BriefCast

BriefCast is an AI-powered podcast generator from RSS feeds.

How It Works

1. You pick a website that offers a RSS feed.
2. BriefCast fetches the feed and coverts it to a short podcast transcript.
3. Finally, the transcript is converted into MP3 audio that you can listen to on any device.

Technologies Used

Frontend

• Svelte for a reactive UIs.
• Tailwind CSS for quick styling
• Howler.js for Audio Player

Backend

• Deno for TypeScript built-in API server.
• Firebase for user management and data storage. Details are below.
• OpenAI GPT-3.5 for generating a podcast transcription from the given RSS feed.
• Google Text-to-Speech API for converting the transcription into podcast-sounding MP3.

How Firebase is Integrated

• Firebase Authentication is used to allow users to sign in/out with their Google accounts as well as email addresses. app/src/lib/firebase.ts handles all utilities for Firebase Authentication. Firebase Firestore stores all transactional data, including podcast playlists, crawled feed contents, and generated transcripts. core/lib/firebase.ts initializes the Firestore client, and core/repository/*.ts handles all Firestore operations. Firebase Analytics tracks user activities on our frontend application, such as screen views. app/src/lib/firebase.ts includes all necessary operations.

Testing Strategy

Frontend

• Unit tests are written for specifically important or complex logic.
• Vitest is used.
• npm run vitest runs all tests.

Backend

• Unit tests are written for specifically important or complex logic.
• Deno built-in test runner is used.
• deno task test runs all tests.

Getting Started

If you want to run this project on your local machine, follow the steps below.

Pre-requisites

1. Install Docker, Docker Compose.
2. Install VSCode and VSCode Remote Dev Containers extension (ms-vscode-remote.remote-containers).

Set up the environment variables

1. Create a .env file in the project root directory.
2. Add the following environment variables to the .env file.

GOOGLE_APPLICATION_CREDENTIALS=<path to your GCP service account key file>
OPEN_AI_API_KEY=<your OpenAI API key>
PUBLIC_FIREBASE_API_KEY=<your Firebase API key>
PUBLIC_FIREBASE_AUTH_DOMAIN=<your Firebase auth domain>
PUBLIC_FIREBASE_PROJECT_ID=<your Firebase project ID>
PUBLIC_FIREBASE_STORAGE_BUCKET=<your Firebase storage bucket>
PUBLIC_FIREBASE_MESSAGING_SENDER_ID=<your Firebase messaging sender ID>
PUBLIC_FIREBASE_APP_ID=<your Firebase app ID>
PUBLIC_FIREBASE_MEASUREMENT_ID=<your Firebase measurement ID>
FIREBASE_ADMIN_EMAIL=<your Firebase admin email>
FIREBASE_ADMIN_PASSWD=<your Firebase admin password>
MEDIA_PATH=<path to the directory to store generated MP3>
LOG_PATH=<path to the directory to store logs>
NPM_CACHE=<path to the directory to store NPM cache>
HOST=<your host, not necessary for the dev environment>

Start the development server

1. Clone this repository.
2. Open the core directory in VSCode.
3. Select Remote-Containers: Reopen in Container
4. Open terminal in the VSCode and run ./run_dev.sh to start the API server.
5. Open the project directory in another VSCode window.
6. Select Remote-Containers: Reopen in Container
7. Open terminal in the VSCode and run npm run dev to start the frontend server.

What's Changed Recently

September 2023

• Add player UI.
• Use howler.js as a podcast player engine.

June 2023

• Delete too old caches periodically.

Show old records

May 2023

• Add UI to show a transript.

April 2023

• Add pause after line break in a transcript.
• Add Language selector on the add podcast page.
• Add voice gender selection option.
• Use podcast title in the transcript instead of the feed's title.
• Generate 'all-combined mp3' for 'Play All'.
• Generate a podcast on add.
• Show toast after adding new podcast.
• Skip podcast generation if feed is not changed and MP3 has been created already.
• Added Terms of Use and Privacy Policy.
• Add 'Play All' button.
• Add loading spinners.
• Fix Google Text-to-Speech API fails on a long text.
• Prompt to add a podcast after signing up.
• Add logger.
• Add 'beta' to the title.
• Add Google Analytics.
• Prepare production environment.

March 2023

• Supported updating user podcasts twice a day.
• Allowed a user to add RSS feeds and listen each feed's podcast from the top page.
• Added sign-in with Google.
• Supported SSL.
• Tuned the default English summarizer to minimize unrelated linking words between topics, referring to this article: https://github.com/openai/openai-cookbook/blob/main/techniques_to_improve_reliability.md
• Inserted 2 second pause between topics so we can tell the beginning of the news more easily.
• Added a Trial feature that supports one-time pod cast generation from any news feed for anyone.
• Implemented automatic CI/CD pipeline with each new code update.

TODOs

• Improve overall UI.
• Add sharing my podcast recipe feature.
• Limit trial generation based on the IP address.
• Limit the total number of generations within a day (300 times a day would be the upper limit for my budget).
• Add 'Add Website' button, allowing users to easily add a podcast by specifying a website URL.
• ...

License

Source code in this repository is licensed under the PolyForm Noncommercial License 1.0.0.