Tag: conversational AI

Prince Mario-Max Schaumburg-Lippe: Microsoft’s AI Voice Stack Transcribes 60 Languages Live

The Half-Second That Changes Everything About Talking to Machines

Here's the honest truth about voice AI: it's never been the intelligence that was the problem. Ask a voice agent a question and the model usually knows the answer. The problem is the wait. You speak, there's a pause, you start talking again, it talks over you. The whole exchange feels like a bad satellite interview, not a conversation.

Microsoft thinks it just fixed that. On October 1, the company's Microsoft AI division announced three new voice models, and the headline number is less than one second: a voice agent built entirely on Microsoft's first-party stack can now complete a full conversational turn in under a second. That's the threshold, according to TechTimes analysis, where an AI voice stops sounding like a walkie-talkie and starts sounding like a phone call.

What Actually Launched

The three models fill out a complete voice pipeline. MAI-Transcribe-2-Streaming is Microsoft's first streaming speech-to-text model — it listens and transcribes continuously, rather than waiting for a finished sentence. MAI-Voice-2.1 handles text-to-speech in 23 languages across 26 locales, and MAI-Voice-2.1-Flash is a lower-latency variant for situations where speed matters most.

All three are in public preview through Microsoft Foundry (alongside Azure AI Speech), plus MAI Playground and Vercel's AI Gateway.

The transcription specs are the part getting the most attention. MAI-Transcribe-2-Streaming supports 60 languages with automatic, continuous language detection — meaning a conversation can switch languages mid-sentence without anyone restarting a session. Interim transcript hypotheses arrive in roughly 100 milliseconds. And Microsoft says it ranks first for both partial and final transcription accuracy on the Artificial Analysis streaming benchmark: a 2.5% final word-error rate, 2.8% first-partial error rate, and 0.13 seconds to finalization. Those are vendor-reported numbers on a benchmark dated September 28, so independent confirmation is still pending — but even with that caveat, the performance bar is clearly high.

Why 60 Languages of Code-Switching Matters More Than Accuracy

Benchmarks get the headlines, but the more interesting feature is the language detection. Real conversations code-switch. A customer service call in Miami might drift between English and Spanish. A family call between generations in Manila might blend Tagalog and English in a single sentence. Most transcription systems handle this badly: you pick a language, and the model fights you when you stray.

Automatic, continuous detection treats multilingual speech the way people actually speak it. That matters enormously for accessibility tools, live translation, and the meetings business — the market where real-time transcription pays its rent. A transcript that follows the conversation instead of the settings menu is a small technical detail with outsized practical effect.

The voice side has its own interesting capability. MAI-Voice-2.1 can take one speaker identity and switch languages while keeping the same recognizable voice, with native accents and local phrasing. Voices can also be replicated from a few seconds of reference audio — consent-gated, Microsoft says, which is the right call given the obvious misuse potential of instant voice cloning.

Pricing That Signals a Land Grab

The pricing tells its own story. Transcription is offered at an introductory $0.54 per audio hour through the end of 2026. Voice generation costs $22 per million characters. That intro price on transcription isn't charity — it's Microsoft buying market share in a space where developers tend to build on whatever stack works first and stay there for years.

And there's the deeper strategic point the brief's research flagged: Microsoft is building its own speech stack rather than licensing one. That means the voice race is now about vertical integration, not just model quality. When the same company owns the models, the cloud, the benchmarks, and the developer platform, the moat isn't any single model — it's the whole pipeline working together.

Where This Goes Next

The obvious near-term winners are call centers, accessibility tools, and live translation services — anything where a half-second of latency is the difference between a product that works and one that frustrates. Longer term, the sub-second turn opens the door to voice agents that can genuinely hold a phone call: appointment booking, customer triage, language tutoring.

There's a broader infrastructure story underneath this, too. Real-time voice agents need serious compute behind them — the kind of capacity hyperscalers are racing to build, like the recent 400MW AI data center project in Japan and new efforts to squeeze more compute from the same power. Voice is latency-sensitive in a way batch transcription never was, and it will push demand for inference capacity closer to users.

Will Microsoft's #1 benchmark claim hold up under independent scrutiny? That's the question worth watching. But even discounting the marketing, the direction is clear: the age of the patient, slow voice bot is ending. The next generation answers before you've finished blinking. The question isn't whether voice agents will feel natural anymore — it's whether we're ready for machines that interrupt us in 60 languages.

The Takeaway

Voice was AI's weakest interface because latency, not intelligence, made it feel dumb. Microsoft's new stack attacks exactly that weakness — and the 60-language auto-detection might matter more than any benchmark. When the technology fades into the background, the conversation can finally begin.

Prince Mario-Max Schaumburg-Lippe: ElevenLabs Hits $22B Valuation, Launches Voice Model v4

Everyone spent the last two years arguing about which chatbot would win. Meanwhile, ElevenLabs went and proved the interface that actually matters is the one people have been using since Alexander Graham Bell: the phone call.

On September 30, the London-based AI voice company completed a $300 million employee tender offer that values it at $22 billion, twice the $11 billion valuation it carried after its $500 million Series D in February. That’s a doubling in seven months. The tender was co-led by Wellington and T. Rowe Price, with participation from existing backers Andreessen Horowitz and Lightspeed and new investors including EQT and Goldman Sachs.

And the same week, the company launched Eleven v4 and v4 Turbo, new text-to-speech models that natively speak, listen, and translate across more than 90 languages used by over 5.5 billion people, at sub-100-millisecond latency. Valuation doubling plus a flagship model launch in one week: that’s a company announcing it intends to own the category.

The number that explains the valuation

Forget the $22 billion for a second. The number that does the explaining is 15 million.

ElevenLabs says its voice agents now handle more than 15 million conversations every week, three times the level in February. And these aren’t demos. The agents are doing real enterprise work: processing refunds, renewing insurance policies, booking appointments. The client list includes Stripe, Deutsche Telekom, DoorDash’s SevenRooms, the insurer Admiral, and the sovereign governments of Ukraine and Greece.

That’s the tell. Chatbots got the headlines, but voice agents got the jobs. There’s a reason: for all the talk about conversational AI, the overwhelming majority of customer interactions at real businesses still happen by voice. The call center is the largest interface in commerce, and it runs on humans having the same conversations thousands of times a day. AI that can hold those conversations naturally, in 90-plus languages, with human-sounding expression, doesn’t need a market to be invented. The market is a phone line.

“We’re already seeing rapid adoption of expressive voice agents by enterprises and governments, who are deploying them in service of consumers and citizens,” said co-founder and CEO Mati Staniszewski. Founded in 2022 by Staniszewski and CTO Piotr Dabkowski, the company has gone from text-to-speech startup to one of Europe’s most valuable startups in four years. The velocity is the story.

Why voice is winning the agent race

There’s a thesis hiding in this valuation, and it’s worth spelling out: chat was the training wheels; voice is the vehicle.

Text chatbots had to teach users a new behavior. Voice agents meet users where they already are. Nobody needs onboarding to have a phone conversation. The elderly customer renewing an insurance policy, the traveler rebooking a flight, the citizen calling a government service line: they all already know how to talk. The AI just has to be good enough at listening and responding that the caller doesn’t notice the difference.

That last part is where the v4 models matter. Sub-100-millisecond latency is the threshold where conversation stops feeling like a walkie-talkie exchange and starts feeling like a person. Expressive speech, the pauses, the emphasis, the warmth, is what makes callers stay on the line instead of mashing zero for a human. ElevenLabs built its name on voice quality back when it was just a text-to-speech tool; now that quality is the moat around the agent business.

The multi-language angle is underrated too. Ninety languages spoken by 5.5 billion people means a single deployment can serve a global customer base without the traditional call-center model of staffing language queues. For governments and multinationals, that’s transformative. Ukraine and Greece using AI voice agents for citizen services is the kind of deployment that would have sounded like science fiction five years ago.

The tender offer tells its own story

One detail worth pausing on: this wasn’t a fundraise. A tender offer lets employees and existing shareholders sell stock to investors, providing liquidity without necessarily raising new capital for the company. The company didn’t need the money. Its people got paid.

That’s a retention weapon in the AI talent war. At $22 billion, with Goldman Sachs and T. Rowe Price buying in, ElevenLabs employees just got a very tangible reason to stay. In a market where top voice-AI researchers can name their price, keeping the team that built the thing is as important as any model release. The v4 launch the same week makes the message complete: we’re winning, we’re shipping, and we’re taking care of our people.

It’s also a signal about where smart money thinks the agent economy is going. The biggest AI investments of 2026 have flowed to agent developers as businesses race to automate customer service and back-office work. Voice is where that automation meets the customer directly. A $22 billion bet says the phone call is not a legacy channel to be replaced. It’s the channel to be upgraded.

What this means for the rest of us

A few practical readouts.

Expect to talk to AI more, and notice it less. Fifteen million conversations a week is still a rounding error against global call volume, but the growth rate is the thing. At 3x in seven months, the crossover point where a meaningful share of routine calls are AI-handled is closer than most people think. The good news: done well, it means no more hold music for a refund.

Voice quality is now a competitive dimension. If you’re building anything customer-facing with AI, the voice matters as much as the brain. The companies winning in this space compete on latency and expressiveness, not just accuracy. Users forgive a slightly wrong answer delivered warmly faster than a correct one delivered like a robot.

The “AI takes jobs” framing misses the point here. The calls being automated are the ones nobody wanted to staff: repetitive, high-volume, emotionally draining. The humans move up to the exceptions, the edge cases, the moments that actually need judgment. That’s been the pattern with every automation wave, and voice AI looks like it’s following the script.

Watch the government angle. Sovereign deployments in Ukraine and Greece are the leading edge of AI in public services. Multilingual, always-available, consistent: it’s a strong pitch for citizen services. Expect more governments to follow, and expect the procurement debates to be lively.

The bigger picture

The chatbot era taught the industry that people will talk to AI. The voice era is teaching it something more valuable: people will talk to AI the way they talk to people, about the boring stuff that keeps businesses running. Refunds. Renewals. Appointments. Fifteen million times a week.

ElevenLabs doubled its valuation in seven months because it found the biggest, most familiar interface in the world and made AI fluent in it. The phone call survived the internet, the smartphone, and the chatbot. Now it’s getting an upgrade.

If all this talk of conversation has you craving the real, unscripted kind, here’s what’s on around New York this week, from jazz nights to night markets. And if you’re the type who does their best thinking over something spicy, the city’s hottest chicken spots are ready when you are.

Prince Mario-Max Schaumburg-Lippe: AI Search Up 200%: Shoppers Now Start Trips in AI Chat

Remember the last time you started a shopping trip by opening a search engine and typing “best running shoes under $150”? That habit is fading fast. A lot of shoppers now open an AI chat first and just talk it through: “I need shoes for marathon training, wide fit, under $150, and I have flat feet.” The results feel closer to what they’d actually buy, and they got there with one question instead of twenty tabs.

Salesforce put hard numbers on that shift this morning. The fourth edition of its State of Commerce report, released September 30, found that agentic search (using an AI assistant as the first step of the shopping journey) grew 200% year over year. It’s a big sample, too: 3,450 commerce professionals (100 of them in Singapore), a consumer survey of 4,690 people, and behavioral data from more than 1.5 billion global shoppers.

The message from retailers? Expectations keep climbing. Eighty-six percent of commerce leaders say AI is raising customer expectations, and 41% say meeting them is harder than ever. Shoppers have tasted what a good AI concierge feels like, and they want it everywhere.

Consumers Moved First; Retailers Are Scrambling to Follow

Here’s the gap that defines this whole story. Only 28% of commerce organizations actually use agentic AI today. But another 52% plan to adopt within six months. Consumers have already changed their behavior; most retailers are still building the thing that serves it.

That’s a classic adoption lag, and it explains why this moment feels so charged. The shoppers are already in the chat. The stores are still figuring out how to be there.

In Asia-Pacific the picture is a little further along. Only 5% of adopters in APAC are still piloting; the largest share, 35%, are prioritizing scaling agentic AI across functions. Scaling, not experimenting. That’s worth noticing: it suggests the experimentation phase is quietly over in the region’s more aggressive markets, and the build-out phase has begun.

Why the Chat Became the Storefront

Think about what a search bar asks of you. Keywords. Filters. Sorting. Page after page of near-identical listings. An AI chat flips the relationship: you describe what you want in plain language, and it does the hunting.

This matters beyond convenience. A conversation is a richer signal than a keyword. When a shopper says “I want a gift for my dad who’s just getting into gardening and hates fiddly tools,” an agent can weigh intent, budget, and taste in a way that “gardening gifts men” never could. The winners in this shift won’t be the retailers with the most SKUs; they’ll be the ones whose AI actually understands what “not fiddly” means.

There’s a real succession happening here. The search bar was the front door of online shopping for twenty years. The chat window is applying for the job, and the +200% number is its résumé. Retail SEO as we know it (keywords, snippets, ranking for “best wireless earbuds”) is giving way to something new: making sure an AI assistant recommends you when a shopper asks.

What This Means If You Sell Things

Three practical takeaways from the report:

Your homepage might be a conversation now. If agentic search is the first step of the journey, the thing a shopper meets first may be an LLM interface, not your website. Retailers need to make their product catalogs legible to AI agents: clean product data, honest availability, prices an agent can trust. The store that feeds the chat well wins the sale.

Personalization is the expectation, not the feature. Eighty-six percent of leaders say AI is raising the bar, and customers can feel the difference between a generic chatbot and one that remembers their size, their budget, their last purchase. Investing in the data behind the chat matters more than the chat itself.

Start where your customers already are. The 52% planning adoption should take heart: the biggest risk isn’t moving too early, it’s the gap between consumer behavior and merchant readiness. Shoppers aren’t waiting for permission.

The broader tech backdrop makes this feel inevitable. AI is quietly rewiring how machines do physical and digital work alike: autonomous trucking routes are being planned by algorithms, robotaxis are multiplying across Texas. Agentic commerce is the same wave hitting the checkout button.

The Optimist’s Take

Strip away the jargon and this is a consumer-benefit story. Less friction. Fewer dead-end searches. Recommendations that understand context instead of just matching keywords. A shopper with specific needs, dietary restrictions, accessibility requirements, a tight budget, gets a personal shopper for free, one that never gets tired and never judges.

Is the technology perfect? Of course not. Agents still hallucinate prices and botch availability, and retailers will have a rough year learning to feed them clean data. But the direction is unmistakable. Two hundred percent growth doesn’t lie: the chat is the new front door of shopping, and it’s open.