I Cloned My Favorite Podcast Host (with AI Voice Cloning)
- Deepgram: tiny.one/NGQ6d0f to get $200 free credit
- Code Tutorial + Overview: tiny.one/J44NVfg
OUTLINE:
0:00 - Introduction
0:40 - Deepgram
1:29 - The Process
2:13 - Background
2:24 - Memories
2:58 - Tone Extraction
3:14 - Create The Audio
3:33 - The Results
OVERVIEW:
I’m Greg Kamradt, and I’m on a mission to figure out how businesses will create more value using AI. In this overview we explore how to clone a person with a language model. We use Shaan Puri (host on My First Million) as a test subject.
We explore 3 tactics to do this:
1) Background Knowledge - Attributes about a person
2) Memories - Specific stories or advice they share
3) Tone Examples/Descriptions - Description of how they speak and examples of vocabulary
It’s important for product teams and businesses to understand these tactics so they can personalize their products. LLMs are a new world and we’re figuring out how to best use them.
Come on the journey as I explore what AI means for business.
Sponsors that help support the channel:
- Deepgram (Transcription Services): tiny.one/NGQ6d0f
- SingleStore (All In One Database): tiny.one/tvvUv6Z
GREG’S INFO:
- Twitter: tiny.one/sIY2j61
- Newsletter: tiny.one/vXzrYJ3
- Website: tiny.one/T948oRT
- LinkedIn: tiny.one/knMMWIw
- Work with me: tiny.one/6AZ890O
- Contact Me: Twitter DM, LinkedIn Message, or contact@dataindependent.com
You are such a great teacher! Nice to meet you and I’m stoked to learn more from you 🎉
Awesome! Thank you Yanik! I appreciate it. What are you building?
@@DataIndependent Hey man, so I'm a backend senior software developer (ruby/rails primarily) but I am so fascinated by the kinds of things one can build with AI that I'm learning python and trying to find resources about AI online. That's where I found you. Any chance you tutor? There are several ideas I have, but in the context of this video, my sibling has an issue where she lost all the recordings of a bunch of coaches for her new startup idea - she. has the scripts and snippets of their voice. So I wanted to see if I could use the two to re-create some coaching audio she lost.
@@yanikjayaram yeah you could do a bunch there. I put out this tweet which is all AI generated x.com/gregkamradt/status/1714342098496078128?s=46 Video/audio are semi easy now but the content of what someone would say is the hard part to build
yoooooo Shaan is gonna love this!
It would be interesting if you could not only do it in their tone of voice, but also translated into a different language. This would automate any dubbing etc and actually have the same voices with the same emotions as the original actor.
Nice - ya that would be pretty cool. I think I've seen some demos of that and I've actually debated putting my channel into other languages but haven't pulled the trigger yet
That's awesome. Were you able to transcript each speaker of the podcast individually? I have also done the transcripts of my favourite podcast (the whisper model is open source, I did it locally on my laptop) but I do not have the different speakers. Cheers, keep it up
Hey thanks! Yep - Deepgram was able to separate out the speakers for me which is nice. They called it 'speaker diarization'
great workflow - thank you for sharing
Thanks Micbab!
Amazing. fasted sub of my life. thank you for the content.
Nice Thank you Amauta. More coming soon!
Greg, that was some change about the past videos! Great edit and new fast format! Congrats
Thanks! What do you think about that new format? The balance between entertainment and education is tough. I'm going to work to tip toe that line well.
WoW 🔥 That is something. Did you do the interaction later with Deepgram itself or did you used our own RAG with a TTS setup, where TTS is powered by Deepgram?
Deepgram was just used for the transcription service - Audio > text. In the future I'm going to use them for streaming which will be fun
Great content, keep it up! Who is getting cloned next?
Nice - who do you want?
this is cool!
Thanks Sam! Been following you since the anti-mba/itch juice days Also - working on my 405 squat and 315 bench as we speak. It's an absolute grind to get there.
How can I reproduce my own voice? Great Video
Thanks! Head over to Plat.ht and try it out
Great idea for virality. He will for sure share this. Btw, this is the only ai channel worth watching
Nice thank you! Shaan ended up giving a shout out which was awesome to see twitter.com/ShaanVP/status/1693722364062961752
Your demo seems fake, the latency is too low to be realistic. You definitely edited and cut the latency. Still great though, but you should disclose that you edit out out the latency in post.
Nice! Thank you for the tip
Always interesting subjects, but really sad that you focus on crappy 'business values' instead of how this thing can change the world to something better than a gazilion idiots fighting for their lives on markets. Guess I have to wait a year for the current system to fall apart for you to talk about open source AI for betterment of humanity. Back to your very important value for business.. sigh..
Oh, and sorry but that Podcast guy sounds like a typical 'life-coach' scammer :)
Hey thanks@@sgramstrup - I needed some help from ChatGPT to turn your comment into a positive tone. Here's what it said "Always interesting subjects": This suggests that the person finds the topics you choose to be intriguing and worth watching. They acknowledge that you consistently pick subjects that captivate their interest. "focus on how this thing can change the world to something better": This indicates that the commenter believes in the potential of the subjects you discuss. They see a broader vision and potential for positive change, which means they think your content is important and impactful. "open source AI for the betterment of humanity": This suggests a topic of interest to the commenter. It could be taken as a suggestion for future content if it aligns with your channel's theme and objectives.
Love this! Great work Greg and sub’d on your website to create my own ‘heroes’ 🥳🤩🦾
nice! Thanks Klammer