermissions 5 --shuffle --output /inContainer/output --length 50

Breakdown

• -v /home/music:/inContainer/music Binds /home/music (Local directory) to /inContainer/music (Directory within the docker container)

• -v /root/.cache/huggingface/hub:/root/.cache/huggingface/hub Binds /root/.cache/huggingface/hub (Local directory) to /root/.cache/huggingface/hub (Directory within the docker container)

• --gpus all Uses any/all GPUs available to the container

• ai-dj Uses the ai-dj image (If this image is unavailable, you must build it locally)

• --music /inContainer/music Reads music from /inContainer/music (Directory found within the container which binds to /home/music)

• --intermissions 5 Plays 5 songs and then an intermission

• --shuffleShuffles the order of the found songs

• --output /inContainer/output Outputs the combined MP3 file to /inContainer/output (Directory within the docker container)

• --length 50 Only processes 50 songs found in /inContainer/music

Resources

Current LLM used for all tests: Storytime-13B

Current Text-to-audio model: Bark