Podcasts - From Content to Publication

This page outlines my workflow for generating podcast content based on species observations. It covers the use of AI tools like ChatGPT and NotebookLM, script preparation, and plans for recording and publishing enriched episodes on Spotify and Apple Podcast.

Last changed on 2025-05-10

Collecting Information

I have created a list of around 40 questions about animal species, covering topics such as distribution, physiology, behavior, conservation status, and scientific importance. This list is parameterized and can be applied to most of the species I observe in the area.

A Python script connects to OpenAI's GPT-4o model to generate an answer for each question. To ensure accurate responses, the system prompt and temperature settings used for each request are crucial.


system_prompt = (
    f"You are a knowledgeable and precise biology assistant answering structured questions "
    f"about the species: {species}, with a focus on the topic: '{topic}'. "
    f"Respond with factual, clear, and well-reasoned explanations, suitable for naturalists and non-specialist biologists. "
    f"Ensure each answer is complete and informative, avoiding vague or overly short responses. "
    f"If the topic is complex (e.g., reproduction, threats, or ecological role), include relevant context, regional variation, "
    f"and real-world examples where appropriate. If reliable information is limited, explain that transparently."
)

response = openai.ChatCompletion.create(
    model=model,
    messages=[
        {"role": "system", "content": system_prompt},
        {"role": "user", "content": question}
    ],
    temperature=0.3
)

The generated answers are compiled into a Markdown file, which is then converted into a PDF.

Creating the Podcast

This file serves as source for generating the podcast with NotebookLM Plus. The prompt to generate the audio overview is very important.


Important:
- Length Audio Overview 45 minutes
- Use metric measurement
- Don't repeat information
- Short break between chapters

The chapters are:
- Intro read: "You're listening to Muuuh Nature Documentaries."
- Distribution and Habitat
- Physical Traits and Adaptations
- Reproduction and Lifecycle
- Diet and Predators
- Ecosystem Role
- Threats and Conservation
- Scientific and Educational Importance
- Outro read: "Credit to Muuuh Nature Documentaries, muuuh.com, owned by Karl-Heinz Müller"

The resulting audio file is downloaded, but directing the audio output is currently limited. If the generated episode is longer than 60 minutes or shorter than 30 minutes, a new version is generated. The audio file is immediately published as a new episode, but it is later imported into DaVinci Resolve for editing. Publishing the file right away allows me to listen to the episode and take notes on any issues that need to be addressed in post-production.

Editing the Podcast in DaVinci Resolve

In DaVinci Resolve, the following edits are done:

AI Tools: Checkerboard to new tracks (this allows generating a more balanced audio later)
Listening to the whole podcast and remove sound glitches
Add field sound recordings if available
Add music to intro, breaks and outro
AI Tools: Create mix
Export audio only as mp3

Publishing the Podcast

NotebookLM Plus is used to create a short description about the podcast published in Spotify.


Can you provide a short description, 2-3 sentences, for the deep dive podcast about this research?

You find the published podcast episodes here:

Content Accuracy

The podcast created using this approach involves two steps, both relying on large language models (LLMs): first, gathering information with ChatGPT, and then generating the podcast script using Gemini. In other words, AI hallucinations and factual inaccuracies can become compounded.

Spotted Hyena (Crocuta crocuta)- Biology, Behavior and Ecology - This episode is special for an obvious reason: the Spotted Hyena is neither native nor introduced in the region I typically cover.

To illustrate the extent of potential errors, I was fortunate to receive a review from Oliver Höner—the world’s leading expert on the Spotted Hyena—regarding the podcast generated about this species. He identified 15 factual errors or imprecise statements in the final version.

Despite the high number of factual errors or imprecise statements, I continue publishing these podcasts—but moving forward, I will carefully review my sources and ensure the final versions are accurate. Even with occasional inaccuracies, these episodes remain highly valuable, offering substantial insight and serving as a useful reference—even for myself—when learning more about the animals I encounter on my field trips.