Jupiter Search

A showcase for indexing jupiter network podcasts using meilisearch. This repository is build in order to provide possible solution to following problems:

• search
• transcription

DISCLAIMER!

Warning! This is a work in progress version to showcase how indexing/transcription might work.

Overview

Project contains two main modules:

• podcast2text a cli tool for downloading RSS feed and transcribing podcast episodes

• search-load a cli tool for loading obtained transcriptions to instance of meilisearch

Building

To build you would need following packages on your system:

• cargo
• pkg-config
• openssl
• ffmpeg

There is a nix flake configured to ship build dependencies just run direnv allow and run:

git submodule update --init --recursive
cargo build --release

To appease the gods of good taste please add following pre commit hook:

git config --local core.hooksPath .githooks

Usage

Run downloading podcasts

Process audio from RSS feed

1. Download the whisper model

mkdir models
# this might be one of:
# "tiny.en" "tiny" "base.en" "base" "small.en" "small" "medium.en" "medium" "large"
model=medium.en
curl --output models/model.bin https://ggml.ggerganov.com/ggml-model-whisper-$model.bin

2. Run the inference on the RSS feed

# get information about the cli
docker run flakm/podcast2text --help

docker run \
    -v $PWD/models:/data/models \
    flakm/podcast2text \
    rss https://feed.jupiter.zone/allshows

Install meilisearch

docker pull getmeili/meilisearch:v0.29
docker run -it --rm \
    -p 7700:7700 \
    -e MEILI_MASTER_KEY='MASTER_KEY'\
    -v $(pwd)/meili_data:/meili_data \
    getmeili/meilisearch:v0.29 \
    meilisearch --env="development"

Run index creation and data loading

Running inference of some audio

1. Download whisper model

mkdir models
# this might be one of:
# "tiny.en" "tiny" "base.en" "base" "small.en" "small" "medium.en" "medium" "large"
model=medium.en
curl --output models/ggml-$model.bin https://ggml.ggerganov.com/ggml-model-whisper-$model.bin

2. Download the example audio from rss feed

curl https://feed.jupiter.zone/link/19057/15745245/55bb5263-04be-43a3-8b92-678072a9cfc8.mp3 -L -o action.mp3

3. Install ffmpeg and put it on PATH variable.

4. Run the inference example

cargo run --release --example=get_transcript -- models/ggml-medium.en.bin action_short.wav | tee output.txt