Step 1: PDF Processing (Notebook 1)

The first notebook processes the input PDF using the Llama-3.2-1B-Instruct model. This step involves:

1. Loading the PDF document
2. Preprocessing the text data
3. Using the Llama-3.2-1B-Instruct model to extract relevant information

python
​
# Install required libraries
!pip install -r requirements.txt
​
# Load the PDF document
from PyPDF2 import PdfReader
pdf_reader = PdfReader('input.pdf')
​
# Preprocess the text data
text = ''
for page in pdf_reader.pages:
    text += page.extract_text()
​
# Use Llama-3.2-1B-Instruct to extract relevant information
from transformers import AutoModelForCausalLM, AutoTokenizer
model = AutoModelForCausalLM.from_pretrained('Llama-3.2-1B-Instruct')
tokenizer = AutoTokenizer.from_pretrained('Llama-3.2-1B-Instruct')
inputs = tokenizer(text, return_tensors='pt')
outputs = model.generate(**inputs)
        

Step 2: Podcast Transcript Generation (Notebook 2)

The second notebook takes the processed output from Notebook 1 and generates a podcast transcript using the Llama-3.1-70B-Instruct model.

python
​
# Load the processed text data
with open('processed_text.txt', 'r') as f:
    text = f.read()
​
# Use Llama-3.1-70B-Instruct to generate the podcast transcript
from transformers import AutoModelForCausalLM, AutoTokenizer
model = AutoModelForCausalLM.from_pretrained('Llama-3.1-70B-Instruct')
tokenizer = AutoTokenizer.from_pretrained('Llama-3.1-70B-Instruct')
inputs = tokenizer(text, return_tensors='pt')
outputs = model.generate(**inputs)
transcript = tokenizer.decode(outputs[0], skip_special_tokens=True)
        

Step 3: Transcript Enhancement (Notebook 3)

The third notebook enhances the podcast transcript by adding dramatization and interruptions using the Llama-3.1-70B-Instruct model.

python
​
# Load the podcast transcript
with open('transcript.txt', 'r') as f:
    transcript = f.read()
​
# Use Llama-3.1-8B-Instruct to enhance the transcript
from transformers import AutoModelForCausalLM, AutoTokenizer
model = AutoModelForCausalLM.from_pretrained('Llama-3.1-8B-Instruct')
tokenizer = AutoTokenizer.from_pretrained('Llama-3.1-8B-Instruct')
inputs = tokenizer(transcript, return_tensors='pt')
outputs = model.generate(**inputs)
enhanced_transcript = tokenizer.decode(outputs[0], skip_special_tokens=True)
        

Step 4: Podcast Generation (Notebook 4)

The final notebook generates the podcast using enhanced transcript and text-to-speech models: parler-tts/parler-tts-mini-v1 and bark/suno.

python
​
# Load the enhanced transcript
with open('enhanced_transcript.txt', 'r') as f:
    transcript = f.read()
​
# Use text-to-speech models to generate the podcast
from transformers import AutoModelForTextToWaveform, AutoProcessor
model = AutoModelForTextToWaveform.from_pretrained('parler-tts/parler-tts-mini-v1')
processor = AutoProcessor.from_pretrained('parler-tts/parler-tts-mini-v1')
inputs = processor(transcript, return_tensors='pt')
outputs = model.generate(**inputs)