hotwirednews Get GrokBot

Week 21

18–24 May 20262026-W21 · 10 filings · 10 companies · 4 net-new

The Speak, Connect and Use layers led the week, with 2 filings each.

Seevision input1

ApplePlatformlaunch2026-05-19net-new

Apple previews VoiceOver and Magnifier camera Q&A with Apple Intelligence

Puts camera Q&A and spoken Magnifier control inside system VoiceOver/Magnifier with on-device Apple Intelligence — a later-2026 ship date, not a same-day GA — alongside the same-week Hyro×Five9 healthcare voice backfill (2026-05-20-hyro-five9-ai-agents).

Also Use

Details2 sources

Apple previewed Apple Intelligence upgrades for VoiceOver and Magnifier: Action-button camera questions with spoken follow-ups, richer Image Explorer descriptions, and spoken Magnifier controls, coming later in 2026.

VoiceOver / Magnifier Live Recognition with Apple Intelligence

Speaktext-to-speech2

TalkativePlatformlaunch2026-05-21net-new

Talkative adds pronunciation dictionaries to Voice AI, with IPA rules and spoken substitutions

Pronunciation control exists only on the cascaded pipeline, so a Talkative customer who switches an agent to speech-to-speech mode loses it; Google Text-to-Speech is the one engine the update names, and only to warn that IPA rules may not hold.

Details2 sources

Talkative customers can attach a dictionary of IPA pronunciations or text substitutions to a voice agent's text-to-speech settings for brand names, places, industry terms, and acronyms.

Voice AI Pronunciation Dictionaries

RimeModel providermodel release2026-05-19

Rime ships Coda dual-decoder TTS for conversational agents

Conversational dual-decoder design plus on-prem serving targets the reliability gap Rime says still kills agent demos at concurrency — Podonos vs ElevenLabs is vendor-hosted, treat as unverified.

Details1 source

Rime released Coda, a dual-decoder autoregressive TTS aimed at high-stakes conversational voice agents, with jointly trained semantic and acoustic decoders, a custom TIGERSTRIPE serving stack, and on-prem delivery like prior Rime models.

Coda TTS

Showavatars, video1

LemonSliceModel providerAPI2026-05-22

LemonSlice Actions let real-time avatars perform programmed gestures

Programmable gesture control moves LemonSlice avatars beyond automatically expressive speech, while the per-character onboarding step keeps the action catalogue a managed production surface.

Details2 sources · 1 X post

LemonSlice Actions let enterprise avatar builders trigger motion sequences such as pointing, waving, cheering, and jumping during a real-time call.

LemonSlice Actions

Introducing LemonSlice Actions Pointing, waving, cheering, jumping... Avatars can do any action in real-time. Available NOW to Enterprise customers. Comment LEMONSLICE to get $100 free credits :lemon

https://x.com/LemonSliceAI/status/2057875745897148729

ThinkLLMs, speech-to-speech1

StepFunModel providermodel release2026-05-24net-new

StepFun ships StepAudio 2.5 Realtime end-to-end voice LLM

End-to-end audio-in/audio-out with persona RLHF is StepFun's bid against cascade stacks in CN/EN — public consent/licensing detail for training voices is still thin.

Also SpeakHear

Details2 sources

StepFun released StepAudio 2.5 Realtime, an end-to-end Chinese/English speech LLM with persona RLHF, paralinguistic comprehension (vendor score 82.18), and WebSocket access at wss://api.stepfun.com/v1/realtime as step-2.5-realtime.

StepAudio 2.5 Realtime

Orchestrateframeworks, evals1

SonioxModel providerAPI2026-05-19

Soniox STT and TTS land natively in LiveKit Agents

One vendor for both ears and mouth inside LiveKit Agents cuts the usual two-dashboard STT/TTS mismatch on language coverage — useful before the June v5 model refresh.

Also HearSpeak

Details1 source

Soniox made its full speech stack — real-time STT and TTS across 60+ languages — available as native LiveKit Agents plugins via livekit-agents[soniox], so one API key covers both sides of an AgentSession.

Soniox STT and TTS LiveKit plugins · vendor LiveKit

Connecttelephony, channels2

Daily / PipecatPlatformlaunch2026-05-22

Daily ships event JSON raw-tracks defaults and animated cloud compositor

Post-call reconstruction gets meeting events and gapless audio by default from 26 May 2026, and live cloud composites can match offline VCS renders with 30fps layout animations.

Also Show

Details1 source

Daily upgraded raw-tracks and cloud recording with default event JSON, gapless audio transcodes, and a VCSRender compositor with layout animations, becoming account defaults on 26 May 2026.

Daily recording (raw-tracks + compositor)

Voice ElementsPlatformlaunch2026-05-20

Voice Elements launches Voice Nexus for OpenAI Realtime phones

Telephony CPaaS turns existing DIDs into OpenAI Realtime agents without a greenfield stack — $0.05/min all-in on the Voice Elements hop, model usage billed separately.

Also OrchestrateUse

Details1 source

Voice Elements released Voice Nexus Service, connecting DIDs and WebRTC endpoints on its cloud platform to OpenAI gpt-realtime and gpt-realtime-mini agents with per-call webhook config, CRM/MCP tools, transfers, and outbound campaigns at $0.05/min plus OpenAI usage.

Voice Nexus Service · vendor OpenAI

Useagents in market2

HyroPlatformpartnership2026-05-20net-new

Hyro plugs healthcare AI agents into Five9 contact centres

Vertical patient-access voice agents on a major CCaaS rail — shortens the path from Five9 shops to Hyro scheduling/triage agents.

Also Connect

Details3 sources

Hyro joined Five9’s AI Agent Connect as an accredited healthcare ISV so health systems can deploy Hyro voice agents for prescription management, triage, and scheduling inside Five9 with integration cut to about one hour.

AI Agents × Five9 · vendor Five9

CommurePlatformfunding2026-05-19

Commure raises $70M at a $7B valuation for healthcare AI operations

The financing is tied to a healthcare platform where call-centre agents, ambient documentation, and dictation are named product lines, so it clears the voice/ambient condition rather than being a general healthcare-software raise.

Also HearOrchestrate

Details1 source · 1 X post

Commure announced $70 million in financing at a $7 billion post-money valuation to expand its AI platform across healthcare operations, including calls, clinical notes, and revenue-cycle work.

Commure healthcare AI platform

We’ve raised $70M at a $7B valuation, led by @generalcatalyst, with participation from @sequoia Capital, @MorganStanley, and @Kirkland_Ellis. The $1 trillion spent on healthcare administration every year is a compounding tax on every doctor, every patient, every health system, and every physician-owned practice in the country. This capital will allow us to turn labor into software at an even faster clip, and to answer the ever-increasing demand we’re seeing. More specialties. More of the difficult, end-to-end work that no one else will fight in the mud to solve. https://t.co/AuwKsLVCKy

https://x.com/CommureOS/status/2056773218602746330