[0:04] **Colin Heilbut - Rime:** Well, Well, Colin, thanks for joining. [0:09] **Colin Heilbut - Rime:** Think you met everybody on the call. You need? Just met RAs, which [0:09] **Colin Heilbut - Rime:** is great. [0:13] **Colin Heilbut - Rime:** Purpose of today is to give you a chance to [0:18] **Colin Heilbut - Rime:** share your skills, talk about how to pretend the pitch to us, [0:21] **Colin Heilbut - Rime:** There's no way to answer. That's what we've kind of talked about. [0:25] **Colin Heilbut - Rime:** How do you think about things? How do you approach a conversation? We'll play different stakeholders within the [0:28] **Colin Heilbut - Rime:** account, that and we'll just hand it over to you. [0:32] **Colin Heilbut - Rime:** Excellent. It's a pleasure to make your coincidence. [0:37] **Colin Heilbut - Rime:** I can, perhaps kick it off with a bit, quick introduction to my [0:44] **Colin:** My name is Colin Hubbard. I'm in fact the newest salesperson here at Rhyme. New to the company, I've been in Speech AI going on [0:47] **Colin:** three, four years now, actually first employee at Deepgram. [0:50] **Colin:** You might have heard of, and then intelligent voice and high security speech recognition. So if you had a whole bunch of [0:55] **Colin:** legal audio from a a court case and it go through the cloud, you could give it to us [0:58] **Colin:** We were one of the first doing GPU inferencing before it was cool to process that real fast. [1:03] **Colin:** I entered a student not long ago to French AI Lab, also in the space [1:07] **Colin:** but really excited about Rhyme. If you wouldn't mind, could we go [1:09] **Colin:** just go around the room very briefly to introduce yourselves and understand your rules at Assurant? [1:17] **Colin Heilbut - Rime:** Sure. I can go first. I am the chief Product Officer at Insurance. I will come through [1:22] **Colin Heilbut - Rime:** all of our product surface areas for [1:23] **Colin Heilbut - Rime:** our direct to consumer business, [1:26] **Colin Heilbut - Rime:** and yeah. [1:31] **Colin Heilbut - Rime:** Excited to hear about Ron, her friends, and, yeah, looking forward to the conversation. [1:46] **Voice 3:** Yeah. I'm the head of on the hours here. And so you know, we take your product and put it in front of our customers and kinda like the first thing that they hear is, like, the voice. Right? So it has to be really good. [1:48] **Voice 3:** Because they're essentially gonna be selling [1:50] **Voice 3:** you know, selling it [1:55] **Voice 3:** for you guys. So so, yeah, we've been trying to [1:58] **Voice 3:** lot of different things. Like, we just kinda want the dust that's out there. So [2:03] **Colin:** I think you've come to the right place. Good. [2:03] **Patrick:** Okay. And I'm Patrick. [2:03] **Colin:** Nice to meet you. Hey. Good to you. [2:05] **Colin:** I lead team of engineers. [2:06] **Patrick:** Focused on [2:07] **Patrick:** implementation [2:12] **Colin:** Cool. And so [2:12] **Voice 4:** I've been doing a little bit of research about your [2:12] **Colin:** amazing company. But do you always [2:12] **Voice 4:** of [2:15] **Colin:** focus on a particular part of the business or [2:21] **Colin Heilbut - Rime:** For us, like, we're we're all focused on the customer experience out of the house. [2:26] **Colin:** Oh, I mean, like you [2:26] **Voice 4:** have some automobile, [2:26] **Colin:** but you do a lot of, the phone warranty stuff or a little bit of everything? [2:29] **Colin Heilbut - Rime:** Most of us focus like, the primary business universe is, the [2:30] **Colin Heilbut - Rime:** warranty [2:30] **Voice 4:** side. [2:31] **Colin Heilbut - Rime:** At Assurant. [2:36] **Colin Heilbut - Rime:** Mhmm. So we have a bunch of different clients, right, within that [2:37] **Colin Heilbut - Rime:** business. [2:37] **Colin:** C [2:37] **Voice 4:** mobile or [2:42] **Colin Heilbut - Rime:** Yeah, like T Mobile, Geico's, like another one, right? Like we do [2:43] **Colin Heilbut - Rime:** like we work across [2:45] **Colin Heilbut - Rime:** all the different clients. [2:48] **Colin Heilbut - Rime:** Most of us around the table represent, like, a core kinda, like, [2:52] **Colin Heilbut - Rime:** C Max product group that looks after this, like, [2:54] **Colin Heilbut - Rime:** digital interactions with customers. [2:59] **Colin:** Yeah. Walt, did you wanna introduce yourself? [3:02] **Colin Heilbut - Rime:** Yeah. Sure. [3:03] **Patrick:** I [3:08] **Patrick:** am in charge of the kind of contact [3:11] **Voice 5:** center operations here. Oh, wow. So I manage [3:14] **Voice 5:** 2,000 seats right now. [3:17] **Voice 5:** And we used to have 3,000. [3:19] **Voice 5:** We're kind of like [3:23] **Voice 5:** trying to balance you know, teeth versus [3:24] **Voice 5:** voice agents. [3:27] **Colin:** So two thousand first distill human seats. [3:29] **Voice 5:** Yep. Exactly. [3:35] **Colin:** Cool. So I wanted to propose an agenda, but, look, we we can always add or subtract as need be. We've [3:40] **Colin:** done some introductions. I wanted to briefly talk about a format that works well for us. Then [3:45] **Colin:** just touch very briefly, you know, what what makes Rhyme different to some of the TTS you make [3:45] **Colin:** currently be using or or heard about [3:50] **Colin:** Obviously, the bulk of today, I really just wanted to talk about VoiceAI at [3:52] **Colin:** at Assurant. I understand that you you have a very successful operation at the moment. [3:57] **Colin:** There are some concerns, I think, in my notes from our partner NICE, who introduced us. [3:59] **Colin:** Something about voice quality, but that could mean anything from [4:04] **Colin:** latency to expressiveness and prosody. So I really wanted just a deep dive on that. [4:08] **Colin:** Then, again, very briefly, I'd want to show you a little taste [4:09] **Colin:** of what we can do. [4:14] **Colin:** And really along the way, please stop me if you have questions or want to [4:18] **Colin:** click on anything, and then I'll leave some special time for questions at the end and talk [4:21] **Colin:** about the next steps. So Nick, everyone else, does that sound okay? [4:22] **Colin:** Yeah. Sounds great. [4:22] **Colin Heilbut - Rime:** We're [4:27] **Colin:** Great. So one small thing. I I just have the worst handwriting in the world. [4:31] **Colin:** So, rather than trying to capture the important things that you're seeing, I have a secure encrypted system here that's [4:36] **Colin:** just gonna be making notes if you talk about a pain point or success criteria. It'll pull that out [4:40] **Colin:** just help me organize the thoughts for the meeting. We'll get rid of it as soon as the meeting's over. I just want to make sure that I [4:43] **Colin:** hear everything you say. That's why I'm here today. Makes sense? [4:48] **Voice 5:** Great. [4:48] **Colin:** Gosh. So very briefly, I introduced myself, but [4:53] **Colin:** tell you about rhymexic, it's a huge coincidence, our we actually have an audience here from [5:00] **Colin:** engineering program at at Stanford where he met Lilly who's also from there dropped out of the PhD program to help build this company since 2022. [5:05] **Colin:** And Brooke as well, also from a linguistic background like Lilly, [5:10] **Colin:** And we commented a little bit differently. So rather than sort of pure [5:14] **Colin:** computational AI approach, have a lot of folks doing that too, but we were [5:16] **Colin:** linguist first and foremost. And so we actually started by building a recording studio. [5:21] **Colin:** So some of the other companies in the industry will use [5:24] **Colin:** pretty much the same data set of podcasts or actors, and it actually works well, you know, [5:29] **Colin:** if you look at the results, there are very strong competitors out there. [5:35] **Colin:** We think our approach is differentiated by spending thousands of hours having real people off the street, have natural, like full duplex conversations. You'd be [5:39] **Colin:** shocked at the results that can get. And I think we actually have a fair amount of proof [5:44] **Colin:** to that effect. We're now doing tens of millions of calls every month [5:48] **Colin:** some of the largest restaurants in the country. Healthcare organizations, telecom and enterprise. [5:52] **Colin:** Some other data that's just very briefly before we get into [5:54] **Colin:** the heart of the conversation about what's going on at Assurant, [5:59] **Colin:** We recently did a 100,000 phone calls. Well, we but there's a [6:03] **Colin:** impartial third party that we commissioned to do that. They did about a third of them with our [6:08] **Colin:** technology, then they used two of the competitors as well. [6:11] **Colin:** The big headline is no one likes receiving cold calls. So there's a lot of hang ups on on humans and [6:12] **Colin:** even best in class AI. [6:17] **Colin:** But there's still some some really important differences. And when you see it, [6:21] **Colin:** the scale here, these small numbers actually add up. So we actually have the lowest hang up rate out [6:22] **Colin:** 30,000 calls to [6:27] **Colin:** Compared to the other ones, people were less likely to hang up on us. They were the fastest [6:28] **Colin:** to actually engage in [6:31] **Colin:** through the surveys, those that responded positively. And, we [6:33] **Colin:** I think that speaks volumes. [6:38] **Colin:** Perhaps more relevant to your business, a different type of data is just some of our large customers. [6:38] **Voice 4:** We could [6:42] **Colin:** potentially share some case studies on these with actual names, but [6:43] **Colin:** 25% increase in call containment. [6:48] **Colin:** Was a a big one that we've we've noticed usually with a telecommunications provider, 20% increase [6:53] **Colin:** sales, Nick. So so you're writing there. Does this seems somewhat relevant or interesting? [6:55] **Colin Heilbut - Rime:** Yeah. I think, like, [7:02] **Colin Heilbut - Rime:** it's relevant. I think that for us, [7:04] **Colin Heilbut - Rime:** we're also trying to figure [7:04] **Voice 5:** out, [7:04] **Colin Heilbut - Rime:** right, [7:10] **Colin Heilbut - Rime:** you know, like, more of an apples to apples comparison. Right? Like, [7:11] **Voice 4:** we [7:12] **Colin Heilbut - Rime:** are [7:17] **Colin Heilbut - Rime:** somewhat nascent on our voice AI journey [7:20] **Colin Heilbut - Rime:** We are still, like, in this transition, like Walt mentioned, from [7:25] **Colin Heilbut - Rime:** call centers to voice agents. Right. In the middle, we have this, [7:26] **Colin Heilbut - Rime:** IPR that we [7:28] **Colin Heilbut - Rime:** deployed today. That's kind of like a smart IVR. [7:30] **Colin Heilbut - Rime:** But [7:35] **Colin Heilbut - Rime:** yeah. So definitely development, like everything you've walked through and, like, that's [7:39] **Colin Heilbut - Rime:** why we're here. And so I'm just writing notes for myself. [7:44] **Colin:** Cool. I'm glad that we're on the right track. Mhmm. [7:44] **Patrick:** Question kind of on [7:48] **Patrick:** before when you're talking about your data collection. [7:48] **Patrick:** I imagine my people just having kind of [7:53] **Patrick:** conversations, like that's one kind of speech too, but you know, when you think about the kinds [7:55] **Voice 5:** talking [7:55] **Patrick:** to somebody in the call center, [7:56] **Patrick:** that's a different kind of [7:58] **Voice 4:** know, [8:02] **Patrick:** Yeah. Style. Is that also stuff that you guys collect as well or [8:06] **Colin:** So I'll I'll talk a little bit later once I actually let you hear to some of our [8:11] **Colin:** how we we think about this. But for example, for our latest model, CODA, we went through [8:15] **Colin:** had a thousand potential voices that were generated from the process that I described, and then we've sort of pair matching to sort [8:20] **Colin:** break it down into certain categories that would be hyper relevant [8:23] **Colin:** to you. And then we have a tool, through our UI that allows you to expose obviously the ones that we've whittled down [8:26] **Colin:** can then pick if you want a formal or professional [8:31] **Colin:** American versus other different languages [8:32] **Colin:** So there's quite a sophisticated process behind it. [8:37] **Colin:** I can always have an expert get back to him a little bit new here, but that's sort of my high level understanding. [8:45] **Colin:** Gosh, so let's just dive in. Nick, don't know if you wanna [8:50] **Colin:** start just to you mentioned you have a solution. Are you able to share the solution provider's [8:52] **Colin:** today through NICE who your TTS provider is? [8:56] **Colin Heilbut - Rime:** Yeah. I mean, I think Pat should share a little bit more, but, like, [9:01] **Colin Heilbut - Rime:** we currently use NICE's cotton energy product. [9:04] **Colin Heilbut - Rime:** Sure. [9:04] **Colin:** Someone familiar [9:04] **Patrick:** with [9:04] **Colin:** that. Is it there? [9:04] **Colin Heilbut - Rime:** It's like [9:08] **Colin Heilbut - Rime:** I think they bought it a few years ago. We've been a nice customer for a [9:08] **Voice 5:** long time. [9:10] **Colin Heilbut - Rime:** But we wanna create, like, [9:14] **Colin Heilbut - Rime:** fully conversational voice agents for us. [9:19] **Colin Heilbut - Rime:** The experience is still like somewhat like deterministic and [9:23] **Colin Heilbut - Rime:** static and so we started to do like some use cases [9:27] **Colin Heilbut - Rime:** fully, like, identity, but still it's, like, only scratching the surface. [9:32] **Colin Heilbut - Rime:** Think from a modeling standpoint, Patrick can keep me honest, [9:35] **Colin Heilbut - Rime:** but I believe like we're just using Microsoft models [9:40] **Patrick:** like [9:40] **Colin Heilbut - Rime:** hyperscaler because [9:40] **Colin:** They're an amazing company, but their models are not [9:44] **Colin:** generally considered stated there. So, yeah, that means there's [9:47] **Colin:** there's definitely some some quick ones there. From what you've heard of the the ones you're using [9:52] **Colin:** so far, I I I just noted again voice quality, but what's [9:53] **Colin:** important to you there and why does that matter? [9:55] **Patrick:** Yeah, mean, from our perspective, like, [9:58] **Patrick:** I mean, look, it's no surprise. [9:58] **Colin:** Like, [10:01] **Colin:** Microsoft voices sound [10:01] **Patrick:** good enough. [10:01] **Voice 5:** Like, [10:06] **Patrick:** think they don't they don't sound great. We've had some [10:09] **Patrick:** customer complaints and things come [10:09] **Voice 5:** in, [10:09] **Voice 4:** but You've had [10:09] **Colin:** customer complaints about the voice quality. [10:09] **Voice 5:** Okay? [10:13] **Patrick:** Yeah. It kinda showed up in [10:13] **Voice 4:** some of the, [10:13] **Patrick:** like, surveys we run, but [10:15] **Patrick:** look, like, it it's working. And I think one of the things we really care about is, like, [10:17] **Patrick:** accuracy and, [10:22] **Patrick:** we've really dialed in the pronunciation and stuff like that with Microsoft. So, like, we feel [10:22] **Colin:** like [10:27] **Colin:** any change we would make, we would need to make sure that, like, that accuracy is maintained to [10:32] **Colin:** You know, when a child learns a new word, they're often only told once. [10:36] **Colin:** And they can start to replicate that. So that's something our team is sort of [10:40] **Colin:** had in mind when we created our system. And certainly with a missed one of our [10:45] **Colin:** cutting edge models, it's not coded, but it has a lot of advantages as well. You can teach it once [10:50] **Colin:** with throughout the system we built, and it will pronounce things extremely precisely according to [10:54] **Colin:** their proper pronunciations. They're minutes on an [10:54] **Voice 5:** API. [10:59] **Colin:** So the the sound quality or the pronunciation is very important to you. [11:04] **Colin:** I think from my perspective, especially building these systems and, like, [11:04] **Patrick:** our team is kind of not only in charge of [11:10] **Patrick:** the initial implementation, but, you know, inevitably, like, something comes up on behalf [11:10] **Colin:** of the one [11:10] **Patrick:** that goes and fixes it. [11:15] **Colin:** And then you have sort of a way of measuring sort [11:15] **Voice 4:** of [11:18] **Colin:** WER and that sort of thing, a word error rate, [11:18] **Voice 5:** anyways? Yeah. [11:18] **Voice 4:** Yeah. [11:18] **Patrick:** I think, like, [11:23] **Patrick:** I mean, just candidly, like, this [11:23] **Voice 5:** is still a very kind [11:23] **Patrick:** of material [11:26] **Patrick:** process, and I've got somebody on my team that's, like, going through. And I think, you know, I just spend a little bit [11:31] **Patrick:** cautious about moving from the kind of static [11:31] **Patrick:** tree where everything is varied [11:36] **Voice 6:** kind of fixed into more of these agentic solutions. But I know this is where [11:36] **Voice 4:** we're gonna [11:37] **Voice 6:** end up. So I just wanna make sure that [11:41] **Voice 6:** as we migrate from where we've done a lot of manual testing to the [11:44] **Voice 6:** things get a little bit out of our control, then you can kind of [11:45] **Voice 6:** help us think [11:45] **Voice 4:** through [11:51] **Colin:** But I don't think it'd be a bit retrend in suggesting anything, but certainly, Myst is [11:55] **Colin:** kind of our more precise tool at the moment and Coda is getting precise and has [11:58] **Colin:** most human voices. So there are some great options for you with [12:00] **Colin:** both. [12:05] **Colin:** Arianne, anything to add in terms of the Asia we wanted to [12:07] **Colin:** you know, would really make a big improvement [12:09] **Colin:** for your new program. [12:11] **Colin:** Yeah. I mean, think like, [12:11] **Patrick:** you know, [12:16] **Patrick:** I was saying we have this 2,000 human agents who really [12:19] **Patrick:** you know, we try to pass everybody on [12:19] **Voice 6:** to [12:24] **Patrick:** one of the human agents eventually. And so, like, for the system that [12:24] **Voice 4:** we're [12:24] **Patrick:** building. [12:24] **Colin:** So there there might be a [12:29] **Colin:** they would speak with an AI and then eventually maybe be transferred to the expert. [12:31] **Patrick:** Exactly. So, like, we're just trying to keep them [12:36] **Patrick:** on the the genetic part of the foot for as long as possible, we can kinda gather all the information. [12:41] **Patrick:** That we then pass on to the human agent. And so if they're collecting [12:44] **Patrick:** you know, speak to an agent earlier, right, then the human agent has to, like, take on that work. [12:49] **Colin:** And well, no one likes more work, but then in terms of the the bottom line to the business, why does that matter? [12:55] **Patrick:** I mean, it's just kind of capacity, right? It's like [12:57] **Colin:** Oh, gosh. So someone I would have to be potentially be on hold if I'm not putting words in your mouth. [13:02] **Colin:** Yeah. Put on hold, [13:02] **Patrick:** like, that stuff might be toxic [13:02] **Colin:** to you. Or you just have to staff more, and then I guess that's [13:04] **Patrick:** Exactly. Yeah. Yeah. All those reasons. [13:07] **Colin:** So some of the success criteria would be [13:12] **Colin:** minimizing the or maximizing the the the time the folks [13:16] **Colin:** spend giving information to the agents and minimizing the amount of time folks Yeah. Are talking to me. [13:17] **Voice 6:** Yeah. [13:21] **Colin:** Success metrics or task completion and call [13:23] **Colin:** completion and [13:28] **Voice 5:** Task [13:28] **Patrick:** completion and call containment. [13:28] **Voice 6:** Yeah. That's [13:28] **Colin:** what I meant. Thank you. [13:29] **Voice 6:** Okay. I [13:29] **Colin:** figured [13:29] **Voice 4:** it out there. [13:34] **Colin:** Walt, sorry. We we've kind of been neglecting you here. [13:38] **Colin:** A lot of folks in the room. Anything you'd like to add before we kind of dive into [13:42] **Colin:** some more specific examples of why Rime might be able to help with some of the the problems we've identified today? [13:47] **Colin:** No. I think it's okay. [13:47] **Voice 5:** I think [13:49] **Voice 5:** my team is just [13:49] **Voice 4:** of [13:51] **Voice 5:** getting laid off. [13:56] **Colin Heilbut - Rime:** Or or put put one I see, like, [13:58] **Colin Heilbut - Rime:** angry customers arriving [13:58] **Voice 4:** on [13:58] **Colin Heilbut - Rime:** wants to ask. Right? Like, [14:03] **Colin Heilbut - Rime:** we don't want a bad experience of some of these emotional having conversations be [14:08] **Colin Heilbut - Rime:** super frustrated by the end of the voice agent conversation that [14:09] **Colin:** Yeah, no, mean, I I I certainly we [14:09] **Patrick:** we [14:14] **Colin:** not involved in in any of that, but what what we are involved in is [14:17] **Colin:** hearing and doing studies on some of these conversations and [14:21] **Colin:** having worked in a few AI labs, a rhyme is one of the first places where where folks will actually hit [14:34] **Colin:** thank you to our 10% of his rent [14:37] **Colin:** was being asked for him because he lost his temper because the AI was so bad that I [14:41] **Colin:** assure you that was not our [14:41] **Voice 4:** system. [14:41] **Colin:** And, yeah, I [14:44] **Colin:** I reached out to the meeting and said, like, this is not, you know, what it the future. [14:46] **Colin:** So we we care a lot about sort of the [14:47] **Colin:** the AI human experiences. [14:53] **Colin:** So if you mind, I'm gonna try a a live [14:54] **Colin:** demo [14:59] **Colin:** We can always go back if unless there's any sort of pinpoints that I didn't capture here. It is [15:00] **Colin:** important that [15:04] **Patrick:** Well, I think just like the [15:05] **Patrick:** infrastructure side to, like, [15:06] **Patrick:** we [15:10] **Patrick:** you know, operate in a compliant environment and know what are the benefits of using. [15:15] **Patrick:** The Azure solution is, like, we're in the Microsoft data centers and so [15:20] **Patrick:** just wanna understand, like, the different options for for serving [15:20] **I'm:** Sure. I'm I'm [15:20] **Voice 6:** not [15:24] **Colin:** super technical myself but I do understand that we have [15:27] **Colin:** APIs available in different regions. But if you wanted to use a cloud that we're not natively on, we have [15:27] **Voice 6:** a [15:32] **Colin:** full on prem deployment that's actually [15:32] **Voice 5:** but it's [15:32] **Colin:** semi automated in [15:32] **I'm:** how it's [15:33] **Colin:** it's managed, and that would absolutely, with a simple Docker container, be [15:37] **Colin:** available to you on Microsoft or on your own [15:42] **Colin:** hardware or lots of different ways to do that [15:48] **Colin:** extremely securely, SOC two Type two working with some of the largest most of ending companies in in the country. So we're definitely [15:53] **Colin:** that road before. Any other questions from a technical perspective? [15:55] **Patrick:** I think so. I mean, like, how are they kind of just, [15:59] **Patrick:** or what I hit or other type kind of metrics, like latency and [16:04] **Colin:** I'm also latency is certainly one place that I hadn't memorized the exact stats and some of the benchmarks [16:09] **Colin:** change day by day, but we're we're towards the the one of the fastest models in the world at at the moment. [16:09] **Colin:** I I don't [16:09] **Voice 6:** think that's the [16:13] **Colin:** the only thing. If something's really fast and doesn't sound great, then [16:18] **Colin:** but we we I I think you'll you'll see for yourself in a moment. We are both fast [16:20] **Colin:** responsive, and I think very human. [16:20] **Voice 5:** Cool. [16:25] **Colin Heilbut - Rime:** I'm excited to get them done. [16:25] **Voice 5:** Okay. [16:27] **Colin:** Was just calling my [16:27] **Colin Heilbut - Rime:** dentist. Let me try this one. [16:32] **Voice 7:** Hello, and thanks for calling NorthStar Device Protection. [16:35] **Voice 7:** Is a demo, and the call may be recorded. [16:37] **I'm:** I'm [16:37] **Voice 7:** NorthStar's virtual claims assistant. [16:39] **Voice 7:** How can I help you today? [16:44] **Colin:** Hey, good afternoon. I'm so sorry to report that I have a [16:45] **Colin:** problem with my iPhone. I'd like to apply for warranty. [16:50] **Voice 7:** I'm sorry to hear about the trouble with your iPhone. [16:54] **Voice 7:** Assist you with the device claim, I first need to verify your identity. [16:57] **Voice 7:** Could you please remind the five digits ZIP code on your policy? [16:59] **Colin:** 94115. [17:08] **Voice 7:** Thanks for that, Colin. I know we'll check your policy now to [17:08] **Voice 7:** see what coverage you have for your iPhone. [17:15] **Voice 7:** Colin, your active policy includes [17:16] **Voice 7:** accidental damage coverage for your iPhone 16 Pro. [17:20] **Voice 7:** To start a claim, when did the damage [17:20] **Colin:** happen? I think it happened on July 11. [17:24] **Colin:** So embarrassing. Thanks, July. [17:26] **Colin:** So, what were you saying? [17:34] **I'm:** Was the [17:34] **Voice 7:** damage accidental, like dropping or popping the [17:37] **Colin:** I was getting ready for this really important presentation. [17:41] **Colin:** I was so excited thinking about these folks I was going to meet and I was building this phone app and I just [17:44] **Colin:** I knocked the darn thing right onto the floor and cracked the screen. [17:45] **Colin:** What a dummy I am. [17:53] **Voice 7:** It sounds like it was accidental damage then. [17:57] **Voice 7:** Did your iPhone still power on and work despite the cracked screen? [17:57] **Colin:** It sure does. [18:02] **Voice 7:** Thanks for confirming that your iPhone still powers on. [18:05] **Voice 7:** Do you still have the device with you now? [18:05] **I'm:** I do. [18:12] **Voice 7:** Just to summarize, you had accidental damage on July 11. Your iPhone is still power gone. [18:16] **Voice 7:** Is that right? [18:16] **Colin:** Yes, [18:16] **I'm:** ma'am. [18:28] **Voice 7:** Your claim submitted with claim number CLMDemo2049 [18:28] **Patrick:** Sounds like you're [18:28] **I'm:** gonna put a [18:31] **Colin:** Sorry. Didn't mind him. What were you saying? [18:39] **Voice 7:** I'm saying I can send a secure document uplift link to your phone to help finish the claim if you'd like. [18:40] **Voice 7:** Like. [18:45] **Voice 7:** Would you like me to send that now? Or is there anything else I [18:45] **Voice 7:** can do for [18:45] **I'm:** you? Yeah, you can send that. [18:51] **Colin:** You said usually you transfer at this point to a like a [18:52] **Colin:** Oh, don't know if our system can do that. [18:55] **Colin:** If you need me more talk to a specialist, [18:55] **Voice 7:** just let me [18:55] **Voice 5:** know. [18:59] **Colin:** Yeah. Can I just can you transfer me to an expert? [19:07] **Voice 7:** Alright. I'll transfer you now to a device claim specialist who already has your full summary. [19:08] **Voice 7:** One moment, please. [19:13] **Patrick:** Hi, Colin. I have your full claim summary [19:13] **Voice 6:** here. [19:16] **Patrick:** How can I assist you with moving your claim forward today? [19:19] **I'm:** Est-ce que vous pouvez m'aider en français s'il vous plaît [19:29] **Voice 7:** Bonjour Colin, j'ai marqué votre réclamation pour dommage accidentel [19:33] **Voice 7:** sur iPhone sales pro a bien été soumis le haut juillet. [19:37] **Colin:** C'est tellement gentil, mais je pense we get the point. [19:38] **I'm:** Merci et bonne après-midi au revoir. [19:48] **Colin:** Yeah. I apologize. I didn't wanna use your brand without your permission, so we [19:52] **Colin:** created Northstar for that purpose. But [19:52] **I'm:** how was the voice quality? [19:56] **Voice 6:** Voice quality is out of reach. Yeah. I'm curious how you did the [19:59] **Voice 6:** language switching. Like, how did [19:59] **I'm:** you navigate that? [20:04] **Colin:** Well, we we have folks that just as I was spinning, like, our first office had to [20:05] **Colin:** recording studio in it. [20:10] **Colin:** We we have folks that are doing recordings for us overseas. So we have native French speakers and [20:15] **Colin:** quite important that you have the speakers speaking their language else. The English model [20:17] **Colin:** can the voices can do it, it does not sound, like like she does. [20:22] **Colin:** And so the live kicks and the logic they're [20:28] **Colin:** the knowledge to switch when it picks up in Deepgram, our STT provider, [20:30] **Colin:** identifies language and [20:35] **Colin:** we go. We support something like eight languages, not the most, but [20:38] **Colin:** I think certainly for this country, we we have a large amount of coverage, and we're always expanding it. And if there are other languages [20:41] **Colin:** were very important to you, that's something I'd always take back to my team. [20:50] **Colin Heilbut - Rime:** I guess, like, what what are the next steps for us to, like, [20:54] **Colin Heilbut - Rime:** try and like, I'll tell you on our side, [20:57] **Colin Heilbut - Rime:** The purpose of today is just to like understand [21:01] **Colin Heilbut - Rime:** a little bit more about rhyme. Right? Think I hear the demo, which sounds great. [21:03] **Colin Heilbut - Rime:** Think the next step for us is just [21:04] **Colin Heilbut - Rime:** like, [21:09] **Colin Heilbut - Rime:** how can we get started and bring more awareness and [21:10] **Colin Heilbut - Rime:** test out the models in our environment. [21:14] **I'm:** So [21:18] **Colin:** this is a just to skipping ahead here because you asked, this is kind of what that would look like. [21:23] **Colin:** You know, as much as we care about and we're happy to show you our latency and the [21:24] **Colin:** know, word error rate [21:30] **Colin:** I think in in our experience, a really successful pilot can actually would would work [21:33] **Colin:** once you're confident on a small amount of real production calls. [21:35] **Colin:** This is what our our customers generally do. [21:39] **Colin:** And we would set success criteria out in advance. So [21:44] **Colin:** for me to tell you what what matters, but looking at the the notes here, it looks like we have [21:48] **Colin:** you know, call completion. Maybe it could be something when we sit down to scope a call or [21:50] **Colin:** you know, the the [21:55] **Colin:** sort of brand pronunciation. [21:55] **Voice 5:** Really, what whatever says [21:58] **Colin:** there are important to you, we would just scope out over whatever period of time. Nick, do you have a [22:00] **Colin:** time period that would make sense? [22:04] **Colin Heilbut - Rime:** I think for us, like, we're looking to make a decision over the next [22:08] **Colin Heilbut - Rime:** like, three to four months. Now, I think there's a [22:11] **Voice 5:** couple phases [22:11] **I'm:** for us. Three to four months [22:11] **Colin:** for next steps. Okay. [22:11] **I'm:** Yeah. [22:12] **Colin Heilbut - Rime:** I think, like, in the interim, [22:14] **I'm:** we [22:17] **Colin Heilbut - Rime:** have, like, a [22:21] **Colin Heilbut - Rime:** dev environment that's, not a production environment, and we'd love to be able [22:26] **Colin Heilbut - Rime:** to implement rhyme and test just like [22:29] **Colin Heilbut - Rime:** with our own ears how it sounds and how it integrates [22:32] **Colin Heilbut - Rime:** Like how can we get started in the short run just like [22:37] **Colin Heilbut - Rime:** getting it set up in that test environment? [22:37] **I'm:** Yeah. Obviously, this is sort of sequential and a [22:42] **I'm:** wouldn't ask you to put something in even in a small production run until you've gone through that phase. So [22:46] **I'm:** you [22:46] **Colin:** know, we're also we we host APIs on on Amazon, so you could be up and running in [22:51] **Colin:** five minutes, technically, maybe a little longer with NICE, but we have a native integration with NICE. [22:53] **Colin:** So that's still SOC two type two. [22:59] **Colin:** Depending on your security requirements. So we can actually maybe get into that [23:01] **Colin:** a little bit later, but [23:06] **Colin:** what is kind of the the process for for going [23:09] **Colin:** even into a a pilot, or how how complicated is that to get approval even to user APIs and [23:12] **Colin:** the the dev environment? [23:12] **Colin Heilbut - Rime:** Yeah. I think, like, [23:18] **Colin Heilbut - Rime:** step one to just get, like, a demo set up right Mhmm. In a [23:20] **Colin Heilbut - Rime:** dev environment is pretty [23:20] **I'm:** easy. Oh, great. [23:29] **I'm:** Try selling to a French bank. [23:29] **Colin:** Not easy. No. [23:29] **Colin Heilbut - Rime:** Definitely not. Or US [23:29] **Voice 5:** So [23:29] **Colin Heilbut - Rime:** like we have the authority like [23:32] **Colin Heilbut - Rime:** we have to do, like, some baseline security checks, [23:32] **Voice 5:** right, [23:32] **Colin Heilbut - Rime:** on, [23:35] **Colin Heilbut - Rime:** you know, your your documentation and whatnot. But, like, we'd love to start testing [23:40] **Colin Heilbut - Rime:** your call API Mhmm. In some constrained demo environment. [23:43] **Colin Heilbut - Rime:** And get it set up. [23:43] **I'm:** We love test. [23:48] **Colin Heilbut - Rime:** And then just being able to allow people to, like, hear it. I don't know if [23:48] **Voice 5:** you have [23:52] **Colin Heilbut - Rime:** suggestions on voices to use. [23:55] **I'm:** We our RRI is [23:55] **Colin:** one of our experts there, and he can [23:59] **Colin:** help guide [23:59] **I'm:** that. So he still can't So, [23:59] **Colin Heilbut - Rime:** that's to me, like, we wanna figure out what we [24:02] **Colin Heilbut - Rime:** is of the best. We want to test it and, like, [24:06] **Colin Heilbut - Rime:** I think we wanna lean on you guys to help us figure out, like, how do we set up the best [24:07] **Colin Heilbut - Rime:** demo test environment Mhmm. [24:09] **Colin Heilbut - Rime:** Then, like, [24:13] **Colin Heilbut - Rime:** you know, that's step one, right, for us. [24:16] **Colin Heilbut - Rime:** Does it sound good in our own, like, environment? Like, [24:19] **Colin Heilbut - Rime:** to our own ears, right? [24:24] **Colin Heilbut - Rime:** Step two is we want to understand what does it take technically to roll this out [24:24] **Colin Heilbut - Rime:** on prem [24:26] **Colin Heilbut - Rime:** for some pilot period. [24:31] **Colin Heilbut - Rime:** Right? And probably, you know, Patrick and his team to kinda, like, [24:33] **Colin Heilbut - Rime:** take that action item, that like, next step. [24:35] **Colin Heilbut - Rime:** And then I would, like, work with [24:39] **Colin Heilbut - Rime:** artists involved on the commercial side to figure out, like, [24:44] **Colin Heilbut - Rime:** kind [24:44] **Voice 5:** of [24:44] **Colin Heilbut - Rime:** That [24:44] **I'm:** that all sounds good. Yeah. We're we're pretty experienced with [24:48] **Colin:** process. Tell me though, Ed, you said you're using Microsoft at the moment. Did they go through this whole process with you? Did they help you [24:52] **Colin:** sort of evaluate which of their voices was best or how hands on was Microsoft? [24:57] **Colin Heilbut - Rime:** They kinda suggested one oh, I should we go with, but, you know, it's [24:58] **Colin Heilbut - Rime:** we just thought it because [25:03] **Colin:** I mean, listen. No no one ever gets [25:03] **I'm:** from [25:03] **Colin:** for buying Microsoft, but I'd like to think of Rime that's, you know, we [25:07] **Colin:** have a more crafts person approach. It's one of our values and we take a lot of [25:07] **Voice 6:** pride in [25:12] **Colin:** sort of customer service and we're quite accessible, even myself. I [25:16] **Colin:** cover many different territories and usually most hours of the day [25:20] **Colin:** available with the Slack or text. [25:20] **I'm:** Okay. Can I actually get your number now while I I [25:20] **Colin:** just mentioned it? Or [25:27] **Colin Heilbut - Rime:** Let's wrap, and then I can I can give you my [25:30] **Colin:** Great? No. It's good that I guess in my my line of work we call it a champion, you know, [25:32] **Colin:** just one person that I can reach out so that can always be accessible [25:33] **Colin:** anytime there. And I [25:38] **Colin:** Yeah, for sure. [25:38] **I'm:** As any other kind of stuff we [25:42] **Colin:** should be aware of down the road, things we can start preparing on our end. Whenever you ask us for a [25:44] **Colin:** security review or anything, we should, have any big surprises. [25:48] **Colin Heilbut - Rime:** I think the general, like, security stuff. Right? Like, [25:53] **Colin Heilbut - Rime:** I'm assuming you've worked with other, like, financial [25:53] **Colin:** services. Airlines, financial services. Yeah. So [25:56] **Colin Heilbut - Rime:** talk to all the data [25:59] **Colin Heilbut - Rime:** productions or requirements. [26:06] **Colin Heilbut - Rime:** I don't think it's something I have at the moment. We have this, like, GenAI form that we have to fill out to, but [26:06] **I'm:** that's [26:06] **Voice 6:** just for any [26:11] **I'm:** Yeah, look, unfortunately, the nature of the TTS system, especially one provided by our partner in Orchestration Layer, [26:16] **I'm:** very familiar with, it connecting is just not [26:16] **I'm:** technically complicated. Certainly, [26:21] **I'm:** maybe [26:21] **Colin:** adding some of the terminology that you might want. There's a little bit of technical stuff there. [26:25] **Colin:** Picking the the But for as as implementations go, I think you'll [26:26] **Voice 5:** be [26:31] **Colin:** pleasantly surprised. [26:31] **Voice 5:** So [26:33] **Colin:** I guess just to kind of wrap up what we've talked about, it sounds like you do have some [26:38] **Colin:** major limitations with the Microsoft Voices that the call completion is a problem. We have some of the others [26:43] **Colin:** written down that the experience for your employees [26:44] **Colin:** and the staffing levels, the cost involved, [26:47] **I'm:** significant and there's a lot of areas to [26:47] **Voice 5:** improve there. [26:51] **Colin:** And does it sound like Walt, maybe I haven't heard from you lately. [26:56] **Colin:** What a lot of what we've discussed today could could address those concerns, or is there anything that left [26:56] **Voice 5:** unaddressed? [27:00] **Voice 5:** No. I think all the [27:01] **Voice 5:** Makes sense. [27:06] **Colin:** Any other questions at all Walter? I [27:09] **Colin:** know if you read about the industry. I'm happy we have just a couple minutes left on the the time allotted. I wanna manage that. [27:10] **Colin:** Carefully. But [27:15] **Colin:** anything at all, [27:15] **I'm:** I'm I'm here. Otherwise, you know, if you'd like, to [27:18] **Colin:** Nick, do do you know what your team's availability is, say, next week for a a scoping call? [27:21] **Colin Heilbut - Rime:** I can check my calendar. I think I've probably [27:26] **Colin Heilbut - Rime:** Wednesday or Thursday. [27:26] **Colin:** Happy to accommodate your schedule, [27:26] **Patrick:** or [27:28] **I'm:** open. [27:28] **Colin:** I just moved from France. [27:30] **Colin Heilbut - Rime:** Awesome. How's France? [27:35] **Colin Heilbut - Rime:** Hot. [27:35] **Voice 5:** And excited for [27:35] **I'm:** their birthday [27:35] **Voice 5:** today. [27:37] **Colin:** National birthday. Right. Yeah. [27:38] **Colin Heilbut - Rime:** That's exciting. [27:48] **Colin Heilbut - Rime:** Cool. Well, thank you. Yeah. Mean, I'm happy to to [27:48] **I'm:** give you a a couple minutes back unless [27:53] **I'm:** already it's Patrick, there's anything at all that I failed to address or [27:55] **Colin:** last minute questions, weird voice AI stuff you'd like to know about? [27:57] **Colin:** No. Sorry to [27:57] **Voice 6:** dig in [27:57] **Patrick:** more difficult. [28:00] **I'm:** Absolutely. Looking forward to it next Tuesday or Thursday. [28:02] **I'm:** Perfect. It's been a pleasure. [28:08] **Voice 5:** Great. Thanks. Cool.