Google's rocking it (while Apple socks)
This is an unusual blog post in that I don't have anything scripted. I am not using AI to edit the text; I am just winging it on the fly by dictating.
Full Bio
I saw that Warren Buffett just bought about $5B worth of Google while he has been trimming Apple 15% in the last quarter.
Apple's only recent "innovation" has been their cut-up sock. I did hear, to their credit, that it was only sold in New York City and that the lines were around the block because of its exclusivity.
Though I got three new iPhone 16s this year for the rest of the family, when I upgrade my iPhone 14 ProMax, I am almost certainly going for a Samsung, probably the s25 Ultra or the latest model. The cameras are way better in the Samsung.
Apple has fallen so far behind in AI that it is about to pay Google $1Billion a year to use Gemini for it's Apple intelligence in Siri !
I am planning to abandon the Apple ecosystem after being an Apple fanboy and avid iOS user since the iPhone 4S. And I may not be the only one..
I just subscribed to Google AI Pro for $20/month to get more flexibility with NotebookLM. The plan actually gives me additional credits across the Google ecosystem, like more images within Google Gemini 3, aka Nano Banana pro.
A 4 minute Podcast on Nano Banana Pro
It's currently THE best image editing software in the world, and you can use it for FREE at https://gemini.google.com/
Tip: Simply add --ar 16:9 etc. at the end to specify aspect ratio, and --res 4K at the end of the prompt to get a high resolution output
NanoBanana Pro is so good that I just wrote a separate blog post about it. https://www.sebastianantony.com/10-mind-blowing-use-cases-for-nano-banana-pro/
As a heavy AI Art user, to earn more credits, I just purchased the Black Friday 50% off "Infinite Plan" special at https://openart.ai/home for $28 a month, paid annually upfront. This allows 24,000 credits a month, enough for 400 images at 2K resolution. Higgsfield was another platform I had considered, but I read that they throttle user speeds on the larger plans.
42Courses
Using NotebookLM, I had uploaded about 28 PDF files and created a podcast about 42 courses. You may recollect that it's the learning platform that I was immersed in during my 6-week sabbatical last year.
Looks like some upgrades were done in the podcast studio?
The UK-based 42 courses are a fantastic experience, and every time a course is released, I am so excited. Their founder, Chris, is a wonderful human being with a great sense of humor. The weekly newsletters have interesting factoids, well-curated links, and witty material.
There is so much to write about 42Courses that it deserves its own blog post. https://www.sebastianantony.com/42-courses/
Finding my Voice
The two podcasts above were created by adding many links to NotebookLM and letting it generate them. If I'm not happy with the podcast, I re-prompt, giving more specific instructions.
All NotebookLM podcasts have the same pair of voices :(
13 Minute Audio only podcast
To be a bit different, I tried using the voice-changing feature in LemonSlice, which creates talking heads videos. Unfortunately, the talking head speech in LemonSlice only requires 60 seconds of training, so my accent wasn't very good.
LemonSlice can separate male and female voices in a NotebookLM podcast (and I believe without using any credits).
I first experimented with adding voice-over to the male parts. Unfortunately, in CapCut, each time you pause it opens up the male voice in a different audio track. And it's hard to remember the exact sequence of the speech, to add your own voice-over in the podcast. In a nutshell, using CapCut to manually voice over your own voice is a royal pain.
Research led me down the rabbit hole of IVC vs PVC. (Instant voice cloning vs. professional voice cloning).
For AI-generated audio, ElevenLabs is widely regarded as the best software. In the interest of science, I signed up for the first month at 50% off at $11.
11 labs has PVC, where you speak for a minimum of 30 minutes, ideally up to 2 hours, and the AI learns to generate audio in your voice.
I discovered that when you use a professionally trained voice to replace an existing audio track, it's not perfect because it has to maintain the original speech's cadence and inflection.
But if you use text-to-speech, it is shockingly good. I saved a snippet of a little story I saw on Facebook or LinkedIn and intended to share it because I bought more than a dozen books just in the past month. Below is this story read in my voice via ElevenLabs, and it sounds exactly like me!!!
Obviously, there is a danger of scams, like
- impersonating somebody at work or a celebrity.
- somebody impersonating you by calling your spouse and saying that your children have been in an accident and need money wired over for hospital bills
To mitigate misuse of your voice, ElevenLabs verifies that you are authorized and that you are only training your own voice. After you record the 30-minute PVC, they ask you to read a random piece of text displayed on the screen. It is only a short phrase, maybe 10-15 seconds, but the voice must match before the PVC is created.
WisprFLOW Speech to Text just raised $25M to accelerate the future of voice-to-text, bringing total funding to $81M. This round was led by Notable Capital, along with Flight Fund, the investment arm of entrepreneur and Diary of a CEO host Steven Bartlett.
For my next NotebookLM podcast, I'm going to try something different. I'm going to play the podcast back through Wispr Flow on my iPhone, getting the complete transcript, and then I'll convert audio of both male and female parts using 11 labs. I will use obviously my PVC for the male part. Let's see how that turns out...
Misc.
Last week, I made a short unlisted music video to celebrate the birthday of one of the managers at work. It took perhaps four or five hours because I had to research the manager's likes, such as favorite taste in music or hobbies. I had to use ChatGPT to create custom lyrics. I found out the manager attended a Metallica concert, so I looked up concerts in Charlotte, identified which tracks were suitable for a birthday celebration, and wrote a parody using custom ChatGPT lyrics. Then I had to make several short video segments in Kling.
This blog post is getting too long, and I'm rambling, so I want to wish you all a wonderful Thanksgiving. For those who are looking for bargains, I wish you all the best on Black Friday and Cyber Monday sales, which are usually the lowest prices of the year.
