Talk by Neil Zeghidour (CEO, Gradium AI, co-creator of the full-duplex voice model Moshi) on why natural-sounding voice AI still falls short of natural conversation. He contrasts cascaded pipelines (speech-to-text, LLM, text-to-speech), which face latency limits from tool calls, with speech-to-speech models that remain half-duplex and cannot backchannel, and argues full-duplex architectures and paralinguistic understanding (tone, hesitation, cultural signals lost in transcription) are needed. Also covers the high cost of TTS at scale and introduces Phoneon, an on-device local TTS approach for privacy and cost. Topics include voice cloning, latency benchmarks, architecture tradeoffs, and a Moshi demonstration. # Channel Identity Observations — Video `01a15ca9-7545-768b-a14a-4460a4c58e6e` ## Explicit Self-Description - **Company mission stated by the speaker:** "Our mission is to unlock the unrealized potential of voice AI." - **Claimed scope — model provider, not vertical solutions:** - "We want to be a main model provider for voice for everyone building voice agents and voice solutions." - "We are not working on orchestration, we are not working on specific verticals. We just make building blocks for people who want to build voice AI." - **Product lines named by the creator:** speech-to-text, text-to-speech, speech-to-speech (transformation, translation, dialogue); "streaming speech-to-text, streaming text-to-speech with voice cloning, semantic VAD." - **On-device product announcement:** "our first step is Gradion Phonon. It's on-device TTS… For us, on-device means it runs on a smartphone CPU… less than 100 million parameter." - **Origin story (creator claim):** "this is a spin-off from a non-profit lab we created 2 years ago… with the funding from philanthropists including Eric Schmidt… Xavier Niel"; prior open-source work cited: "Moshi… the first speech-to-speech model for conversation… Pocket TTS." - **Recruiting/hiring signal:** "if you want to join us to bring her to life… you can apply and join us." - **Call to action:** "you can use us on gradium.ai." ## Intended Audience - **Developers and voice AI builders:** framed around what "everyone building voice agents" needs, with technical content on latency budgets (e.g., "the entire stack… to be around 200 milliseconds"), cascaded vs. speech-to-speech architectures, and VAD. - **Conference/industry audience:** "Thanks for all being here today… we had a lot of great presentations around voice this morning"; references to "customers, from investors." - **Startup/consumer-app builders:** targeted by the Phonon pitch — "The goal for us is to allow people to create consumer apps with voice… without having to lose money on the on API." ## Claimed Scope of Expertise / Positioning - **Positioning against competitors:** - On commoditization: "I have a strong strict opposition to some of our competitors say that voice is a commodity now. I think it's completely false." - Names rivals as context: "the best voice AI company in the world which is 11 labs"; "Cezanne AI"; OpenAI's Advanced Voice Mode; Kokoro ("it doesn't have voice cloning"). - **Claimed differentiators:** voice cloning from "10 seconds of your voice"; fast TTS ("very fast TTS"); Moshi claimed as "still the only full duplex model"; Phonon claimed as "better than all the existing on-device models." - **Competitor critique offered as expertise:** hyperscaler voice modes "run at a loss"; TTS costs are the bottleneck for consumer voice apps. ## Observable Presentation Patterns (Demonstrated, Not Just Claimed) - **Live/recorded demonstration style:** multiple demos — cloned podcaster audio, an on-stage "rich mini" gym-bro demo ("it was very risky… I was very scared"), a "live coded travel agent," and a Rick-and-Morty-style Phonon clip. - **Pragmatic, self-critical tone:** acknowledges his own product's limits — "this needs polishing"; "some model was very stupid… it was not an agent, it had no tool call"; admits past positioning shift: "I used to be really at war against cascaded systems" but now sees them as practical. - **Honest framing of field-wide gaps:** "we're still not there, right? The latency is still quite high"; "this is mostly just a glorified text model with a voice around it." - **Recurring rhetorical device:** the *Her* movie analogy, explicitly acknowledged as "the most overused, most annoying analogy… At the same time it's extremely relevant." ## Claims vs. Demonstrated Evidence — Separation Notes - **Creator claims (uncorroborated by material alone):** being "a main model provider"; Moshi being "the only full duplex model"; Phonon being best-in-class on-device; funding history. - **Demonstrated in material:** voice cloning output (Joel clip); full-duplex conversation robustness (Moshi spaceship dialogue); half-duplex failure mode (hotel-room demo); filler-based tool-call latency mitigation (travel agent demo, with speaker admitting imperfection). No private motivations were inferred; all observations are grounded in quoted transcript passages. # Delivery Observations: Neil Zeghidour, Gradium AI ## Tone - **Casual, self-deprecating, and conversational** throughout, even while presenting technical content: - > "Again, I don't want to be mean to anyone, so I will also make fun of myself." - > "Okay, so we have basically I was a bit mean, right?" - **Candid and blunt** when evaluating competitors and the industry: - > "I'm not going to mention the cost of the API of our competitors, but voice is very expensive." - > "At the same time, some model was very stupid." ## Teaching Habits - **Demo-heavy delivery**: repeatedly pauses explanation to play audio/video clips (voice cloning demo, *Her* movie scenes, hotel-room conversation video, Moshi demo, live-coded travel agent, on-device TTS demo). - > "So, just because I always do a live demos, I will do a very short improvised live demo with something that was live coded very quickly, so it may be go very wrong." - **Admits risk and nervousness about live demos**, which recurs: - > "it was very risky. I recognize I was very scared when I did it." - **Contrast-based explanation**: repeatedly frames points as gap between promise and reality ("Her moment" vs. current systems; human conversation vs. half duplex). - **Defines terms explicitly** before arguing about them: - > "So, I think a big confusion is what is speech to speech? So, speech to speech is the idea that now instead of having the three blocks, you only have one..." ## Recurring Phrases & Verbal Patterns - **Filler-heavy, hedging speech**: "you know," "so basically," "and so on and so forth" appear very frequently across the talk: - > "So basically we train voice models, speech-to-text, text-to-speech, speech-to-speech whether it's transformation, translation, speech-to-speech dialogue and so on and so on." - **Self-corrections and stumbles** ("Gradian/Gradio," "the the goal," "I I think"), consistent with improvised or lightly rehearsed delivery. - **Rhetorical framing questions** to structure transitions: - > "So, what does it take to get there?" - > "But then we know that latency of cascaded system is inherently high, right?" ## Interaction with Venue/Audience - **Repeated requests for A/V adjustment**, observable multiple times: - > "Can we have much more volume, please?" - > "And maybe I could ask one last time to increase a bit the volume." - **Directly addresses the conference context** ("we had a lot of great presentations around voice this morning"), referencing earlier speakers by name (Samuel, the "presenter this morning"). ## Honesty About Limitations - A notable pattern of **flagging uncertainty and imperfection in his own demos**: - > "So, this needs polishing is the main idea" - > "it may be go very wrong" - Also candid about his company's past product: - > "some model was very stupid. So, basically, it was, you know, just useless." ## Uncertainty Notes - Whether requests for volume and live-demo nervousness reflect stable habits cannot be confirmed from a single talk; they are observed here as repeated *within* this presentation. - No curated metadata was supplied, so delivery context is limited to the transcript. # Position Record: "Voice AI: When Is the 'Her' Moment?" — Neil Zeghidour, CEO, Gradium AI ## Speaker All positions below are attributed to **Neil Zeghidour**, CEO of Gradium AI (per video metadata), speaking at a conference presentation. ## Company Framing (Stated Claims) - **Gradium's mission**: "unlock the unrealized potential of voice AI" — training speech-to-text, text-to-speech, and speech-to-speech models (transformation, translation, dialogue). - **Business model**: a model provider / building blocks for voice agents; explicitly *not* working on orchestration or verticals. - **Origins**: spin-off from a non-profit lab founded two years ago, funded by philanthropists including Eric Schmidt and Xavier Niel; lab produced **Moshi** (claimed first speech-to-speech conversation model), speech-to-speech translation, and Pocket TTS (CPU model). ## Core Argument: The "Her" Moment Has Not Arrived - Zeghidour calls the *Her* analogy the "most overused, most annoying" in the field, yet "extremely relevant" — a tension he acknowledges and retains. - Current demos (he cites ElevenLabs — "the best voice AI company in the world" — and his own live demos) are: - **Latency too high** - Unable to handle **simultaneous speaking** - "Mostly just a glorified text model with a voice around it" — anything not in text cannot be leveraged - He concedes intelligence is improving, which he says explains traction in voice agents. ## Latency and Tool Calls - Even a fast TTS is "more than 200 milliseconds," while human conversation requires the *entire* stack (understanding, answer, speech) at ~200ms — so cascade approaches cannot sound human on latency alone. - **Bottleneck shift claim**: the main bottleneck is now the **tool call** (500ms–4s latency), making TTS latency optimization "not the most relevant anymore." - Proposed mitigation: **fillers** — model splits into a tool call plus conversational filler, re-inserting results naturally (demonstrated with an improvised travel-agent demo he flags as possibly going "very wrong" and "needs polishing"). ## Speech-to-Speech and Full Duplex - Definition offered: speech-to-speech is one model taking speech in and outputting speech, reducing latency but still not a human conversation. - **Key claim**: "every single speech to speech model except Moshi is half duplex" — either listening or speaking, never overlapping. He names OpenAI's advanced voice model and Sesame AI (which he says he's "a huge fan of") as still half duplex. - Full duplex matters because human conversation includes overlap, coughing, **back channeling** ("mhm"), which "breaks completely" in half-duplex models. Back channeling varies culturally — e.g., in Japanese it's politeness; overlap can reach 20% of time. - He demos Moshi (co-founder Alex in video) as "the most robust to this day" conversational experience — robust to noise and overlapping speakers, ~2 years old. ## Honest Self-Criticism of Moshi (Preserved Contradictions) - Moshi is full duplex, but Zeghidour admits: "some model was very stupid" — no tools, no agent capability, "pointless" after a few minutes, impossible to use in production due to **no observability**, and no real paralinguistic understanding. - On **paralinguistics**: speech-to-speech models *could* capture cues (the information isn't lost), but if trained on audio versions of instruct datasets with factual Q&A, "why would it even try to capture this information?" ## Central Thesis > "As long as we are not able to give to this kind of very natural sounding models the same level of reliability, intelligence, and personalization as cascaded systems, I don't see a path towards them replacing cascaded systems." - He notes he was "really at war against cascaded systems" previously — a stated reversal; cascaded systems are "so practical and so convenient." ## Scalability, Cost, and Privacy - **Cost claim**: voice is very expensive; "the voice mode of most hyperscalers is run at a loss." TTS consumes most costs; he claims to have seen teams "burning their fundraising in TTS bills." - **Privacy**: more openness to AI increases desire for local, private data (citing fear of database hacks). - **Solution offered**: **Gradium Phonon** — on-device TTS running on a smartphone CPU (his definition of on-device), under 100M parameters, claimed better than existing on-device models; contrasted with Kokoro, which "doesn't have voice cloning." Open to private beta; goal is enabling consumer apps to scale without API losses. ## Conclusion - **Direct opposition to competitors' claim that "voice is a commodity"**: "I think it's completely false. Voice is very challenging. The last mile is going to be the most difficult to solve." - Path forward framed as "science and engineering" leading to the *Her* moment — an analogy he both mocks and embraces. # Speaker Attribution & Provenance Record **Video:** *Voice AI: when is the "Her" moment? — Neil Zeghidour, CEO, Gradium AI* **Video ID:** `01a15ca9-7545-768b-a14a-4460a4c58e6e` ## Speakers | Speaker | Role | Evidence | |---|---|---| | **Neil Zeghidour** | **Creator / Presenter** — CEO of Gradium AI | Title metadata; he introduces Gradium ("Quick intro about Gradian"), presents the talk live | | **Alex** (co-founder of Gradium) | Featured in demo footage | "here you'll see my uh co-founder Alex talking to it" | | **"Joel"** (podcaster, named by speaker) | Voice-cloned in demo | "I can let sit say by a famous podcaster that I cloned this morning… I guess you recognize maybe Joel" — voice clone, not a live speaker | | **AI-generated demo voices** | System outputs (not human speakers) | Gradium voice-clone demo; "Logan, your gym bro" gym-demo voice; "Colin from Wanderlust Travel"; Moshi full-duplex demo; Gradium Phonon Rick-and-Morty-style demo | | **Movie clips ("Her")** | Third-party media excerpts | Samantha/AI dialogue scenes played from the film | | **Audience** | Unidentified | Closing "Woo!" | ## Referenced Sources & Third-Party Claims - **Film:** *Her* — multiple scenes played; speaker calls it "the most overused, most annoying analogy" in the field. - **11 labs demo** — speaker calls it "the best voice AI company in the world" (his characterization); government AI-helper demo clip played. - **Samuel** (earlier presenter) — cited for a presentation on cascaded systems. - **Moshi** — Gradium/parent lab's open-source full-duplex speech-to-speech model; speaker claims it is "still the only full duplex model" and that every other speech-to-speech model is half duplex. - **Nvidia** — "recently published the personal plex model based on it [Moshi]". - **OpenAI Advanced Voice Model** and **Cezanne AI** — referenced as examples of half-duplex speech-to-speech models ("I'm a huge fan"). - **Open Router** — cited for tool-call latency claims (500 ms–4 s). - **Kokoro** — referenced on-device TTS competitor; noted by speaker as lacking voice cloning. - **Backers of parent non-profit lab:** Eric Schmidt, "Odolant Odolant" (as transcribed), Xavier Niel — philanthropic funding, per speaker. ## Promotional Segments (Flagged) - **Gradium voice-clone demo** (Joel podcast-clone clip) — product showcase. - **Gym-bro demo with "rich mini"** — Gradium streaming voice model demo. - **Live travel-agent demo** — "live coded" Gradium demo with filler technique. - **Moshi full-duplex demo** — lab's own model demo. - **Gradium Phonon announcement** — on-device TTS (<100M params), smartphone CPU, **open to private beta** — explicit product promotion. - **Closing call to action** — "you can use us on gradium.ai" and hiring pitch ("if you want to join us to bring Her to life"). ## Uncertainty & Availability Notes - Speaker name appears in transcript as **"Gradian/Gradion/Gradio"** inconsistently; canonical spelling taken from title (**Gradium**). - Funder name transcribed as "Odolant Odolant" — unclear/garbled; uncertainty retained. - No curated metadata (comment availability, sources) supplied beyond title; **comment availability unknown**. - All technical claims (latency figures, cost-at-loss claims for hyperscalers' voice modes) are **speaker's own claims**, unverified. Hi everyone. Thanks for all being here today. Really happy to to talk a bit. So it's the right time for this talk because we had a lot of great presentations around voice this morning. I want you to take a bit, you know, time to reflect where we are in the modestly where we are right now in the in the voice AI domain and what is left to do as main challenges. Quick intro about Gradian. So we are Our mission is to unlock the unrealized potential of voice AI. So basically we train voice models, speech-to-text, text-to-speech, speech-to-speech whether it's transformation, translation, speech-to-speech dialogue and so on and so forth. We want to be a main model provider for voice for everyone building voice agents and voice solutions. So we are not working on orchestration, we are not working on specific verticals. We just make building blocks for people who want to build voice AI. I mean I can let sit say by a famous podcaster that I cloned this morning. Uh There should be audio that is output. Man, have you seen what Gradian is doing with voice cloning? It's kind of crazy, like seriously. Can we have a bit of uh 10 seconds of your voice, that's it. Can we have much more volume, please? Man, have you seen what Gradian is doing with voice cloning? It's kind of crazy, like seriously. You record like 10 seconds of your voice, that's it. And the system analyzes the tone, the pitch, the accent, all those little quirks that make your voice yours. Then boom, you type text and it talks back. Okay, I guess you recognize maybe Joel again. I hope you did. Uh so basically this is a spin-off from a non-profit lab we created 2 years ago uh with the funding uh from philanthropists including Eric Schmidt uh Odolant Odolant Xavier Niel. The main idea was to create a lab that has open research uh and uh we focus mostly on speech. So we developed Moshi uh which was the first uh speech to speech model for conversation, uh speech to speech translation, Pocket TTS most recently a CPU model. Uh and we decided to also create this for-profit structure to uh make a product that can be used in production beyond open source. So the the goal of the talk is uh uh basically based on the Her movie. So uh this has been the most overused, most annoying uh analogy I I think in the field. At the same time it's extremely relevant uh because it's a 13-year-old and if you look at one of the introduction scenes, so that's when the main character uh meets uh his AI voice Samantha for the first time. It still sounds like it was, you know, anticipating I think what what interaction could look like. Oh, what what do I call you? Do you have a name? Um yes. Samantha. Hey, where did you get that name from? I gave it to myself, actually. How come? Cuz I like the sound of it. Samantha. And you know, then came the like the the trend of the Her moment. Uh and we got so many Her moments on Twitter and in real life. Again, I don't want to be mean to anyone, so I will also make fun of myself. It's more like, you know, trying to be pragmatic about what was the promise and where we are right now. This is a a very recent uh demo from for me what is the best voice AI company in the world which is 11 labs. Hello, you're speaking to the government's AI helper. How can I help you today? I would like to start a new business. To start a new business, you'll need to choose a business structure and name, complete the registration, and upload required documents. And this one is a demo that I did two months ago with a rich mini. So, our presenter this morning do you know but it's it was very risky. I recognize I was very scared when I did it. But basically putting our streaming voice models into a into a rich mini. Action, just say the word. Okay, maybe you can do something a bit fun. I want to get you know to improve my health overall and so I'm looking for a for a bro to go to the gym. Can you take the personality and the voice of a gym addict? Hey, I'm Logan, your gym bro. Let's crush those gains together. You ready to lift, sweat, and feel awesome? I've got your back, no excuses, just results. So Okay, this it's it's fine. It it it sounds more natural than it used to. But in both cases, you know I mean, we're still not there, right? The latency is still quite high. The ability to handle simultaneous speaking between the user and the system is not there. Intelligence starts to become much better and I think that's also why we see all this traction around voice intelligence because there are agents actually to whom we can give a voice and they can be useful. And also this is mostly just a glorified text model with a voice around it. And so you know, anything that is not in the text will not be able to be leveraged. And maybe I could ask one last time to increase a bit the volume. I think that could be even better. Um And so what does it take to get there? So, basically we had very nice presentation this morning by Samuel about cascaded systems. So, I'll go quickly. Speech-to-text, LLM, text-to-speech, that's the classic Cascade. In our case, we do streaming speech-to-text, streaming text-to-speech with voice cloning, semantic VAD, the classic stuff. So, latency, you know, getting to a fast conversation. So, we have a very fast TTS. So, that's the latency for our TTS compared to a few other models. But, you know, that means that just the TTS is still more than 200 milliseconds. While in a human conversation, you need the entire stack of understanding, producing an answer, and pronouncing it to be around 200 milliseconds. So, that will not None of this will allow to have conversations that sounds that sound human. And this is just latency for text conversation. There is no tool call, no actual task that is performed. Now, that's another scene from the from the movie. Uh Okay, let's start with your emails. You have several thousand emails regarding LA Weekly, but it looks like you haven't worked there in many years. Oh, yeah. Okay, so yeah, it just went instantly into for all the emails and gathered information and so on and so forth. Obviously, we are not there at all. Today, if you have a a voice agent that is supposed to use a tool, you're going to wait. The tool call, you know, we are fighting for latency of the TTS, trying to grab 10 milliseconds, 20 milliseconds, and then you have a tool call or open router that is going to have a latency between 500 milliseconds and 4 seconds. So, in a way, sometimes you feel like we're fighting for something that is not the most relevant anymore. I think now the main bottleneck is becoming the tool call, which is very unpredictable. And we should have models that are resilient to very complex tool calls. So, what we are uh you know, one solution to do that is to have fillers. So, basically, you know, your LLM it splits into two things. So, it sends a tool call. And while it waits for getting the result back, it can keep the conversation going in a natural way. and then it retrieves the result, and it tries to insert it back naturally in the conversation. So, just because I always do a live demos, I will do a very short improvised live demo with something that was live coded very quickly, so it may be go very wrong. But the idea here is to have like just a live coded travel agent. And that can when I ask it to retrieve places, it's going to try to find a nice thing to say about the location I'm going to. Hello. This is Colin from Wanderlust Travel. I see you're looking to book a trip for two people from April 10th to the 13th, 2026. Where would you like to travel to for your getaway? I want to go to Tokyo. Tokyo is such an incredible choice. It's a fascinating mix of ultra-modern skyscrapers and beautiful, peaceful shrines. I found some fantastic options for you. You could stay at the Fairmont for a sweet So, this needs polishing is the main idea, you know, is that while it was trying to gather things, it doesn't really know how long it will take to get the data. It tries to say a nice thing about about the place you're going to. So, that's one way to get latency that is going to be more controlled and more reliable even despite the complexity of the thing. But then we know that latency of cascaded system is inherently high, right? And so then what I hear a lot from customers, from investors, and so on is what about speech to speech? So, I think a big confusion is what is speech to speech? So, speech to speech is the idea that now instead of having the three blocks, you only have one that does everything together, right? So, instead of having audio that goes in speech to text and then LM text to speech, you have a model that takes speech as input and it's going to output speech. And that reduces latency a lot, but that's still not a human conversation. In particular, because every single speech to speech model except Moshi is half duplex. What that mean is even the best uh speech to speech model we could argue I don't know maybe that's the advanced voice model of Open AI or Cezanne AI I'm a huge fan of uh the Cezanne AI uh uh voice voice model is still half duplex. What it means is that the model is either listening or it's speaking. And it cannot handle the ambiguity of human conversation where you can have overlap between uh people speaking on one another. A lot of things happening. You cough. Uh you know, you just do mhm which is called back channeling and that it breaks completely. Full duplex that's a human conversation. It varies between cultures and languages but for example in Japanese it's a sign of politeness and that you are actively listening to do a lot of back channeling. So you say mhm uh uh constantly when the other person is speaking and you get up to 20% of the time that is overlapped between the people, right? So that's what makes a conversation human and so I showed this video in my hotel room this morning just to you know that's a speech to speech model but that's not a full duplex model and you see how it can it can become annoying. Hey, how's it going? Hey, I'm doing really well. Thanks for asking. How about you? How's your day going so far? Uh I'm great. Uh you know, I'm preparing to talk about voice AI and how far we are from um uh you know um from the Her movie and uh yeah, I just like to brainstorm a bit about it with you. Oh, that's a great topic. Yeah, I'd love to help you brainstorm. Are you thinking more Exactly. Yeah, I was thinking \>\> No, no, I didn't mean to interrupt. You know, I was just saying yeah, like that. You can just keep going. Uh you know, don't mind me. That's something I typically do. Uh no worries at all. Yeah, I was just going to say we could break it down into a few aspects like Yeah, exactly. \>\> please stop stop interrupting. You know, it's called back channeling. Uh humans do it all the time. It shows that you're just following the conversation that you don't you know, like interrupt you in your flow. Just just going. Uh got it. Thanks for letting me know. Yeah. Oh, come on. Okay, so we have basically I was a bit mean, right? That's you know, that's my point. It's not it's not an actual conversation, right? And it can become very annoying in particular. That's also why you know, most of the voice AI demos they are shot in an empty like a quiet room next to the phone and so on. A lot of things can break. So, what we did instead in our case was Uh sorry, where am I in my presentation? I'm right here with Mushi. was uh the forceful duplex system. So, here you'll see my uh co-founder Alex talking to it. Uh it's almost 2 year old now. I think still kind of uh aged well because what you'll see is um you know, they are going to talk on one another constantly and it's just fine. The the planet is Sirius 22. Can you plot a trajectory course to it, please? \>\> Yes, sir. Okay. How long is it going to take us to get there? \>\> it out. It's approximately 5 months to get there. Okay, that's that's not too bad. Uh do you think we have all we need on board the ship to start the mission? \>\> We have everything we need. Okay. So, you know, even when the model has guessed what you're going to say, it start answering before you're done. At the same time you can talk over it and it's not ignoring what you're saying. It's really you know, consider it afterwards and so on. And so you you have what is the most robust uh to this day uh conversational experience robust to noise, to a lot of people speaking and so on and so forth. But now if we compare it to uh uh to the her Do you know what I'm thinking right now? Well, I take it from your tone that you're challenging me. Maybe because you're curious how I work. Do you want to know how I work? Yeah, actually. So, maybe this was not very clear, but this snippet here it's the AI understanding that the character is a bit uncomfortable. So, that's paralinguistic understanding. You know, it's understanding all the cues that are come from the way people speak. Technically, that is in Mushi, that is in any speech to speech model because this information is not lost. However, if you don't exploit this information uh to make your model say relevant things, it's never going to exploit it. If you train it on the audio version of a instruct data set uh and it's just factual question answering, what why would it even try to capture this information? So, basically, Moshi, I think we we saw still the only full duplex model. Uh recently, Nvidia published the personal plex model based on it. Um what was great is still the the flow of it is just honestly impossible to match, I guess. Um it's conversational and uh and very robust. At the same time, some model was very stupid. So, basically, it was, you know, just useless. You could talk for for a few minutes and then it was a bit pointless, you know? Because it was not an agent, it had no tool call, no ability to do anything. Uh it's impossible to use in production something that is as no observability. You don't know if the it's very hard to detect if someone said something that should be not accepted and so on and so forth. And there was no real uh paralinguistic uh understanding. So, the main takeaway for me is uh we know that this nature of interactive model that are going to be full duplex, that are going to be um really the that will be the way to get to an interaction that is as natural as you would have with a human. But as long as we don't are able we are not able to give to this kind of very natural sounding uh models the same level of reliability intelligence, and personalization as cascaded systems, I don't see a path towards uh uh them, you know, replacing cascaded systems. So, I used to be really at war against cascaded systems. I think they are so practical and so convenient that the main challenge, honestly, I think this we solve that with the Moshi. Uh anybody who implements it and train it on beta data will have something that is sounds just indistinguishable from humans. But all of this is going all the challenge going to happen here. Uh a last point is the scalability. So now, let's say you have the best uh speech-to-speech model, okay? Uh it solves all the stuff that I've talked to talked about. If we take again the analogy from the movie, um you will talk to it maybe several hours a day, or it will always be on because, you know, it's on your computer when you work and you asynchronously ask things to it and so on and so forth. Um I'm not going to mention the cost of the API of our competitors, but voice is very expensive. Everybody in this room probably knows it. Uh the voice mode of most hyperscalers is run at a loss. It's a gigantic multimodal model, and they lose money every time they they you use it. But it's, you know, it's kind of a marketing thing and so on, it's fine. But now, if we want to make it an actual profitable product that people are using at a massive scale, just not going to work. And uh in particular, anyone who tried developing a consumer app with voice uh realized that LLM now is uh almost nothing in terms of cost. All the TTS, speech-to-text is very cheap as well. Uh diarization is affordable. Uh TTS is really what is going to consume most of the And we I saw people burning their fundraising in TTS bills, and they don't even get the opportunity to get their user base to grow. So, another aspect is privacy. The more you're going to open to your eyes, the more um you know, you'll want it to be more controlled and private and feel more comfortable that things are not shared publicly. In particular, uh we see now people with mythos being afraid that an any single database was going to be hacked in a few months. And so, you know, you will be more comfortable if all your private data is uh is local. And so, to solve that, our first step is uh Gradion Phonon. Uh it's uh on-device TTS. So, on-device means a lot of things. For some people, on-device means it runs on a gamer GPU. For us, on-device means it runs on a smartphone CPU. And so, it's a very small model, uh less than 100 million parameter, and for its size, it works quite well. It's better than all the existing on-device models. Kokoro is is a good one, but it doesn't have voice cloning, and I'm out of time, so I would just play a short demo if I can. Yes. Uh jeez, Morty, stop looking for a signal. Gradio phone app runs locally on the CPU, which means high fidelity neural speech right here without those intergalactic cloud government hacks. It is about simplicity with no servers and no waiting, just a smooth, quiet performance that stays on the device with local processing and total privacy. What else? Local processing? That sounds like the machine is making its own donut. Wait, if the CPU is doing all the work, can it sprinkle some Yeah, so this runs on the smartphone CPU, which means that you can use that to power, you know, any kind of voice application without paying a single cent of API fee. So, we open to private beta for this model. The goal for us is to allow people to create consumer apps with voice, basically, and that they are able to scale the usage without having to lose money on the on API. So, the conclusion, the path forward is for us, I have a strong strict opposition to some of our competitors say that voice is a commodity now. I think it's completely false. Voice is very challenging. The last mile is going to be the most difficult to solve, and for us, it's really about science and engineering, and it will get to us to the her movie, sorry. So, you can use us on gradium.ai, and if you want to join us to bring her to life. Yeah, that's another I I Now I'm using this analogy. No, yes, if you want to to to work on very exciting voice models, you can apply and and and join us. Thanks a lot for your attention. \>\> Woo!