FiCast (formly FiCast-GPT)

A Python package for processing & producing PodCast overlayed with music aligned to the podcast : "Fi-Casts" (LoFi + Podcast)

Table of Contents

• Introduction
• Installation
• Usage
• Optional: Using Tortoise-TTS for Local Inference
• Contributing
• License

Introduction

Tool for processing & producing musical podcasts w/ thoughtful multi-agents

Installation

• Installing from Test PyPI

  python3 -m pip install -i https://test.pypi.org/simple/ ficast-gpt
  

• Installing from Test PyPI using poetry:

  poetry add ficast-gpt --source https://test.pypi.org/simple/
  

• Required environment / keys configuration
  1. conf/OAI_CONFIG_LIST.txt following the syntax of PyAutogen, the framework this package is based on to configure your LLM clients.
     • Note: Please add the followin safety settings to the model configuration to avoid safety-related filter errors during generation.

     "safety_settings": [
       {
           "category": "HARM_CATEGORY_HARASSMENT",
           "threshold": "BLOCK_NONE"
       },
       {
           "category": "HARM_CATEGORY_HATE_SPEECH",
           "threshold": "BLOCK_NONE"
       },
       {
           "category": "HARM_CATEGORY_SEXUALLY_EXPLICIT",
           "threshold": "BLOCK_NONE"
       },
       {
           "category": "HARM_CATEGORY_DANGEROUS_CONTENT",
           "threshold": "BLOCK_NONE"
       }
       ]
     

  2. ElevenLabs Transcription API Key in .env as such (see example)

     ELEVENLABS_API_KEY=<Your_Elevelabs_API_Key>
     

     Optionally, you could use Tortoise-TTS for Local Inference. This set-up is more complex and is currently under development.

Usage

• Environment

Check out the tutorial notebook for the following example.

from ficast.character.podcast import Podcaster
from ficast.conversation.podcast import Podcast

my_podcast = Podcast(
  # define podcast topic
  topic="health & longevity", 
  # define number of rounds
  n_rounds=20, 
)

# Initialize participant instances
joe_rogan = Podcaster(
  name="Joe Rogan", 
  description="Joe Rogan is a funny and popular podcast host",
  model="gpt-3.5-turbo",
  role="host"
)
david_sinclair = Podcaster(
  name="David Sinclair",
  description="David Sinclair is a Harvard Medical Expert",
  model="gpt-3.5-turbo"
  # default role as a guest
)

# Aadd both podcast participants to conversation
my_podcast.add([joe_rogan, david_sinclair])

# Create the podcast dialogues
my_script = my_podcast.create()

# check the scripts
# raw string script
my_podcast.raw_script
# return chat history from script (warning: this part can fail due to the inconsistency of formatting that may result from individual dialogues from the LLMs)
my_podcast.script
# a structured json script parsed 
my_podcast.json_script

from ficast.assembly.ficast import FiCast
fantasy_ficast = FiCast(conversation=my_podcast)

# Convert the conversation to an audio podcast
my_audio = fantasy_ficast.to_podcast()

# Save / Play Audio!
from datetime import datetime
fantasy_ficast.save_podcast()

# play audio
from elevenlabs import play
play(bytes(my_audio, 'latin-1'))

Managing Git Submodules

This repository contains submodules for two key components of the FiCast project:

1. FiCast Web Frontend: The web frontend for FiCast is located in the apps/web directory and is a separate repository hosted privately at FiCast-Frontend.
2. Tortoise-TTS: The text-to-speech (TTS) engine is located in the tortoise-tts directory and is hosted as a separate submodule.

Cloning the Repository with Submodules

When cloning this repository for the first time, you need to initialize the submodules to ensure you get the content for both apps/web and tortoise-tts.

Run the following commands to clone the repository and initialize the submodules:

git clone git@github.com:supermomo668/FiCast-GPT.git
cd FiCast-GPT
git submodule update --init --recursive

This will ensure that both the FiCast Web Frontend (apps/web) and Tortoise-TTS (tortoise-tts) submodules are correctly fetched.

Autogen Workflow Diagram (An Illustration of how the conversation takes place)

Optional: Using Tortoise-TTS for Local Inference

If you prefer to use the Tortoise-TTS for local inference instead of the ElevenLabs API, you need to pull the tortoise-tts submodule. Follow the instructions below to set it up.

[Alternative] Steps to Pull the Tortoise-TTS Submodule

• Only the commands if you need the submodule:

1. Initialize and Update the Submodule:

   git submodule init
   git submodule update

2. Clone the Submodule Directly (if not already cloned):

   git submodule add https://github.com/neonbjb/tortoise-tts.git path/to/tortoise-tts

3. Setup Tortoise-TTS for Inference:

   Follow any additional setup instructions provided in the tortoise-tts repository to complete the configuration for local inference.

Important: What NOT to Do If Not Using the Submodule

If you do not need to use the Tortoise-TTS submodule for local inference, do not run the submodule initialization and update commands. Additionally, you can avoid pulling the submodule by skipping the steps mentioned above.

Contributing

Guidelines for contributing to the project.

License

Apache 2.0