Did you know that we have LinkedIn, Instagram and X accounts
that you can follow?

Hi {{first_name|everyone}},

Microsoft is trying to cut the waiting out of voice AI, and the most interesting part is a transcription model that starts guessing your words before you've finished the sentence.

We've also got a study on whether AI-assisted writing still feels like yours, tools for adding an agent to your product, getting a project online fast, and improving AI-search visibility, plus an animated sci-fi film made with AI.

Let's get right into it.

🤿 DEEP DIVE

The quick take

Transcript Starts Before You Finish Speaking

A lot of voice AI comes down to something pretty mundane.

Waiting.

Waiting for someone to finish talking. Waiting for the transcript. Waiting for the model to think. Then waiting again for the reply to become audio.

Microsoft AI is trying to squeeze that delay from both ends with three new models, and the transcription side is probably the most interesting here.

MAI-Transcribe-2-Streaming doesn't wait until you've finished speaking to start producing text. It begins generating early guesses, called partials, just over 100 milliseconds after receiving audio, then keeps revising them as more context arrives.

That means an agent doesn't necessarily need to wait for the whole sentence either.

It can potentially start reasoning or calling tools while you're still talking.

The transcript keeps changing

Those early words aren't treated as final.

The model produces a partial transcript, updates it as you continue speaking, then commits a stable version. Microsoft says it supports 60 languages with continuous automatic language detection.

On Artificial Analysis, it ranks first for accuracy on both final and partial transcripts. Microsoft also says its internal evaluations found words appeared in the transcript twice as fast as its closest competitor for things like live dictation and subtitling.

The introductory price is $0.54 per hour of audio through the end of the year.

Then the reply has to speak

The other half of the loop is MAI-Voice-2.1 and its faster sibling, MAI-Voice-2.1-Flash.

Both support 23 languages and 26 locales, with a single voice able to switch languages while keeping the same speaker identity and using a native accent in each language.

Flash is the speed-focused version. Microsoft says it can generate 45 seconds of audio with 150 milliseconds of end-to-end latency, with 55% faster model inference and pricing of $15 per million characters. The standard Voice 2.1 model costs $22 per million characters.

Both also support voice cloning from a few seconds of reference audio, with consent guardrails built in.

And I think the bigger idea here is the combination.

A voice agent has to hear, understand, decide and speak fast enough that the interaction still feels like a conversation. Faster transcription gives the system more time to start working. Faster speech generation reduces the delay at the other end.

So the saved milliseconds aren't really the product by themselves. They're extra room for the agent to do everything in between.

🤝 FROM OUR FRIENDS

A quick share we thought you might find useful

A third-generation LA roaster ships 50+ coffees — single origins, flavored favorites, decafs, even teas — days after roasting. Mix and match Keurig®-compatible boxes for tiered savings up to 34% off. First order? Take an extra 15% off, automatically.

Taste It Fresh

🔥 WHAT’S THE BUZZ

What people are building, sharing, and talking about

What I found interesting is that doing more with AI didn’t necessarily make the writing feel more like your own. Prompting and editing both seem to help with that, just in different ways, but even when people did both, writing without AI still felt more personally theirs.

⚒ TOOL SNAPSHOTS

Futuristic tools within AI, no-code, and productivity

🤖 AI MADE THIS

Interesting and inspiring creations made with AI

Palm Pets imagines genetically engineered animals becoming pocket-sized consumer products

Palm Pets imagines a biotech breakthrough that turns exotic animals into tiny pets you can hold in your hand. K.A. Betancourt built the 14-minute animated sci-fi film with a mix of AI image, video and music tools, creating an entire fictional consumer world around the idea. It’s cute on the surface, but things don't stay cute for long. The film is currently participating in the Higgsfield Global Film Festival.

AI used
Confirmed tools include Higgsfield Cinema Studio and Elements, Higgsfield Soul 2.0 with Moodboards, Nano Banana, GPT Image and Suno for the original soundtrack.

ℹ️ ABOUT US

The Intelligent Worker helps you to be more productive at work with AI, automation, no-code, and other technologies.

We like real, practical, and tangible use-cases and hate hand-wavy, theoretical, and abstract concepts that don’t drive real-world outcomes.

Our mission is to empower individuals, boost their productivity, and future-proof their careers.

We read all your comments - please provide your feedback!