--- title: Microsoft’s new voice AI can start working out what you... description: Microsoft introduced three new voice AI models in Microsoft Foundry on October 1: MAI-Transcribe-2-Streaming, MAI-Voice-2.1, and MAI-Voice-2.1-Flash. The... canonical: https://daily.dev/posts/microsoft-s-new-voice-ai-can-start-working-out-what-you-want-before-you-finish-speaking-tqpal1wrh twitter:card: summary_large_image twitter:site: @dailydotdev og:type: website og:site_name: daily.dev og:title: Microsoft’s new voice AI can start working out what you want before you finish speaking | daily.dev og:description: Microsoft introduced three new voice AI models in Microsoft Foundry on October 1: MAI-Transcribe-2-Streaming, MAI-Voice-2.1, and MAI-Voice-2.1-Flash. The... og:url: https://daily.dev/posts/microsoft-s-new-voice-ai-can-start-working-out-what-you-want-before-you-finish-speaking-tqpal1wrh og:image: https://api.daily.dev/og/posts/TQpAl1Wrh.png og:image:alt: Microsoft’s new voice AI can start working out what you want before you finish speaking og:image:width: 1200 og:image:height: 630 og:locale: en --- > ## Documentation Index > Fetch the complete documentation index at: https://daily.dev/llms.txt > Use this file to discover all available pages before exploring further. # Microsoft’s new voice AI can start working out what you want before you finish speaking **[The New Stack](https://daily.dev/sources/newstack)** · 6 min read · 1 upvotes · 0 comments ## Summary Microsoft introduced three new voice AI models in Microsoft Foundry on October 1: MAI-Transcribe-2-Streaming, MAI-Voice-2.1, and MAI-Voice-2.1-Flash. The streaming transcription model produces first-pass partial text within low hundreds of milliseconds, letting agents start reasoning before a sentence ends, and it debuts at #1 for accuracy on the Artificial Analysis leaderboard. Phonely CEO Will Bodewes calls the move a sign voice AI is ready for wider use, noting his own company hit 120-millisecond responses by owning its model queue. Phonely researcher Amit Kiran Rege counters that speech in and out aren't the hard part; the hard part is detecting silence and turn-taking, since cutting latency in half turned missed interruptions into audible ones two-thirds of the time. ## Full article daily.dev links to this article rather than hosting it. Read it at the original source: ## Questions this post answers ### What is MAI-Transcribe-2-Streaming and how fast does it produce transcripts? MAI-Transcribe-2-Streaming is a Microsoft speech recognition model in Microsoft Foundry that turns live speech into text as it's spoken. It produces initial partial hypotheses within the low hundreds of milliseconds of receiving audio, then refines them as more context arrives before committing to a stable transcript once the utterance ends. It debuted at #1 for accuracy on the Artificial Analysis leaderboard for both partial and final transcripts. _Teams building real-time voice agents can track streaming transcription benchmarks like this on daily.dev._ ### Why does reducing latency in a voice AI agent cause it to interrupt callers more often? Lower latency can turn silent, unheard interruptions into audible ones. Phonely found its agent jumped in too early two-thirds of the time, but those attempts were usually canceled silently because the caller resumed speaking before the agent's reply started. Cutting latency in half meant the agent's replies started sooner, so those same early decisions became audible interruptions instead of being masked. _Developers tuning voice agent latency weigh these turn-taking trade-offs daily, follow the space on daily.dev._ ### What's the hardest technical problem in building voice AI agents according to Phonely? The hardest problem isn't speech-to-text or text-to-speech, it's detecting turn-taking, specifically figuring out whether a caller has finished speaking or is just pausing mid-sentence. Phonely's machine learning researcher Amit Kiran Rege says Microsoft's new models address the two ends of the pipeline (speech in and speech out) but not this middle silence-detection problem, which Phonely encountered when scaling to millions of calls. _Anyone evaluating voice agent platforms can compare these architecture trade-offs on daily.dev._ ## Similar posts on daily.dev - [Voice Is Coming to AI Agents Faster Than Most People Realize](https://daily.dev/posts/voice-is-coming-to-ai-agents-faster-than-most-people-realize-sf61jdypa) · C\# Corner · 1 upvotes · 0 comments - [These execs think voice AI hasn’t reached its ChatGPT moment yet](https://daily.dev/posts/these-execs-think-voice-ai-hasn-t-reached-its-chatgpt-moment-yet-mb4zndoay) · TechCrunch · 0 upvotes · 0 comments --- Tags: [#ai-agents](https://daily.dev/tags/ai-agents), [#speech-recognition](https://daily.dev/tags/speech-recognition), [#text-to-speech](https://daily.dev/tags/text-to-speech), [#voice-ai](https://daily.dev/tags/voice-ai) [View this post on daily.dev](https://daily.dev/posts/microsoft-s-new-voice-ai-can-start-working-out-what-you-want-before-you-finish-speaking-tqpal1wrh) ```json {"@context":"https://schema.org","@graph":[{"@type":"Organization","@id":"https://daily.dev/#organization","name":"daily.dev","url":"https://daily.dev","logo":{"@type":"ImageObject","url":"https://daily.dev/apple-touch-icon.png","width":180,"height":180},"sameAs":["https://twitter.com/dailydotdev","https://github.com/dailydotdev","https://www.linkedin.com/company/daily-dev-ltd"]},{"@type":"WebSite","@id":"https://daily.dev/#website","url":"https://daily.dev","name":"daily.dev","publisher":{"@id":"https://daily.dev/#organization"},"potentialAction":{"@type":"SearchAction","target":{"@type":"EntryPoint","urlTemplate":"https://daily.dev/search?q={search_term_string}"},"query-input":"required name=search_term_string"}}]} {"@context":"https://schema.org","@type":"TechArticle","headline":"Microsoft’s new voice AI can start working out what you want before you finish speaking","url":"https://daily.dev/posts/microsoft-s-new-voice-ai-can-start-working-out-what-you-want-before-you-finish-speaking-tqpal1wrh","mainEntityOfPage":{"@type":"WebPage","@id":"https://daily.dev/posts/microsoft-s-new-voice-ai-can-start-working-out-what-you-want-before-you-finish-speaking-tqpal1wrh"},"datePublished":"2026-10-23T12:04:28.657Z","dateModified":"2026-10-23T12:05:06.336Z","description":"Microsoft introduced three new voice AI models in Microsoft Foundry on October 1: MAI-Transcribe-2-Streaming, MAI-Voice-2.1, and MAI-Voice-2.1-Flash. The...","image":"https://media.daily.dev/image/upload/f_auto,q_auto/v1/posts/43c4f603438a51ff4796739d6a5e1e89?_a=AQAEuop","thumbnailUrl":"https://media.daily.dev/image/upload/f_auto,q_auto/v1/posts/43c4f603438a51ff4796739d6a5e1e89?_a=AQAEuop","isAccessibleForFree":true,"articleSection":"The New Stack","inLanguage":"en","publisher":{"@type":"Organization","name":"daily.dev","url":"https://daily.dev","logo":{"@type":"ImageObject","url":"https://daily.dev/apple-touch-icon.png","width":180,"height":180}},"author":{"@type":"Organization","name":"The New Stack","logo":"https://media.daily.dev/image/upload/t_logo,f_auto/v1/logos/newstack","url":"https://daily.dev/sources/newstack"},"commentCount":0,"discussionUrl":"https://daily.dev/posts/microsoft-s-new-voice-ai-can-start-working-out-what-you-want-before-you-finish-speaking-tqpal1wrh","interactionStatistic":[{"@type":"InteractionCounter","interactionType":{"@type":"LikeAction"},"userInteractionCount":1},{"@type":"InteractionCounter","interactionType":{"@type":"CommentAction"},"userInteractionCount":0}],"keywords":"ai-agents,speech-recognition,text-to-speech,voice-ai","timeRequired":"PT6M"} {"@context":"https://schema.org","@type":"BreadcrumbList","itemListElement":[{"@type":"ListItem","position":1,"name":"Home","item":"https://daily.dev"},{"@type":"ListItem","position":2,"name":"The New Stack","item":"https://daily.dev/sources/newstack"},{"@type":"ListItem","position":3,"name":"Microsoft’s new voice AI can start working out what you want before you finish speaking"}]} {"@context":"https://schema.org","@type":"FAQPage","@id":"https://daily.dev/posts/microsoft-s-new-voice-ai-can-start-working-out-what-you-want-before-you-finish-speaking-tqpal1wrh#faq","mainEntity":[{"@type":"Question","name":"What is MAI-Transcribe-2-Streaming and how fast does it produce transcripts?","acceptedAnswer":{"@type":"Answer","text":"MAI-Transcribe-2-Streaming is a Microsoft speech recognition model in Microsoft Foundry that turns live speech into text as it's spoken. It produces initial partial hypotheses within the low hundreds of milliseconds of receiving audio, then refines them as more context arrives before committing to a stable transcript once the utterance ends. It debuted at #1 for accuracy on the Artificial Analysis leaderboard for both partial and final transcripts. Teams building real-time voice agents can track streaming transcription benchmarks like this on daily.dev."}},{"@type":"Question","name":"Why does reducing latency in a voice AI agent cause it to interrupt callers more often?","acceptedAnswer":{"@type":"Answer","text":"Lower latency can turn silent, unheard interruptions into audible ones. Phonely found its agent jumped in too early two-thirds of the time, but those attempts were usually canceled silently because the caller resumed speaking before the agent's reply started. Cutting latency in half meant the agent's replies started sooner, so those same early decisions became audible interruptions instead of being masked. Developers tuning voice agent latency weigh these turn-taking trade-offs daily, follow the space on daily.dev."}},{"@type":"Question","name":"What's the hardest technical problem in building voice AI agents according to Phonely?","acceptedAnswer":{"@type":"Answer","text":"The hardest problem isn't speech-to-text or text-to-speech, it's detecting turn-taking, specifically figuring out whether a caller has finished speaking or is just pausing mid-sentence. Phonely's machine learning researcher Amit Kiran Rege says Microsoft's new models address the two ends of the pipeline (speech in and speech out) but not this middle silence-detection problem, which Phonely encountered when scaling to millions of calls. Anyone evaluating voice agent platforms can compare these architecture trade-offs on daily.dev."}}]} ```