Inside the AI Audio Revolution: Translating, Conversing, & Composing | [Sidecar Sync Episode 98]

Summary:

In this audio-focused episode of the Sidecar Sync, Mallory Mejias and Amith Nagarajan explore three groundbreaking advancements in AI audio technology—starting with Google’s real-time voice translation for seamless cross-language communication. They then unpack Microsoft’s open-source Vibe Voice, capable of generating full-length, multi-speaker podcasts, before diving into ElevenLabs Music, an AI tool creating fully-licensed, studio-quality tracks from simple prompts. Tune in for laughs, live demos, and serious implications for associations, from global content reach to AI-driven personalization. Plus, hear the hilarious tale of Amith’s early-morning car debacle and Mallory’s AI-generated bounce anthem for associations. This one hits all the right notes!
Timestamps:

📅 Find out more digitalNow 2025 and register now:

https://digitalnow.sidecar.ai/

🤖 Join the AI Mastermind:

https://sidecar.ai/association-ai-mas...

🔎 Check out Sidecar's AI Learning Hub and get your Association AI Professional (AAiP) certification:

https://learn.sidecar.ai/

📕 Download ‘Ascend 2nd Edition: Unlocking the Power of AI for Associations’ for FREE

https://sidecar.ai/ai

🛠 AI Tools and Resources Mentioned in This Episode:

Google’s Real-Time Translate ➡ https://blog.google/products/translate/language-learning-live-translate
Microsoft Vibe Voice ➡ https://microsoft.github.io/VibeVoice
ElevenLabs Music ➡ https://elevenlabs.io/music
HeyGen ➡ https://www.heygen.com
Notebook LM ➡ https://notebooklm.google
Rasa.io ➡ https://rasa.io
LM Studio ➡ https://lmstudio.ai
Suno ➡ https://www.suno.ai
Sidecar Sync: Online Learning Course Creation at 26x Speed ➡ https://podcasts.apple.com/us/podcast/online-learning-course-creation-at-26x-speed-86/id1714325496?i=1000712572863

👍 Please Like & Subscribe!

https://www.linkedin.com/company/sidecar-global

https://twitter.com/sidecarglobal

https://www.youtube.com/@SidecarSync

Follow Sidecar on LinkedIn

⚙️ Other Resources from Sidecar: 

• Sidecar Blog
• Sidecar Community
• digitalNow Conference
• Upcoming Webinars and Events
• Association AI Mastermind Group

More about Your Hosts:

Amith Nagarajan is the Chairman of Blue Cypress 🔗 https://BlueCypress.io, a family of purpose-driven companies and proud practitioners of Conscious Capitalism. The Blue Cypress companies focus on helping associations, non-profits, and other purpose-driven organizations achieve long-term success. Amith is also an active early-stage investor in B2B SaaS companies. He’s had the good fortune of nearly three decades of success as an entrepreneur and enjoys helping others in their journey.

📣 Follow Amith on LinkedIn:
https://linkedin.com/amithnagarajan

Mallory Mejias is passionate about creating opportunities for association professionals to learn, grow, and better serve their members using artificial intelligence. She enjoys blending creativity and innovation to produce fresh, meaningful content for the association space.

📣 Follow Mallory on Linkedin:
https://linkedin.com/mallorymejias

Read the Transcript

🤖 Please note this transcript was generated using (you guessed it) AI, so please excuse any errors 🤖

[00:00:00] Amith: Welcome to the Sidecar Sync Podcast, your home for all things innovation, artificial intelligence, and associations.

[00:00:14] Greetings, everybody, and welcome to the Sidecar Sync Your Home for content at the intersection of all things associations and the exciting world of artificial intelligence. My name is Amith Nagarajan,

[00:00:26] Mallory: and my name is Mallory Mejias.

[00:00:29] Amith: We are your hosts, and as always, we have a really fun episode planned for you guys.

[00:00:34] Uh, this one, it's actually fitting for our format since we are a podcast and we do things primarily via audio. Although if you haven't checked out our YouTube channel, and if you do wanna see us and see demos from time to time, check out YouTube 'cause we're there as well. But since we are an audio delivery format primarily, this is a perfect set of topics today because there's so much happening in the world of audio.

[00:00:55] AI and uh, it's gonna be exciting. It's gonna be a lot of fun. So, Mallory, what's going on? [00:01:00]

[00:01:00] Mallory: You know, not too much for me. Amith. I've had a pretty casual easy morning. I don't think it's been the same for you, everybody. If we have any, uh, listeners in New Orleans and you saw Amme pushing a car in neutral into a parking spot this morning, um, that's what he was doing.

[00:01:17] So am mean. Yeah. Give us a little recap. How was, how's your day been so far?

[00:01:21] Amith: You know, uh, now that I'm sitting in an air conditioned office recording the pod, I am, I am good. Um, it's not quite as crazy hot this morning in early September, just, just after Labor Day is when we're recording this, and, uh, I'm thankful for that, although it's still pretty, pretty, pretty warm.

[00:01:37] Anyway. Um, my wife picked up, uh, a car from the shop, uh, late last week that we're about to give to our 16-year-old. She's about to start driving, which is. Really exciting and also terrifying. Uh, but we got this older car that we have, uh, fixed up a little before and when she brought it back, she's like, oh man, the engine light's on, so can you take it in this week and get that checked out?

[00:01:55] I'm like, oh yeah, sure, I'll do that. You know, sometime later this week. So I decided to take it in today, [00:02:00] and I notice when I get in the car, I'm like, oh, uh, there's no gas in the car. It's on E. So I drive down to this little gas station that's near her house and actually I needed to talk to my wife about something else.

[00:02:09] So I call her up right at the same time. I'm like, oh, I'm getting gas and I'm gonna take it into the dealership later. And, uh, she goes, oh, don't get gas. I'm like, I'm like, well, it's, it's not empty. She's like, no, no. I just, I just was at the gas station. I just, just fill it up. Like, that's really weird. She's like, yeah.

[00:02:23] It was strange. It was like, I tried to fill it up, but it's, it's clearly a full tank. 'cause she's like, unless I don't know how to pump gas anymore, um, this car is full. So I'm like, all right, cool. So like, I. Basically go close the gas tank and get back in the car and start driving to the office, which is only a couple miles away.

[00:02:39] Anyway, long story short is I'm circling the office area, and if you're familiar with New Orleans and Magazine Street, you know, it's a pretty busy, uh, street and, um, I find a little spot to cross the street and punch the accelerator, uh, to get through this little gap. And, uh, right in the middle of the street, the car dies.

[00:02:55] So it just starts, you know, sputtering out. Um. I'm like, oh man, that Sure, [00:03:00] sure. I mean, I wasn't thinking through all of this in real time and I'm like, it sure felt like the car was running outta gas, because that's happened to me a few times in the past in my life, you know? Um, anyway, long story short is I pushed the car outta the way and got a little exercise.

[00:03:12] It turns out the car was in fact empty and, um, there must've been some computer glitch when she tried to fill it up last week. So, but I got some bonus exercise this morning, so I'm all, I'm all good.

[00:03:23] Mallory: That is quite a morning a me. Yeah. He, he shot me a teams message and said, can we push this back 30 minutes?

[00:03:29] I just, my car just died on me a block from here. I had to push it into an illegal spot. I had to reread it because I thought, wait, what happened? But, um, we were just talking about how there's some situations, at least right now, that AI can't always fix. Yep. Though maybe there were some, you know, clever things that you could have done with Claude voice.

[00:03:47] I'm thinking.

[00:03:49] Amith: Maybe, I mean, you know, good old school dials in the car told me the car wasn't empty. Um, and I'm like, oh, it's not empty. And she's like, no, no, the gauge is definitely broken. So I'm like, all [00:04:00] right, well I trust my, trust, my wife a lot more than I trust the old car. So, uh, but yeah, it turns out I think she was just in a hurry.

[00:04:06] 'cause the engine light that she had reported was also a, uh, a low tire pressure light. So anyway, um, all good. I'm gonna go take care of that myself later today when I go back to another gas station and fill up that tire and then that car will be good to go for my teen driver. So, uh, all good. Oh man,

[00:04:23] Mallory: I've gotta say am I've never had a car run out of gas on me, really?

[00:04:27] So I don't know. Yep. Well, yeah, I'm younger than you, but I will say I'm always pretty tuned into that, that gas gauge. Yeah.

[00:04:35] Amith: It's happened to me, I don't know, three or four times in my life probably. So, um. It's a good experience for you. You know, you run outta gas, you're on the interstate, you have to figure out how to get cars safely to the side of the road, and then you have to hike a couple miles to the nearest gas station, get ripped off for a plastic can come back.

[00:04:51] You know, it's, it's part of, it's part of the experience, I think. But yeah, that around, yeah, that sounds like

[00:04:56] Mallory: a blast. I mean, you know, maybe this weekend I might do it on purpose and just kind of see [00:05:00] what that, what that experience is like. I mean, maybe

[00:05:01] Amith: that's an experience we can offer Our, uh, friends coming to digital now this November.

[00:05:05] Um. We can simulate that or something like that. So speaking additional. Now, I promise everybody

[00:05:10] Mallory: listening, we will not make you be in a car that runs outta gas. Trust me, November

[00:05:14] Amith: 2nd through fifth in Chicago, you'll be helping us fill up gas tanks. So,

[00:05:20] Mallory: ah man, please join us. It'll be a great time. Well, as a myth mentioned, we are talking about AI audio today, particularly three.

[00:05:28] Audio AI releases that show how far we've come from the old robotic text to speech. First, we're gonna be talking about Google's realtime translation, then Microsoft's vibe, voice that can generate up to 90 minute podcasts with multiple speakers, more than two, and then finally 11 labs, music that turns prompts into commercial ready songs with vocals.

[00:05:50] And of course, we've got some fun examples to share with you all as well. So first, Google's new real-time voice translate feature is rolling out as of mid 2025, [00:06:00] primarily through Google Meet and the Pixel 10 smartphones. What makes this remarkable is it doesn't just translate words. It uses Gemini AI and Google's latest tensor chip sets to preserve the speaker's tone, rhythm, and emotional expression.

[00:06:15] The technology creates a synthetic voice that closely mirrors the original speaker, so you're not just understanding what they said, but how they set it. The implementation is particularly clever. The original voice remains faintly audible in the background while the translated voice overlays it. This dual layer approach gives listeners context and creates more natural interaction than reading subtitles or waiting for consecutive interpretation.

[00:06:39] Currently it's limited to English and Spanish in beta, but Italian, German, and Portuguese are coming soon. Also, a note on the Pixel 10, Google Smartphone, everything processes on device for privacy, so you can have a phone conversation with someone speaking a different language and hear their translated voice in real time.

[00:06:58] The latency is impressively [00:07:00] low, making actual conversation possible rather than those awkward delays that we're used to with translation apps. For associations, I think this tech could particularly be transformative. Think about international conferences where attendees can network naturally across language barriers.

[00:07:15] Member services could support anyone regardless of language. Board meetings with international chapters could become seamless educational content reaches global audiences. Without dubbing or subtitles, this tech potentially has the opportunity to make every association global. So Amme, what are your initial thoughts on real-time translation from Google?

[00:07:39] Amith: I find it extremely exciting and I think your point about the association market, uh, being at the center of opportunity for this technology is spot on. Associations fundamentally are about connecting people. And so if you think about how we do that and how we scale that there's a wide range of languages and dialects and voices that are out [00:08:00] there and what we want.

[00:08:01] Is to connect people in an original and authentic way. And sure in-person events could be affected by this. I think associations could also offer a trusted way to connect online through various type types of technology that, uh, uses the association as a conduit, leverages technology like this and offers people from totally different backgrounds to be able to communicate in a native way.

[00:08:24] Uh, I find that extremely exciting. So to me, uh, what you mentioned about the on device piece of the equation is really also worth noting. We've talked a lot on the sidecar sync about the progression of models becoming smarter and also becoming smaller. So the smaller parts, what I wanna focus on for a second, every six to 12 months, you have not only a doubling in AI power, but you have.

[00:08:51] Roughly having, or many cases, it's more like 80, 90% reduction in cost. And along with that, part of what's happening is the models are getting [00:09:00] smaller and smaller. They're getting compressed. Now on the frontier, the most powerful models, the most intelligent models are staying the same size or even growing in terms of size.

[00:09:09] The models at a given size are dramatically more intelligent. So it's kinda like, you know, pound for pound, the strength level of the model is so much higher, right? If you think of it in, in those kinds of terms. So, um, in order to run portions of this technology on the Google Pixel device, um, you had to make the model a lot faster and a lot smaller.

[00:09:30] And that's important. Uh, the other thing we gotta remember is smartphones are not these, you know, dumb devices. They're actually pretty sophisticated computers. The computing power, the memory, the bandwidth capabilities through 5G. These are incredible computing devices we have in our hands. And so being able to put a little bit of the processing, it's not only good in terms of latency and privacy, it's just a smart way to take advantage of the compute power that's out there in the world.

[00:09:55] So I think it's a great example of what's to come. Um, I think that. I shouldn't [00:10:00] say, I think that I am hopeful that Apple, uh, and everyone else that's in the audio and voice game will get into this. I hope apps like WhatsApp will jump into this, that that's an app that connects, you know, I think a billion plus people on a daily basis through text and audio, uh, all over the world.

[00:10:16] So I find it incredibly exciting.

[00:10:20] Mallory: It seems like we had already seen big strides in translations and in dubbing through tools like Hagen. So is the real innovation here, Ames, the real time component, the fact that it can happen basically as it's happening.

[00:10:35] Amith: So from my point of view, the real-time nature of this is definitely the big step change and Hagen, uh, is a great kind of analog to compare this to.

[00:10:45] So Hagen is a tool that we've been using and still use actively. Uh, we've had it, uh, in part of our product portfolio, technology portfolio for a couple years. We've been demoing it, uh, we've shared it on the podcast, but essentially allows you to take a video. Of any individual [00:11:00] speaking. And that video can be uploaded to hey gen servers, uh, through their website and it will convert it into a number of different formats.

[00:11:07] Uh, one of the format changes it can do is, is, uh, language to language. So I have a video of me speaking Japanese and Spanish and Czech, and I don't speak any of those languages. So, uh, it's incredibly powerful and really cool, but it's batch. Based meaning you have to upload a video or a series of videos, and sometimes you have to wait half an hour or an hour to get your video back.

[00:11:28] Uh, so it's really useful for learning content. Um, we use it a lot for translation of learning content. Um, it's useful for a lot of things that where asynchronous, you know, communication is fine. Uh, but for synchronous real-time communication, the technology hasn't yet. Been available and now it's, and so that's the kind of pace we're on with AI because people were blown away by, Hey, gen, probably people are blown away by it today who haven't seen it yet, but certainly two years ago it was absolutely state of the art.

[00:11:58] He didn't mind waiting 15 minutes [00:12:00] for a 32nd or 62nd video to come back to you, uh, translated. Um, but now with this capability, you can do pretty impressive real time translation. Now keep in mind, hey Jen, it is video. Audio, and this is just audio. Um, the audio part's super important because that's kind of our primary modality, uh, of, you know, people in general as we communicate through verbal language.

[00:12:22] And, uh, so we're really good at picking up on any subtleties, anything that's off slightly, it just feels strange. So the bar is quite high. Uh, but that's also why I like what you had said about their approach to implementing it, where the original voice is still maintained. You can kind of hear it. Um, it's almost as if a translator's speaking, but it, it sounds like the person's there.

[00:12:41] I suspect that will eventually go away. Uh, but I think people are just gonna take a minute to get used to this and try it out. I haven't tried it out myself yet and I'm looking forward to it

[00:12:51] Mallory: and Right. I think it's still in beta now, so I'm not even sure if you can try it out unless you're kind of a part of that small group right now.

[00:12:58] But again, it'll be on the [00:13:00] Pixel 10 smartphones and Google Meet. I'm assuming also Microsoft will follow suit, or I'm hoping kind of we see some other major companies enter this game, but Amit, when it's more widely available, if you were an association leader, where does your mind immediately go in terms of a, a near term use case?

[00:13:18] Amith: Well, the first thing this should light turn on as a light bulb for you is, uh, the ability to translate all of your existing content. So, forgetting about the realtime use case for a second and just say, Hey, um, as a fundamental technology ability. We now have the ability to do translation of audio with high level of accuracy, really, really close to perfect in the speaker's natural voice.

[00:13:40] And so if you have an archive of fairly popular pieces of content, let's say there are audio proceedings from conferences or a. Maybe you have a podcast of your own and he'd like to make it available in multiple languages. The the bar is so low, it's something you can do. And I would be surprised if the various platforms [00:14:00] like learning management systems and podcasting platforms don't just have options that you can check boxes on to say, Hey, publish this in X, Y, and Z languages.

[00:14:08] So. You should be thinking about that strategically though, because you might be the American Association of blah, blah, blah, whatever that industry is. Um, but that doesn't mean that's what your audience is or necessarily needs to be. So if your limitation is geographic. From a historical perspective that makes sense.

[00:14:26] And you certainly wanna serve the geography that is home base, but your content probably can be leveraged far beyond your borders. Uh, I think this is also true, uh, perhaps even more so as an opportunity for associations that are. Not native and not located natively English speaking countries. So, um, there's a lot of great content and a lot of associations doing great work that are outside of the United States, Canada, Australia, uk, et cetera.

[00:14:51] So, um, you know, outside of kind of the primarily English speaking world, um, there are people who would prefer to publish in their own language and now they can do that and they can still get an [00:15:00] English version that's, you know, just as good or maybe even better in some ways.

[00:15:04] Mallory: Do you think there's any reason an association should not attempt to be more globalized?

[00:15:10] In terms of its reach.

[00:15:12] Amith: I mean, if you're the association of, you know, kind of anti-globalization or something like that, that's fine. Or you know, maybe if you just prefer to not have as much opportunities as possible. I mean there probably are some legitimate reasons for that. Um, but I think the primary reason historically has been if your association just hasn't had the resources to do this, you have a hard enough time keeping up with your English language content and making that good.

[00:15:36] Uh. This just puts another opportunity on your radar. Now, I'm not suggesting to you that this becomes the number one priority immediately. It's just ar. It's an opportunity that needs to be cataloged in your brain as something you can do with ai. And so when you're thinking about your strategy and engagement, maybe.

[00:15:52] The global market makes sense. Now, it might not because from a focus perspective, um, you might have your hands full and there might be so much opportunity [00:16:00] in your local market. I'm simply saying that you shouldn't intentionally kind of push away an audience outside of your locale unless there's like a really good reason for that.

[00:16:10] And there, there might be, there might be national security issues for some associations or things like that. So I'm sure there's some legitimate uh, reasons. But I think generally for most associations that are. They're there to promote a particular sector or improve the quality of the services in a particular profession, or in some cases, their enthusiast organizations.

[00:16:29] Just try to bring people together who enjoy a particular sport or hobby or something like that. Well, I mean, that certainly makes sense, right? To bring people together from the global community.

[00:16:39] Mallory: Absolutely. And thinking about its impact on associations. As they exist now in global reach, but also the downstream effects on your members.

[00:16:49] As you all know, if you listen to the podcast before, my husband works in healthcare and I often hear really the terrible stories of like these translators that they have to call in that are doing the best that they can. Um, the [00:17:00] lags kind of information that gets misconstrued, and also the discomfort of patients having to kind of, when they're in an maybe an urgent moment.

[00:17:08] Where they need care immediately, especially like in emergency rooms or urgent cares, having to express themselves and then wait for a translator. So I think real time translation in many industries and many professions will really change things for the better.

[00:17:23] Amith: I agree. I think it's an exciting thing to have available and, uh, in associations or conduits for connecting people.

[00:17:29] So there are opportunities here. And think of this, you know, a lot of times what we do in the pod is we'll cover a general purpose technology and we'll say, look, this is a capability, and this might've been the other way to look at it, is, um, previously the resourcing might've. Been, you know, extremely expensive, right?

[00:17:44] To have a voiceover actor to do a translation for you would be hundreds of dollars per hour or something like that, right? And you're not gonna go do that for all of your content because the constraints are cost and time. And also, you know, your content tends to get stale pretty quickly in most [00:18:00] industries.

[00:18:00] You know, after a few years, the content, you know, doesn't really have much of a shelf life remaining. So it doesn't make sense to necessarily invest a ton of money or time in that. But if you can have it essentially for free. Then it makes sense to consider that that's a really important piece. Now, your overall content strategy is something you've gotta think about deeply because a big part of the problem isn't a lack of content.

[00:18:21] Your people, your members, and your audience aren't typically telling your association, man, you guys just don't have enough stuff on your website. It's just so sparse. There's hardly anything there. I can't, you know, I just run out, I run out of things to browse. You know, that's, that is not what your members generally tell you.

[00:18:36] Most of 'em are saying they're having a really hard time finding what they need. They probably know that it's somewhere deep inside there. Um, so you might be saying, well, but now if I add like multiple language versions of everything, it's gonna make it even more complex. Uh. And that is true if you maintain the current mindset around how you organize your content and how you deliver content.

[00:18:54] That's also where AI can help, uh, through things like personalization, where you actually deliver stuff to people [00:19:00] proactively. It's based on what they are likely to want, which is again, very much democratized these days. You can do personalization at scale across all of your asset categories. In a very affordable way nowadays.

[00:19:10] Um, there's great tools for that. And of course there are knowledge products out there that can help you very easily answer any question against a broad corpus of data and quickly become the greatest expert on your content. So there, there are AI solutions out there, um, that can help solve that problem.

[00:19:26] So if you have that constraint and cons as a concern, which is the consumability of multiple languages of content, I, I also think you have to, you have to look at that side of it, which is how do you manage this stuff and how do you deliver it?

[00:19:38] Mallory: I want to move to Microsoft Vibe Voice. Uh, Microsoft just released Five Voice as open source, and we think it's a game changer for audio content creation.

[00:19:48] It can generate up to 90 minutes of expressive multi-speaker conversational audio from Plain Tech. So four up to four distinct speakers, having natural conversations with [00:20:00] realistic turn taking, emotional delivery, and even emergent properties like background music appearing when contextually appropriate.

[00:20:07] The technical achievement here is staggering. So previous systems could barely manage short outputs or even two speakers before losing consistency Vibe voice uses continuous speech. Tokenizes combining a large language model for managing dialogue context with a diffusion decoder for high fidelity acoustic output.

[00:20:26] The 1.5 billion parameter model runs on consumer GPUs, meaning you could run this entirely locally. What's particularly impressive is the conversational realism. The system doesn't just alternate speakers reading lines. It creates natural interruptions, overlapping speech, laughter, and emotional reactions.

[00:20:45] It's optimized for English and Chinese right now supports emotion control and even has basic singing synthesis as an emergent property. Because it's MIT licensed, you can use it freely for research and commercial purposes. For associations. We've [00:21:00] kind of talked about some of these already, but you could turn a research report into an engaging podcast discussion automatically.

[00:21:07] Board meeting minutes could become accessible. Audio summaries. Educational content, could it be instantly available in multiple formats? You could create personalized audio briefings for different member segments, all generated from text. So I did a little dabbling with this. I tried to use the demo myself, uh, in my browser.

[00:21:26] It was a little bit buggy, so I ended up pulling an example from the Microsoft announcement. We're gonna play that here now.

[00:21:34] Demo: Welcome to Tech Forward, the show that unpacks the biggest stories and technology. I'm your host, Alice, and today we are diving into one of the most anticipated and frankly. Most chaotic tech launches of the year Open AI's, GPT Pie.

[00:21:49] The hype was immense. The teasers and leaks building for weeks. On August 7th, it finally dropped promising a new era of artificial intelligence to help us [00:22:00] make sense of it all. We have two fantastic guests, Andrew, a senior AI industry analyst who has been tracking this launch closely. Welcome, Andrew.

[00:22:09] Great to be here, Alice. It's certainly been an eventful launch and we also have Frank, a tech enthusiast and a super user who has been deep in the community forums.

[00:22:19] Mallory: So as you can see, we hear those multiple speakers. In my opinion, the delivery is pretty natural. I still think Notebook LM is maybe just a bit more natural in terms of audio delivery.

[00:22:30] Um, Amit, what is your take on the open source Microsoft vibe voice?

[00:22:37] Amith: I would say the trend line continues to show that with a very small model, just like we talked about in the last topic, you know, smaller models running on device, being able to do pretty impressive stuff. I mean, notebook, lm I would agree with you probably has a little bit of a leg up right now in terms of the overall quality, although their length is not nearly as right.

[00:22:57] Um. As long as I think, I think they go to like eight or [00:23:00] 15 minutes maybe, but 90 minutes. That's a lot. I mean, that's like a Lex Friedman podcast or something, right? It's a very, very long format. Um, I don't think a lot of people want, you know, unless you're, unless you're dealing with a very specialty podcast like the acquired podcast or something like that, most people don't have the ability to pay attention for three, four hours or whatever.

[00:23:17] So I think that's not necessarily the biggest, uh, distinction. But being able to run it locally means a lot of things. One is the cost is extremely low, uh, close to free, essentially. And it also means you can make changes to it over time, so you can tie it into different systems you have. Um, there's a lot of cool stuff you can do with this.

[00:23:36] So one of the things that I'd point out too is that you can think of this as another tool in your tool belt. Just like real trans realtime translation. You should be thinking, okay. What benefit could I essentially provide in my value delivery to my audience through automatic summarization of content through audio, right?

[00:23:57] And I like, I like the idea of the deeply [00:24:00] personalized versions that you mentioned. So take for example, a newsletter where you say, oh, this is a newsletter of content picked out just for Mallory. Wouldn't it be cool if there was a little button in there that also said, Hey, listen to an audio summary of today's newsletter, and it's today's newsletter summary for Mallory, right?

[00:24:18] Mm-hmm. That is something that you can do with vibe, voice, and really exciting. Um, you guys who are listening probably know that one of the companies in our family is raso.io, which does personalized newsletters and. You can bet that that is on the ra, uh, the radar for a future, uh, opportunity for, for that company.

[00:24:33] Uh, it's just an obvious thing to be able to say, Hey, wouldn't you love to be able to listen to your newsletter? Um, and there's so many other opportunities like that because naturally, you know, we do like to listen to a lot of things. There's lots of situations where audio is just more natural than reading.

[00:24:47] And, uh, depending on the individual, sometimes they prefer audio over text even when they can't sit down and read. A lot of people listen to audio books, for example. So. I find this really exciting, the democratization of models, open [00:25:00] source being the key way that's done means they're available for free, they're available to download, they're really easy to run.

[00:25:06] These things, even I would say a savvy consumer and non-technical person can download tools like Notebook, uh, sorry, not notebook, L lm, LM Studio, all these, all these, these terms are jumbled up even in my head and I'm thinking about this stuff all the time. Um, so LM Studio is a tool that I run on the Mac where you can run a bunch of local models.

[00:25:26] Um, and there's, there's lots of ways to do that. So I find it really exciting. I think, I think associations need to put on their creative cap here and say, what can we do with this? It's kinda like the first time you were staring a chat GPT and it was a blank screen. You're like, what do I type into this?

[00:25:41] This is weird. This is interesting. I think it's the same thing. You have to go and experiment with this stuff.

[00:25:47] Mallory: Mm-hmm. As you were talking, Amit, I just had a thought too, you know how we do prediction episodes at the end of every year, thinking that probably next year I would predict that we're gonna see maybe a 90 minute [00:26:00] podcast that's totally AI generated rank in like the top 10 or 20 or 50 podcasts, which will be pretty insane.

[00:26:07] Um, to think about how an AI generated podcast will likely be making a ton of money in the near term. Yeah.

[00:26:14] Amith: It's interesting because, um, like the, the thing that we're doing with Sidecar, with the learning Hub that, uh, many of you're familiar with, we have all this great content around AI specifically for associations, and we built this tool that many of you have heard of called Learning Content Agent.

[00:26:31] We actually had a, a previous episode of the Sidecar Sync where Jason and Mallory had, uh, talked about that in some detail. We'll link to that in the show notes. But, um, the learning content agent is essentially a way of producing content at scale that can change continuously, uh, which means it never gets stale.

[00:26:47] It's super important for our content. So it's about ai, so it's changing literally every day. Uh, but it also allows us to have dimensionality, meaning that we can have the same content presented. Uh, to multiple different audiences with different context. [00:27:00] Uh, so through a variety of settings and configuration, we can create, uh, learning content for using our AI learning content as a baseline for any industry.

[00:27:09] And so we're partnering with associations to create what we call learning hub for members, which are co-branded websites that the association puts out there and it's got content. AI for x, X being your profession, whether that's architecture or engineering, or branch of medicine or accounting or whatever the case may be.

[00:27:27] And the value there is that you're delivering not generic AI education, it's AI education specifically tailored to your audience. And we've built an entire technology pipeline that's very heavy on ai, obviously to make that possible for us to do with really just some clicks and, and obviously there's some work to set it up correctly and have some layer that's industry specific, but, um.

[00:27:48] The, the analog to this is, you know, in the world of, uh, podcasting, you have great content coming from humans across a wide variety of different podcasts. What if you create, [00:28:00] could create a version of the sidecar sync for a bunch of other industries as well? Right? And, you know, you and I can't do that ourselves, but, uh, it would be kind of cool to be able to do a version of this that's targeted to a whole bunch of other sectors as we partner with associations.

[00:28:13] Uh, to deliver a learning content. So, um, I think that podcast that's fully synthesized by AI is, it's not only gonna be a thing. I think that in some cases it'll be better because it'll be deeply personalized. You can create a podcast just for one person.

[00:28:27] Mallory: We could do a Mallory and Amit podcast all when you break down in your car and it runs out of gas, telling you exactly what to do next.

[00:28:34] Amith: There you go.

[00:28:35] Mallory: Um, I wanted to ask you, AMI, I know we talk about a lot, kind of, there are situations where human connection and relationships is of the utmost importance, and there are situations where you can lean on AI generated video and audio. Talking about the AI Learning Hub specifically that you mentioned with sidecar, what has been the reception on the AI generated video and voice?

[00:28:58] Because I imagine we have some people [00:29:00] listening thinking, well, it, it's hard to decide when to use one versus the other.

[00:29:05] Amith: So if you, if you've gone through our learning hub content, but if you haven't taken a course with us in the last couple months. Jump back on and, and check it out. Um, all the content now, as Mallory mentioned, is AI avatars and AI voices.

[00:29:18] All the content is still created by us, but we're constantly modifying it. And then every time we modify it, we press a button and then a few hours later, the course is updated in our LMS automatically. Uh, that's. Pretty magical actually. So to be able to do that, you, you can't really have people re-record mm-hmm.

[00:29:34] Stuff all the time. It just doesn't scale. Um, but if you listen to or watched our content on the Learning Hub prior to Iran, I think it was maybe at the end of May when we switched over, um, you would've, uh, heard a lot of Mallory's voice, a lot of my voice and several other, uh, folks across our organization.

[00:29:50] Um, and we replaced that with AI avatars. So to get back to your question, uh, the feedback has been consistently very, very positive. Um, there have been a few people who've said, Hey, I [00:30:00] prefer the, the, the humans. And, and I get that. Mm-hmm. Um, and I also kinda like that a little bit since, uh, we recorded the last version, a little ego

[00:30:06] Mallory: stroke.

[00:30:07] Amith: Well, a little bit, but, uh, you know, but, but the reality is, is most people. Are, we have to remind them that it's an ai. So we are a big believer in the idea of transparency in AI use, essentially disclosing that you use ai. So this is not an AI po generated podcast at the moment, but if it was, we would definitely tell you that at the beginning and reinforced that.

[00:30:27] And we do that with all of our learning content. At the beginning of each course, the instructor introduces themselves as an AI avatar and there's actually a section, uh, in the learning hub that explains the whole idea and, and why we do that. The feedback's been real positive and a lot of people have said, oh yeah, I forgot that was an ai.

[00:30:41] So to me that's the best compliment of AI is that it, it just didn't, really wasn't factor, it didn't, it didn't bother them. It didn't necessarily make it better either. It's just, it was essentially on par with the typical human instructor.

[00:30:54] Mallory: Mm-hmm. And I think, too, circling back to my own question, it's really important to.

[00:30:59] Keep in [00:31:00] focus the outcome of your educational content. So you want your learner to walk away with the most recent cutting edge knowledge and retain it. And I do feel like with AI avatars, as Amit said, educational content goes stale pretty darn quickly in the world of ai, like by the hour, by the day. So being able to update that content on the fly means our learners are learning more.

[00:31:23] Um. Than they could prior. So even though it is an ego stroke, and I do enjoy delivering educational content myself, Amit, I know you do too. Knowing that our learners are walking away with the most recent information makes me feel good.

[00:31:37] Amith: I was talking to Jason the other day. Jason's in charge of our learning, uh, curriculum at Sidecar, and he and I were chatting about the opportunity for synchronous education delivery to compliment async.

[00:31:47] So the online asynchronous format, of course, scales really nicely. You can have effectively, an unlimited number of people take that at any time of the day and Eddie Pace that they'd like, which is wonderful. Um, but there's something [00:32:00] about. Um, instructor led, um, synchronous cohort based learning. That's really interesting and really powerful.

[00:32:07] I think it's the group of people you're with. I think it's the live interaction, uh, if you're face to face, like in actual rooms sometimes. I think that's really powerful. Uh, so, uh, we're looking at ways to introduce additional, uh, it's all AI learning, that's what we're focused on, of course, but synchronous learning opportunities, whether they're workshops from time to time that are live sometimes in person, but things that can kind of level up the experience, right?

[00:32:32] Mm-hmm. So the a IP and the core curriculum we have online will be our foundational layer, but because of the ai, we have the bandwidth. To take on those kinds of ideas. And I think the same thing could be said for many associations is if you can take a lot of the current work that you do in your education department or in your conferences department or whoever's responsible for this stuff, and automate a large portion of it, not only can your content be better and more up better because it's more up to date, [00:33:00] uh, and you can potentially go in multiple languages as we're talking about.

[00:33:03] But also can free up your staff to potentially do those really high touch human connection things that Mallory, you introduced the segment with. I think that that connectivity, that person to person relationship building, that is what associations are all about. That's what people get energy from. You know, they go to an annual conference like ASAs annual meeting or digital now or whatever, and they come back going, man, that was awesome.

[00:33:24] It was really great. Yes, I learned a lot. Um, but I also just, I had a great time, which is of course very important and I. I met people, I, I had a different perspective coming back from that, and you, you can't get that from any form of ai, from my point of view.

[00:33:39] Mallory: Mm-hmm. There's an energy, hard to put into words exactly as something tangible, but an energy when you come back from an in-person event where you think, Ooh, like, what will I do next?

[00:33:47] For sure.

[00:33:48] Amith: Yep. It's usually positive.

[00:33:51] Mallory: Unless you're doing a simulated car breakdown, like we're doing it digital now this year, but okay. Moving on to topic [00:34:00] three. This will be a fun one. 11 Labs is no stranger to fantastic AI generated audio. We use 11 labs. AI Learning hub that we were just talking about, but they're entering the music scene with AI that generates complete studio quality songs with both vocals and instruments from text descriptions with all rights cleared for commercial use.

[00:34:22] We're talking about professional grade tracks with expressive vocals and multiple languages, complex arrangements and solid production quality. What sets 11 labs apart from competitors like Suno, which we have covered on the pod before, is their approach to copyright. So they partnered with Merlin Network and Cobalt Music Group to ensure their model is trained on license data.

[00:34:43] Every output is legally clear for commercial use. No copyright disputes, no surprise lawsuits. Users can specify genre, mood, instrumentation, structure, and lyrics. The platform handles everything from AL scores to pop songs in English, Spanish, German, and [00:35:00] Japanese. The platform also offers section based editing, so you can adjust individual parts of a song after generation change the chorus lyrics modify the bridge arrangement, or alter the overall sound.

[00:35:12] The AI vocals are pretty realistic and expressive as you'll hear. So if you need a custom anthem for your annual conference or background music for educational videos, maybe a jingle for your new AI generated podcast, mood music for your gala, 11 labs, music can help. So I want to share a, a fun example that I generated with.

[00:35:32] 11 labs music, especially when I'm testing out AI tools because I guess I'm such a frequent user, I tend to go for like some of the hardest use cases. So for this one, I asked it to generate a song about, you know, it AI for associations in the style of a New Orleans bounce music song, which is like a very particular style of music.

[00:35:54] If you all haven't heard it, go to YouTube, big Frida's a, a big New Orleans bounce artist here. Or [00:36:00] not here. I'm not in New Orleans anymore, but in New Orleans. And so I asked it in that style, which I didn't think it would do a great job with. I'm gonna play this song for you and then we'll talk about it after.

[00:36:12] Demo: Drop everybody. Bounce a y in the house. Let go.

[00:36:22] Yo, listen up. Associations take no clue. Take a note. Be the a y, awake. We gotta stay afloat. Crew, they afloat. Hey, the driven moves. That's how we compete. Compete. Serve your members better. Let that knowledge speak.

[00:36:42] Learn fly, shake it up, smart. Get your strategy right. Knowledge is power. Win the on with.

[00:36:55] Mallory: So as you can see, it was a fun song for sure. Personally, I don't [00:37:00] think it's in the style of New Orleans bounce music. I don't know if there was like any training data available on that style, but it was really fun. Honestly, it sounds like a, a real song, right? It doesn't sound like it was AI generated, in my opinion.

[00:37:12] There were a couple of little mistakes there. For example, in the text of the lyrics there, um, it said crew and then like, I guess the crew was like the background group of the song that was supposed to be saying something. But actually the song was saying Crew. Crew, which you might have heard over and over again.

[00:37:30] I think it was good overall. Amit, what was your take on this, uh, fun song from 11 labs.

[00:37:36] Amith: My, my point of view is, uh, that this is, I think this is something once again that people can weave into their thinking on what they're capable of doing. Mm-hmm. You might not have thought about multiple languages. You might not thought that you could do real time translation and that you might not have thought that custom music was in your budget or something that your association could potentially tackle and weave into the [00:38:00] production quality for your upcoming event or for.

[00:38:03] Any other use cases that Mallory mentioned. And while this may not be, you know, the Grammy award-winning type of thing at that level, but it, it's something that's quite high in the utility scale for business. So I think it's really exciting. Uh. I'm not an artist myself, so I don't really have a great ear for music, so I can't really comment on anything beyond that.

[00:38:23] But I do think that, um, for many use cases, the quality of the, uh, outputs from these types of models are exceptional and it's truly stunning to see where we're at now relative to even a couple of years ago. Um, you know, the, the thing about, uh, the copyright stuff you started to mention I think is worth digging into.

[00:38:41] In some more detail. That's exciting. I think first of all, associations tend to be on the conservative side of things when it comes to ensuring that they're, they're doing the right thing, which is good, uh, but also that they're on the right side of, of not having their content. Being misappropriated is a big concern for many associations.

[00:38:58] Most associations are [00:39:00] not, uh, in the business of music, but, uh, they can appreciate the fact that they don't want, you know, artists who've produced music and, you know, music labels don't want their content being misappropriated in any way. Now the conversation about Suno. Or, um, you know, in the case of, uh, OpenAI and the New York Times, which is a similar lawsuit in the context of text, uh, as opposed to audio, um, you know, you have, there are different opinions about that.

[00:39:25] Like is it really copyright infringement if a model is trained on something or is it not? And the courts are going to give us insight on that in the coming years, and I think that's very interesting. But I, I do think that licensing deals like this are really smart. They're gonna keep happening for the bigger players.

[00:39:40] They're gonna have the checkbooks where they can afford to bring in those kinds of partners on the front end. Um, one other thing I'd say is it's kind of like, you know, Napster may have been the original way that millions of people first got exposed to digital music. Um, and then Spotify came along and that was kind of like the [00:40:00] legit way of doing it.

[00:40:00] That became a big business. And so I think. Maybe Suno is the Napster of, uh, I'm not suggesting they are by the way, but maybe they're kind of at that phase 'cause they came really early and that 11 labs, you know, sees an opportunity here to do it a little bit differently and come up with maybe a more mature product.

[00:40:16] And, you know, what they've been doing for years with audio, I guess is not all that different modality wise from singing and music. So it makes a ton of sense. It's a really good leverage point for them.

[00:40:26] Mallory: I wanted to dive into to Suno versus 11 Labs. I think there's another company called uio, uh, that was being sued for copyright infringement.

[00:40:36] To me, there's a sense of peace that comes from using a tool like 11 Labs Music, where, you know, whatever you use at your conferences, at your galas is gonna be safe and protected. Um, but I can also understand when you're trying to be an early mover and these tools, these applications are popping up and you're interested in using them.

[00:40:53] Amit, do you think. And it's such an unclear landscape, I, and nobody has the answer, but do you think it's best [00:41:00] for associations to kind of wait and see, um, when they get more clarity from like the training data that these companies are using before using something like that? I know in our Intro to AI webinar that we hosted sidecar, pretty much every single one we would get asked questions about like.

[00:41:16] Are we allowed to use this information from chat GBT that we created? Can we use these images from Midjourney? Can we use this music from Suno? So what, what are your thoughts on that?

[00:41:25] Amith: Yeah, I think, I think the side of like the outputs you're using with AI models, there's a lot of clarity that's come down over the last couple years that you're probably in pretty good shape, uh, in, in most cases.

[00:41:36] So I think there's less to be concerned there. I think that the biggest concern people seem to have right now is that when they put their stuff into a model, what happens to it. So if you, for example, say, Hey, I'm gonna allow chat PT to connect to my Google Drive because I'd love to enable the feature where chat PT can be smart enough to understand.

[00:41:55] All of my Google docs, right? And they, they rolled that out, uh, I think a couple weeks ago with GPT five, [00:42:00] uh, or maybe they had that a little bit before then. But a number of these other companies have that. There's also MCP, which we've covered on the pod before, which allows you to connect AI systems to a variety of other databases and, and file systems and so forth.

[00:42:11] And so a lot of people are wondering, well, what does that mean actually? Does that mean that OpenAI and philanthropic and Google have the right to use my content for model training? The short answer is any legitimate company in the space, which is not necessarily a very long list, but any legitimate company in the space that's providing models, um, is going to give you a pretty strong license agreement that clearly articulates that they're not going to use your content for model training.

[00:42:36] Um. It all boils down to how much you trust them. So it's kind of like cloud storage providers. Um, Mallory, do you use Dropbox or Box or Google Drive for like, for personal Google Drive, typically? Yeah. Yeah, yeah. So, and, and a lot of people use Google Drive. A ton of people use Dropbox and people will put like super sensitive stuff on there, right?

[00:42:55] Um, financial data and like personal records and all this stuff. Medical [00:43:00] records, right? People will put that in their, in their Google Drive or their Dropbox without thinking about it. And, um. While that, you know, those drives are, are quite secure, they're encrypted in transit, encrypted at rest, uh, there's a lot of things that, uh, these cloud storage providers do that are really good safeguards.

[00:43:18] Ultimately, it is possible for Google and Dropbox and Microsoft, these other cloud storage providers to actually read your content. It is possible for Google to read anything that you have in Google Drive. Uh, yet you store a ton of sensitive data in Google Drive. I use a ton of different. Cloud storage myself.

[00:43:36] Um, and so if you trust a vendor. For cloud storage, you probably should be willing to trust them for model inference is my point of view. Because in order to do something that is not allowed of them with your data, they'd have to violate the license agreement. It doesn't mean it wouldn't happen. I'm not saying that you're, there's no such thing as a, a bulletproof way of dealing with this.

[00:43:58] Um. But it is [00:44:00] definitely something that's reasonable to seem that if you go with a fly by night model provider or you know, some random inference provider that pops up out of the blue that has some cool model on their website and you allow them to connect to your data, well, you know, that might be a little bit sketchier.

[00:44:15] So I'm not suggesting also that. The bigger the company, the better. So OpenAI is probably the largest player in the AI space right now. Right? Uh, I'm not saying to you that they're more trustworthy than Anthropic or, or Google. Uh, I don't have an opinion on that, that I really would share publicly. I, I have my views on each of these companies, but I think that all the bigger players in this space have license agreements that say what you need to hear.

[00:44:38] Uh, and then it boils down to what do you think of the company and do you, do you want to entrust them? It's some of the same decision making you, you'd make if like. Oh, am I gonna host my stuff on Amazon or on Microsoft or on Google Cloud? It's the same kind of thinking here. Um, so. That's the main way I'd put it.

[00:44:55] And that's the number one concern coming back to copyright that people have is protecting their own stuff. Mm-hmm. [00:45:00] Um, they are concerned with the outputs they generate not getting them in trouble. They want indemnity. They wanna have assurance that they're not gonna have problems. And then most associations do care.

[00:45:08] That the models are not infringing on someone else because Right. You know, they don't wanna be bad actors. And, um, you know, it's kinda like, I think Adobe did a really good job with this over the last couple years at the Adobe Creative Suite. They introduced generative AI very early on, but they built their own models and they had licensing deals, uh, with imaging companies and other companies that provided them the, the content for training.

[00:45:29] They, they made a big deal outta that. They said, look. Mm-hmm. Their, their customers are creative professionals who create these things. The last thing they wanna do is say, Hey, uh, we've ripped you off and we're gonna use that stuff to automate your own jobs. That's, you know, they, they took a very strong stance on that, uh, right away.

[00:45:44] And I think, I think a lot of companies are starting to do that. So, and if you think about the amount of money being spent on ai, a, a proper licensing deal is really not an unreasonable cost considering how much they're spending on everything else.

[00:45:56] Mallory: Mm-hmm. And I, I will say, I mean, if it comes [00:46:00] down to it for me to use Suno versus 11 Labs, knowing what we know about these licensing deals, I'll, I'll use 11 labs music.

[00:46:05] And I think that is a way to retain, I don't know whether it's moral or ethical or some sense of control over the AI that you're using, at least in my opinion.

[00:46:16] Amith: Yeah, I feel the same way. I mean, I, I view myself as a creator as well. Not a creative, but like a creator of content and software and things like that.

[00:46:23] And I, I think that my intellectual property should be respected by others. So I certainly wanna respect their intellectual property. And I think Mo most people in general feel that way, at least in the western world, and most, uh, associations that I know feel that way as well. So I think it's a good topic to, to always, uh, have front and center as you're evaluating these tools.

[00:46:43] Mallory: Amit and listeners, can you hear my dog whining in the background? I'm curious.

[00:46:47] Amith: No, I can't.

[00:46:48] Mallory: Okay. She's in her crate and I think she's, she's signaling to me that it's probably time that we wrap up the episode. Yeah. Good timing. Oh, and she just barked, so she might, she might know that we're talking about her, but Amit, we talked about a lot of AI [00:47:00] audio today, which we love talking about on the podcast.

[00:47:03] Do you feel like all of our listeners should be. At least considering ways that they can integrate stuff like this into their organizations.

[00:47:12] Amith: A hundred percent. If you're not experimenting with these things, you are missing out on, not necessarily even the use of that particular thing, but you're missing out on the reps.

[00:47:22] So if you wanna be good at ai, you need to put in the reps, just like in anything else in life. If you want to be good at it, you've gotta put in the work, put in the reps. You gotta do it every day. And the more diversity of your training, if you think of it that way, the better. So if you're like, oh, I'm all about ai, I'm using chat PT all the time.

[00:47:39] That's cool. That's kinda like going out and working out one arm at the gym over and over and over again. Right? You need to kind of mix it up, do different exercises, try different things. Um, it's good to get some variety. And so what it does is if you go experiment with 11 labs music and you're like, ah, I don't know, I'm not gonna use music in my association, just kind of.

[00:47:58] Thinking, well go do it anyway. [00:48:00] Maybe it'll just be fun, right? Share it with your kids, share it with your colleagues. Do something silly. Um, that's not bad for you, first of all. And you'll probably get some aspect of your neural network lit up that would not have otherwise been activated, and you would learn something and have some connection drawn that could be helpful.

[00:48:15] So I would, I would encourage everyone to go experiment with this stuff. Try, you know, different things. You know, be silly with it too. It's not just about like, what's the use case that I can directly take and put into production? This is all new technology. Even those of us that are spending most of our waking hours in ai, um, we don't have all the answers.

[00:48:32] We're coming up with this script as as we go.

[00:48:36] Mallory: You heard it here. Everyone keep building your AI muscles. Thank you all for tuning into the Sidecar Sync as one of your AI gyms that you can attend, and we will see you all next week.

[00:48:48] Amith: Thanks for tuning into the Sidecar Sync Podcast. If you want to dive deeper into anything mentioned in this episode, please check out the links in our show notes, and if you're looking for.

[00:48:58] In depth AI [00:49:00] education for you, your entire team or your members head to sidecar ai.