Another leap in AI - PODCASTS!

[GregB]

Prepare to have your socks, underwear and everything else, perhaps even your skin, blown off!!  Even the cynic in me is shocked!

 

A couple of days ago Google release a new product that leverage their own AI Chat system (Gemini 1.5).

Called NotebookLM (https://notebooklm.google/) it takes written text and generates an audio conversation/podcast.

 

It is FREE to use.

 

“Ho-hum”, I thought, “how crap is THIS going to be?”

 

I simply gave it the script I had used for my most recent Episode (copy/pasting from Word into a Google Doc in my Google account).  The scrip is HERE if you want a quick squiz ... ...  https://docs.google.com/document/d/1z5gx0oMIAMPU553XB9nOTAigzbmg_FjWJF98o96kPVs/edit?usp=sharing

 

I didn’t think to time it, but it took less than 5 minutes to generate a 10-min WAV file that I could play and/or download.

 

Not only did it convert my first-person narrative to a conversation, but the conversation is VERY organic, the voices SOUND natural (albeit American-style enthusiastic) AND the spoken words are an amazingly good re-worded amalgam of what I had written.  It has kept the essence and has NOT made any mistakes regarding content, names, titles etc.

 

I’ve converted the WAV to MP3 for inclusion here. 

 

 

The WAV was mono only.

 

It would be very easy to edit this file and insert my own musical extracts.

 

The limitations are that all AI podcasts would start to SOUND the same ... sonically, accents, verbal/conversational rhythms, etc..   

 

But the concept and actualisation are just astonishing.  It certainly sound a lot more lively than my droning/stumbling monologues ... 

https://podcasters.spotify.com/pod/show/dumbit/episodes/6-The-Prescient-viewed-from-the-Future-e2nd9gt

 

 

Greg

[MisterB]

It's interesting that one side of Google puts out that, and another side puts out this - https://support.google.com/youtube/answer/14328491?hl=en.

 

I just tried it on one of my scripts. Pretty amazing results. It really sounds like a radio program. The accuracy was about the same as your typical reporting too. Definitely ad libbed a bit too much, but pretty real sounding. The problem for me is that I find the whole two person radio interview thing really annoying.

[GregB]

The other thing I didn't mention, as it seem a bit TOO nit-picky considering the amazing result, was that the 'podcast chat' I generated was a) far too Amercian, and b) far too 'young' (i.e. vapid and overly enthusiastic).  "Amazing", "100%", "totally", "right?", "so inspiring", etc., etc..  All reasons I've stopped subscribing to a few REAL podcasts involving young songwriters/artists.
 

Instead of high-energy and exclamatory, I prefer laid-back and contemplative.  

 

But then again, as always, this is the first toe-in-the-water of AI generating 'conversations'. In just another 6 months we'll  probably be able to dial in the number of speakers, their age, sex, accent, energy level and percentile contribution.

[MisterB]

I think that regardless of the actual generated audio “radio programme”, this is the best tool i’ve ever seen to help with writing material for a podcast. 

[john]

It even adds breathing, which completely adds to the feeling of authenticity, as does talking over each other which is done at an understated level.

Impressive.

 

Much better than a lot of the voice to text AI which often screws up numbers and when to interpret letters as words or letters to be read out.

 

Not long before you can choose the characters, perspective, emotional perspectives and accents. I know some of that is there in other tools, but not to this level of realism.

 

Interesting Greg!