Conversation is the new reading: Listen to a podcast about your notes or readings with free tools
The full text of this with examples is available on practical-ai.notion.site. And you can listen to a conversation about this text via NotebookLM.
ℹ️ What is text to podcast
Thanks to advances in both Large Language Models and voice synthesis, it is now possible to create very realistic multimedia versions of written documents. And they don’t just have to be simple audio versions of the text but even conversations about them in the form of a podcast or radio interview.
There are several free tools that will allow you to do this. This guide covers two
💬 Text-to-conversation: A new reading modality
Listening to text is becoming increasingly popular (as we can see with the rise of audiobooks and text-to-speech apps) and it has many advantages for cognition but also some limitations. The main one is metacognitive control. Listening while reading often improves attention but the attention becomes linear. In other words, it is hard to jump back and forth in the text to control the flow of information. It is also impossible to skim or scan while listening and thus other elements of metacognition become more difficult.
🧠 Metacognitive control
Conversation (when done in a natural way as all the text-to-podcast tools seem to be able to do) comes with built-in metacognitive controls. Spoken language in a conversational context has very different structure of reference and co-reference that is different even from a typical monologue. Conversation partners focus on maintaining attention and reinforce relationships that the written text can rely on vision to maintain. They also engage in much more repetition and conversation repair.
This can get quite messy in real life and not easy for somebody listening in passively but radio-style conversations improve on this mode by introducing an implicit third party on whose behalf the questions often get asked. Thus podcasts are often more powerful way of letting a novice in on a difficult subject than straight non-fiction narratives.
🪜Cognitive scaffolding
The best role of a conversation about a topic is as cognitive scaffolding. To build context and a sense of familiarity with a wide range of concepts. This is much harder through linear reading, particularly when one does not have sufficient context.
🤙 Strengths and weaknesses of the current technology
🗣️ Amazing voices with intonation
One thing that was holding text-to-speech back now was voices. Not so much the pronunciation of words or sentences but larger intonation patterns.
But this new generation of synthetic voices has completely solved this problem. We can now listen to completely realistic conversation that sounds as good as anything on the radio. So, we’re moving from text-to-speech to text-to-conversation.
🔡 Translation power of LLMs
The new Large Language Models are really good at taking any text and summarising it mostly accurately. They can also very easily and accurately translate this into conversation.
👁️ Hallucination and control of LLMs
However, the biggest problem at the moment is that LLMs do still make mistakes and are relatively hard to control over long stretches of context.
They have also been fine-tuned to be overly helpful in a particular way and may steer the conversation in ways that does not represent the text.
📌 Opportunities: Individualisation
The biggest opportunities here are giving each individual to create podcasts in the style of conversation that best suits them and even giving them a chance to change as their needs and/or preferences changes.
This is relatively trivial to enable through custom instructions - as some tools already do. But it is not straightforward for individuals to create or fine-tune these instructions. So the challenge is not just in creating the interfaces but also guiding users in taking advantage of them.
1️⃣ Converting notes to podcast with NotebookLM
NotebookLM is a free tool for AI-assisted notetaking. You can upload as many as 50 source documents, make your notes and combine them with AI summaries. But a new feature of NotebookLM is the possibility to create an audio overview.
👉 Example
Here’s an example of a podcast of a guide. Here’s a link directly to the NotebookLM.
Recommended by LinkedIn
?How to use this
NotebookLM is free, you can create your account on Notebooklm.google . How to create a podcast:
You can listen:
🤙 Pros and cons
2️⃣ Make a podcast out of a paper with Lettercast.ai
Lettercast.ai is a free service that can
Note: It is not clear what the privacy policies or future pricing plans are. Only use it for public documents such as papers or blog posts.
👉 Example
Lettercast.ai also offers predefined feeds of some predefined sources of information:
?How to use it
You can create an account for free or even use it without an account. You can create a podcast in two steps:
🏁 Getting started
🎧 Listening options
You can listen in two ways:
➕ Advanced creation options
There are other advanced options you have:
♻️ Alternatives
There are many other tools that focus on creating podcasts using AI voices. Some examples are:
Global Symbols and Senior Research Fellow University of Southampton
3moThank you so much for this really interesting post. 😊
Head of Technology with over 16 years of experience in e-Learning Operations and Technology.
3moGreat insight on the adaptation of LLMs, thank you Dominik Lukes
Marketing Manager | Digital Marketing Specialist | Developing Marketing Automation and Workflows to increase team efficiency | Blog/Podcast Host at Untapped: Higher Education
3moFascinating stuff Dominik Lukes I wonder if this might make it a bit easier to disseminate notes after a conference, company town hall or some other long-form discussion. Any advice on how well it shortens/summarizes long documentation?