Mato Voice is out | Sep 16
Show notes
What the episode covers
Mato Voice, Mato's new in-house text-to-speech model, is live, and the launch post runs a blind-style comparison against Gemini and ElevenLabs using one shared script.
Blake and Elena Reyes unpack what the launch post shows and what it leaves out, including the build-versus-license tradeoff, ElevenLabs' 2026 model reshuffle, and what permissioned cloning actually requires for creators publishing on Mato.
- Mato Voice launches with permissioned cloning, original voice design, and guided Studio sessions.
- Gemini's Speech Arena Elo ranking (dated April 15, 2026) and its 24-language speaker handoffs are compared against Mato's blind test results.
- ElevenLabs deprecated Turbo v2/v2.5 for Flash v2.5, now recommends v3 for finished production, and splits cloning into a $6 Instant tier versus Professional Voice Cloning.
- Wikipedia reports ElevenLabs raised $500 million at an $11 billion valuation with IPO chatter and a $1 billion voice-restoration pledge.
- Guided Studio sessions move intro/outro work rather than remove it, and the cost is a different bill, not automatic savings.
Week 39, 2026
Full changelog and posts at heymato.com.
📣 We Want to Hear from You!
Timeline
In this episode
7 moments worth skipping to. The timecodes match the player above.
- 0:15Introduction
- 1:28What Mato Voice actually is
- 4:36How Gemini says it stacks up
- 7:23The turn: ElevenLabs isn't standing still either
- 11:22What the launch post left out
- 15:18What this means for anyone publishing on Mato
- 19:01Outro
Quick answers
Straight from the episode
The questions this one settles, without the listen.
- What is Mato Voice?
- Mato Voice is Mato's own in-house voice model for podcast conversation. It includes permissioned voice cloning, original voice design, and guided Studio sessions, and was launched with a blind-style test comparing one script read by Gemini, ElevenLabs, and Mato Voice back to back.
- How does Gemini's speech model compare in the test?
- Gemini pitches itself around multi-speaker podcast use, cites a Speech Arena Elo blind-comparison ranking dated April 15, 2026, offers refined speaker handoffs across 24 languages, and partners with Wondercraft via Convo Mode and Director Mode. However, the hosts noted a blind test on a clean script may not predict how it performs in a messy real interview.
- What changes has ElevenLabs made to its voice models recently?
- ElevenLabs deprecated Turbo v2/v2.5 in favor of Flash v2.5, now recommends v3 for finished podcast or audiobook production while reserving Flash for live voice agents, and splits cloning into a $6 Instant Voice Cloning starter tier and a longer, more demanding Professional Voice Cloning path.
- Is Mato Voice cheaper than using a professional voice actor?
- Not necessarily. The hosts concluded that guided Studio sessions move the intro/outro work rather than eliminate it, so it's a different kind of cost (your time versus a studio fee) rather than a straightforward savings. Some creators will still choose to book a professional actor.
- What did Mato's launch post leave out about ElevenLabs?
- The launch post only compares voice quality and doesn't mention ElevenLabs' company scale or roadmap, including a reported $500 million raise at an $11 billion valuation, IPO chatter from February 2026, a $1 billion free voice-restoration pledge from March 2026, and features like a browser timeline editor, a transcript generator, and 32-language Professional Voice Cloning.
- Does the Mato Voice launch require existing publishers to change their workflow?
- No. The hosts concluded that nothing in the launch forces a workflow change for existing Mato publishers this week, since permissioned cloning still requires an explicit consent conversation and Studio sessions simply relocate existing production work rather than remove it.
Transcript
The full conversation
Every word of the episode, 3,476 of them, in the order they were said.
Read the transcriptHide the transcript
Elena ReyesOkay, wait, so you actually tried it, Mato Voice, the new model?
BlakeLast night. Didn't sleep great after.
Elena ReyesThat bad?
BlakeThat good. This is Made on Mato. I'm Blake with Elena Reyes, and I need you to understand, I did not expect to feel anything about a text-to-speech model.
Elena ReyesYou felt something?
BlakeI felt something.
Elena ReyesOkay, now I have to know what broke you.
BlakeNot yet. There's a comparison in the launch post that I want to get into first. I'll tell you what got me by the time we land this.
Elena ReyesAgainst who?
BlakeTwo names you'd guess.
Elena ReyesGemini and ElevenLabs.
BlakeMm.
Elena ReyesSee, I want to talk about ElevenLabs specifically because they have not exactly been sitting still this year.
BlakeMoney moves?
Elena ReyesModel moves, lineup moves. We'll get there.
BlakeThere's a whole reshuffle over there this year. New models, deprecated ones, funding news, and it's kind of wild timing given what Mato just did.
Elena ReyesNow you've got me properly curious, but that's a segment away, not right now.
BlakeOkay, but first.
Elena ReyesFirst, what it actually does.
BlakeRight. No framing, no reading the blog post to each other.
Elena ReyesJust the thing.
BlakeJust the thing.
Elena ReyesSo Mato Voice is exactly what it sounds like. Mato built its own voice model just for podcast conversation, not licensing, not reselling, their own thing.
BlakeWhich is a bigger swing than it sounds like.
Elena ReyesRight?
BlakeI mean, most of these companies just bolt somebody else's engine onto their product and call it a feature.
Elena ReyesTotally. And the launch post doesn't just tell you it sounds good. It makes you sit through a test. Same script three times. Gemini reads it, then ElevenLabs, then Mato Voice back to back.
BlakeWhich is smart, honestly. You can't argue with your own ears the way you can argue with a paragraph of marketing copy.
Elena ReyesExactly. No adjectives, just, "Here, listen."
BlakeOkay, but does it actually hold up?
Elena ReyesWe'll get there. There's more under the hood too. Permissioned cloning is in there.
BlakeMeaning you can clone a real voice, but only with that person signing off.
Elena ReyesRight, and then there's original voice design. So you're not cloning anyone. You're building a voice from scratch.
BlakeHuh.
Elena ReyesAnd guided studio sessions on top of that, which sounds like the onboarding layer walking you through actually setting a voice up.
BlakeSo it's not one feature. It's three things stacked together.
Elena ReyesYeah. Cloning, design, and the guided setup.
BlakeHere's what I keep chewing on, though. Why build the model at all? Licensing ElevenLabs or Gemini's voice stuff and slapping a UI on it is so much cheaper.
Elena ReyesCheaper, faster, less risk.
BlakeRight. So either they think the existing models are wrong for podcast audio specifically, or they just wanted to own the whole thing top to bottom.
Elena ReyesI'd guess it's the first one. Podcast conversation isn't the same as, like, an audiobook narrator or a customer service bot.
BlakeSure, but that's a bet. If the model's mediocre, they own that mediocrity now. No vendor to blame.
Elena ReyesThat's the risk of doing it yourself, yeah.
BlakeI'm not saying it's wrong. I'm saying it's a real decision with a real downside if it doesn't land.
Elena ReyesFair. But think about the upside. They control the roadmap, the pricing, all of it forever.
BlakeAssuming it's actually good.
Elena ReyesAssuming it's actually good.
BlakeThink about it from the cost side too, though. If you're licensing someone else's model, every voice read you generate is basically a metered cost forever, and their pricing can change whenever they want.
Elena ReyesRight. Whereas if you own the model, the marginal cost of generating one more minute of audio is basically whatever your own servers cost you.
BlakeExactly. And there's a compounding thing too. If the vendor discontinues the exact model you built your workflow around, you're stuck migrating everything on their timeline, not yours.
Elena ReyesThat actually happens. Companies retire model versions all the time and tell everyone to just move to the new one.
BlakeRight. So owning it means Mato controls that migration schedule instead of getting stuck by surprise.
Elena ReyesOkay, that's a real argument for building in-house. I'll give you that one.
BlakeBut it cuts both ways. Now, if Mato's model needs an upgrade, that's Mato's engineering time, not somebody else's.
Elena ReyesSure. But that's true of literally any core piece of your product you decide matters enough to own.
BlakeWhich brings us back to that comparison they set up.
Elena ReyesRight. And before we say whether Mato Voice wins that comparison, it matters what it's actually being measured against.
BlakeBecause Gemini's voice work and ElevenLabs aren't the same kind of competitor.
Elena ReyesNot even close. So let's look at what each one's actually bringing to that test. Okay, so Gemini, DeepMind's own page for this stuff, the speech generation models, says you can build multi-speaker podcasts, interviews, dialogue, distinct character voices, basically the same pitch as Mato Voice.
BlakeRight, but pitch is the word. Anyone can say distinct character voices. Doesn't mean it sounds distinct.
Elena ReyesSure, but they've got a number to back it. There's an Elo-based blind comparison thing they run. They call it Speech Arena. It's how they rank naturalness against other voices.
BlakeElo like chess ratings.
Elena ReyesYeah, exactly. Same math. People listen blind, pick which one sounds more human, rating goes up or down.
BlakeOkay. As of when, though? Because a leaderboard from a year ago is basically fiction in this field.
Elena ReyesThey date it April 15th, 2026 on that page.
BlakeSo five months old by the time we're talking about it. Not nothing, but this space moves fast enough that I wouldn't die on that hill.
Elena ReyesFair. There's a separate blog post too, a Google one, that gets more specific. Says they refine the TTS model for smoother handoffs between speakers.
BlakeHandoffs, meaning the awkward part where one voice cuts into another.
Elena ReyesRight, that transition. And they say it holds up across 24 languages, consistent character voices in each one.
Blake24 is a lot to claim consistency across.
Elena ReyesAnd there's a partner mentioned, Wondercraft. The post says Wondercraft runs on Gemini TTS for two of its features, something called Convo Mode and Director Mode.
BlakeNever heard of either.
Elena ReyesMe neither. But the point is, someone's actually shipping product on top of this, not just running a demo.
BlakeHere's my problem with all of it, though. An Elo ranking on a scripted line is a controlled room. My actual test is a forty-minute interview where somebody stumbles, laughs, talks over the other person.
Elena ReyesYou think the blind test doesn't transfer?
BlakeI think it tells you who wins when the material is clean. It doesn't tell you who wins when it's messy, which is every real episode.
Elena ReyesYeah. No scripted demo has ever had someone go, "Wait, sorry, what was I saying?"
BlakeExactly. So take the ranking as a floor, not a verdict.
Elena ReyesNoted. Short pause. Okay, before we get into ElevenLabs, quick word from the people who actually made this show possible.
BlakeMato.
Elena ReyesMato, yeah. If you've been thinking about launching a podcast for your company or your brand, now's a pretty good time to actually do it.
BlakeFor a limited time, when you sign up for Mato, they'll throw in a custom cover design, white glove onboarding, your own custom jingle, and a spot on their promotional list.
Elena ReyesThat's a package worth five thousand dollars.
BlakeHead to hiimato.com to get started.
Elena ReyesOkay, so that's Gemini's case for itself. Solid on paper, untested on the kind of conversation we actually have.
BlakeAnd Gemini's not even the name most people think of first in this space.
Elena ReyesNo. They think ElevenLabs.
BlakeWhich is the other name we teased at the top, and that company has not exactly been quiet this year. Elena, ElevenLabs spent this year rebuilding the engine room while everyone was watching the front door.
Elena ReyesWhat's that mean exactly?
BlakeThey killed off Turbo V2 and V2.5. Both of them gone, replaced by Flash V2.5.
Elena ReyesWait, they just retired two models?
BlakeYep, and now they're telling people, "Don't use Flash for a podcast or an audiobook."
Elena ReyesThen what's Flash even for?
BlakeIf you're making an actual finished episode, they want you on V3, and Flash is for live voice agents and real-time conversations.
Elena ReyesSo it's two lanes. Fast lane for live chat, quality lane for anything you're actually going to publish.
BlakeRight. Tells you something about where they think the real demand sits.
Elena ReyesThat they don't trust one model to do both jobs well.
BlakeOr that agent builders complained loud enough they had to split it.
Elena ReyesSure. Let's be generous about it.
BlakeThere's a cloning split too.
Elena ReyesGo on.
BlakeInstant voice cloning starts on their Starter tier, six bucks a month, and Professional voice cloning needs way more training audio.
Elena ReyesSix dollars? That's basically nothing.
BlakeFor a rough clone, sure. The Professional version is a different animal. Hours of material, way more setup.
Elena ReyesSo the cheap tier gets you close enough. The expensive one gets you the real thing.
BlakeBasically.
Elena ReyesSo walk me through the actual decision for someone running a company podcast, though. Say you're producing four episodes a month for a brand, and you're picking between Flash and V3.
BlakeIf you're publishing finished episodes, ElevenLabs is telling you flat out to use V3. Flash is explicitly not for that job anymore.
Elena ReyesSo even if Flash is cheaper, you'd be using the tool they're telling you not to use for the actual product you ship.
BlakeRight. And that's a different posture than most vendors take. Usually, it's use whichever, we don't care. Here, they're drawing a hard line between live and finished.
Elena ReyesWhich tells you they've probably seen people using Flash for finished episodes and getting a worse result.
BlakeThat would be my guess, yeah. Though the review doesn't spell out why they drew the line, just that they did.
Elena ReyesFair. We shouldn't invent the reason, but practically it means the decision tree got simpler for a publisher. Live agent, use Flash. Actual episode, use V3. No in-between anymore.
BlakeAnd on the cloning side, six dollars gets you instant cloning. But if you actually want a professional grade clone of your own voice for regular use, that's the tier that needs a lot more training audio to get right.
Elena ReyesSo the six dollar tier is more of a let's see if this is even useful tier.
BlakeThat's how I'd read it. It's not the tier you'd build a permanent show voice on.
Elena ReyesOkay. Here's what I keep chewing on. Mato just put its own voice model out into a market where the biggest name in the space renamed half its lineup this same year.
BlakeRight.
Elena ReyesIs that good timing or bad timing?
BlakeI'd call it good timing. If the leader's still moving pieces around, there's no fixed target to lose to.
Elena ReyesI don't buy it. If they're this deep into rebuilding, they clearly learned something about where the demand's actually heading. Mato's betting against people who already ran that experiment.
BlakeOr Mato skips the years they spent finding out Turbo wasn't good enough.
Elena ReyesThat's a generous way to describe catching up.
BlakeI'm not saying it's settled. I'm saying it's not an obvious loss either.
Elena ReyesHere's a concrete version of it, though. If I'm a small show weighing Mato Voice against ElevenLabs today, and ElevenLabs is mid-reshuffle, do I wait to see where they land, or do I just go with whoever's stable right now?
BlakeI'd go with whoever's stable right now. A moving target vendor is a worse pick for a production workflow than a slightly behind but settled one.
Elena ReyesEven if the settled one might be worse a year from now.
BlakeEven then, because you can always migrate later. What you can't get back is the time you spent rebuilding your workflow twice because you picked the vendor mid-transition.
Elena ReyesOkay, that's the first thing you've said today I fully agree with.
BlakeDon't get used to it.
Elena ReyesWe're not deciding this one today.
BlakeNope. But there's a piece of this whole story the Mato launch post skipped completely.
Elena ReyesOkay, so Blake, you said there's stuff ElevenLabs never put in front of Mato's launch post. Spill it.
BlakeWikipedia's page on ElevenLabs has this whole section Mato conveniently skipped. ElevenLabs raised five hundred million dollars at an eleven billion dollar valuation back in February, and there's IPO talk on top of that.
Elena ReyesWait, eleven billion? That's not a startup anymore. That's just a company-shaped mountain of money.
BlakeRight. And it gets stranger. They also promised a billion dollars in free voice restoration tech to a million people who've permanently lost their voice.
Elena ReyesHold on. A billion dollars in free tech? Like, actually free, not free trial free?
BlakeThat's what the page says. March announcement.
Elena ReyesOkay. That's genuinely a nice thing to do. I'll give them that.
BlakeSure. But none of that, the raise, the IPO chatter, the restoration pledge, shows up anywhere in Mato's post.
Elena ReyesWell, why would it? It's their launch post, not ElevenLabs' quarterly filing.
BlakeFair. But it changes how you read the comparison. Mato's framing ElevenLabs as what? The incumbent to beat on quality. Incumbents with eleven billion dollars behind them don't stay still for long.
Elena ReyesYeah, I mean, come on, they're clearly not hurting for cash to throw at the next model.
BlakeHere's the thing that actually matters for somebody building a show on any of these platforms though. When a company's raised that much money and there's IPO chatter, the pressure changes. Growth and margins start mattering in ways they didn't when it was just a smaller startup.
Elena ReyesSo it's not that the product gets worse tomorrow, it's that the incentives shift.
BlakeRight. Pricing tends to firm up. Free tiers tend to shrink. That's just the pattern with well-funded companies as they mature.
Elena ReyesIs there anything in the Wikipedia entry that says ElevenLabs is actually doing any of that though? Or are we just speculating?
BlakeWe're speculating. The entry just gives us the raise, the valuation, and the IPO chatter. It doesn't say anything about pricing changes.
Elena ReyesGood, let's be clear about that then. We're flagging a pattern, not reporting a change that's already happened.
BlakeExactly. And to be fair, the voice restoration pledge cuts the other way a little. A company squeezing every dollar out of customers doesn't usually also promise a billion dollars in free access to people who need it.
Elena ReyesTrue. Though a big company can do both a margin push and a goodwill program at the same time. They're not mutually exclusive.
BlakeThere's more too. Their own podcast page talks about a timeline editor in the browser, a transcript generator that pulls straight from Spotify or Apple Podcasts links, and professional voice cloning that lets you speak in thirty-two languages in your own cloned voice.
Elena ReyesWait, thirty-two? Didn't we already say something in the twenties for one of these?
BlakeDifferent feature, different number. That one's the cloning tier, not the multi-speaker handoffs thing.
Elena ReyesGot it. So they've got the editor, the transcripts, the cloning. That's basically a whole publishing workflow, not just the voice model.
BlakeWhich is the thing the launch post doesn't grapple with. Mato's pitching a voice. ElevenLabs is pitching the whole pipeline around the voice.
Elena ReyesIs that a knock on Mato though? Or is that just different product, different day one?
BlakeI'd say it's neither, and Mato didn't have to mention it. I just think listeners should know the comparison in that post is voice quality only. It's not comparing company size or roadmap or what happens if ElevenLabs actually goes public.
Elena ReyesRight, right. Nobody puts an eleven billion dollar valuation and IPO chatter in a launch post right next to a link inviting you to go try the free comparison.
BlakeThey don't have to. It's not dishonest. It's just incomplete on purpose.
Elena ReyesSure. Just know what you're actually weighing when you listen to that comparison.
BlakeRight. That's the whole of what got left out.
Elena ReyesOkay, so if you're actually booking a guest on Mato this week, what do you do with the cloning piece?
BlakeYou ask them first, out loud, on the record, before you touch record on anything.
Elena ReyesNot a checkbox buried in a form.
BlakeRight. Because a checkbox doesn't tell you if the guest actually understood what they agreed to.
Elena ReyesSo it's basically just being a decent human and asking.
BlakePretty much. Say, "Hey, we might clone your voice for the intro. Are you good with that?"
Elena ReyesAnd if they say no?
BlakeThen you don't have a voice clone. You have a normal recording. The sky doesn't fall.
Elena ReyesWait, so this is less a feature and more a policy you have to actually build into your prep call.
BlakeThat's the part the launch post skips. It shows you the toggle. It doesn't show you the conversation you need to have before you flip it.
Elena ReyesWhich honestly is the boring part nobody wants to write a blog about.
BlakeExactly the part that gets people sued though.
Elena ReyesFair.
BlakeSo step one, permissioned cloning means permission is a conversation, not a setting.
Elena ReyesGot it. Okay, second thing, the guided studio sessions. Does that mean I don't need a voice actor for my intro anymore?
BlakeDepends what you're using the voice actor for.
Elena ReyesI mean the ten seconds where a nice voice says the show name and welcomes people in.
BlakeIf it's just that, a studio session probably gets you there. You're recording your own read, guided, cleaned up.
Elena ReyesRight. So it's not replacing the actor, it's replacing the studio booking.
BlakeKind of. The work doesn't disappear, it moves. Somebody still has to write the read, still has to sit through a session and get the take right.
Elena ReyesIs that actually cheaper though, or just different?
BlakeDifferent, mostly. You're trading a studio session fee for your own time sitting through Mato's guided flow. If your time's worth more than what you'd pay a voice actor for thirty seconds, you haven't saved anything.
Elena ReyesRight. It's not free. It's just a different bill.
BlakeExactly. Some people will still book the actor because they want a voice that isn't theirs, or because they'd rather pay money than spend an afternoon in a studio session.
Elena ReyesSo the honest pitch is you have the option now, not you should switch.
BlakeRight. Nobody's obligated to prefer their own voice over a professional one just because the tool exists.
Elena ReyesSo less you don't need a voice actor, more you don't need to hire one for thirty seconds of copy.
BlakeRight. And Mato's own post is pretty clear. This whole thing was built around podcast conversation specifically. Not some general text-to-speech tool with a podcast skin slapped on top.
Elena ReyesMeaning what? Practically?
BlakeMeaning the sessions are built for back and forth, for two people talking, not for reading a paragraph into a microphone alone.
Elena ReyesThat tracks with everything else we've talked about today, honestly.
BlakeIt does.
Elena ReyesSo permission is a conversation. Studio moves the work instead of erasing it. What's the actual takeaway for somebody hitting publish Thursday?
BlakeNothing about your workflow breaks this week. Nobody's making you clone anybody. Nobody's making you touch Studio.
Elena ReyesIt's there if you want it.
BlakeRight. And if you don't, your show sounds exactly like it did last week.
Elena ReyesWhich given everything we just went through with Gemini and ElevenLabs feels like the actual headline. Nothing forces your hand.
BlakeNothing forces your hand. That's genuinely the whole update.
Elena ReyesSix segments to get to nothing forces your hand.
BlakeWelcome to the Changelog.
Elena ReyesOkay, so that's the Changelog and the blog post for this week.
BlakeNext week, we'll see if anybody's actually used the clone toggle.
Elena ReyesOr if it just sits there looking pretty. So that's the whole picture this week. Mato Voice, Gemini, ElevenLabs, all of it.
BlakeAnd none of it changes what you do on Monday.
Elena ReyesWhich is kind of the point.
BlakeActually, remember what I promised up top? What actually broke me? It was that exact three-way script comparison. Same lines, three reads, and somewhere in the Mato Voice read, I just stopped noticing it was synthetic. That's the whole thing. That's what kept me up.
Elena ReyesYeah, go listen before you decide who you believe. That's kind of the whole point of the post.
BlakeFull changelog and both blog posts are up at himato.com if you want the details we skipped.
Elena ReyesAnd hey, if you're running a show on Mato and want to come talk to us about it, email hello@himato.com.
BlakeWe mean that. We read those.
Elena ReyesSubscribe wherever you're listening so you catch the next one.
BlakeAnd if you've got thoughts on whether Mato should have built its own model instead of licensing, that email address is where that goes to.
Elena ReyesWe'll actually read that one out loud on air if it's good.
BlakeAll right.
Elena ReyesI'm still gonna go mess with the voice designer tonight, not gonna lie.
BlakeReport back.
More episodes
Keep listening
Other episodes of Made on Mato, newest first.
- MCP access ships for AI assistants | Sep 14Sep 15, 2026 · 19 min
- Mato takes Mucker Capital funding | Sep 8Sep 8, 2026 · 18 min
- New embed player ships | Sep 1Sep 1, 2026 · 19 min
- The Blueprint Engine Explains Itself | Aug 24Aug 26, 2026 · 19 min
Sources
Where this came from
6 reports behind the episode. Every one of them opens where it was published.
- Introducing Mato Voice: what it is and how it soundsheymato.com
- ElevenLabs Voice Cloning Review 2026: v3, Scribe & Agents | Covalcoval.ai
- ElevenLabsen.wikipedia.org
- Gemini Audio – Speech generation — Google DeepMinddeepmind.google
- Improving Gemini Text-to-Speech models for better control and capabilitiesblog.google
- Podcast editing and voice generation with AIelevenlabs.io
