hotwirednews Get GrokBot

Week 10

2–8 Mar 20262026-W10 · 20 filings · 14 companies · 8 net-new

The Use layer led the week with 5 of 20 filings, from Salesforce, SuperDial, Singtel, Tennr and 1 more.

Hearspeech-to-text4

SpeechmaticsModel providermodel release2026-03-06

Speechmatics ships an Arabic–English bilingual speech model for code-switching

A single model covers mixed-language clinical and contact-centre audio across cloud, on-premises, and on-device deployments; the 6.3% WER and 35% comparison are Speechmatics benchmarks, not independent results.

Details3 sources · 2 X posts

The production model handles Gulf, Egyptian, and Levantine Arabic alongside English across real-time, batch, on-premises, and on-device deployments, with a bilingual medical variant.

Arabic–English bilingual speech model · vendor NVIDIA

A production-ready Arabic–English bilingual model that handles dialects and English simultaneously across real-time, batch, on-prem, and on-device.

https://x.com/Speechmatics/status/2031341290575556758

The world's most accurate Arabic–English bilingual model is now available.

https://x.com/Speechmatics/status/2031803508454814020
LiveKitPlatformlaunch2026-03-03

LiveKit Agents 1.4.4 adds Telnyx, AssemblyAI u3-rt-pro, and ElevenLabs STT knobs

One hear row for all STT provider ships in 1.4.4. Class-2: folded quiet patch(es) 1.4.3 into this row.

Details2 sources

livekit-agents 1.4.4 adds Telnyx STT, AssemblyAI u3-rt-pro with mid-session updates, and ElevenLabs STT keyterms plus usage events.

livekit-agents 1.4.4

AssemblyAIModel providermodel release2026-03-03net-new

AssemblyAI brings Universal-3 Pro to streaming transcription for voice agents

Real-time speaker labels and entity detection target the names, numbers, and specialised terms that voice-agent transcripts often lose; AssemblyAI lists 99+ languages but gives no independent benchmark in the launch materials.

Details3 sources · 1 X post

Developers can use Universal-3 Pro for live audio with real-time speaker labels, entity detection, code-switching, and support for more than 99 languages.

Universal-3 Pro Streaming

Universal-3-Pro is now available for streaming — bringing AssemblyAI's most accurate speech model to live audio for the first time.

https://x.com/AssemblyAI/status/2028872510599836141
Daily / PipecatPlatformlaunch2026-03-02

Pipecat 0.0.104 adds AssemblyAI u3-rt-pro and Azure private STT endpoints

U3 Pro turn detection and private Azure STT unblock regulated and Contact-Centre deployments.

Details3 sources

pipecat-ai 0.0.104 adds AssemblyAI u3-rt-pro with built-in turn detection, Azure Speech private endpoints, and strongly typed runtime service settings objects.

pipecat-ai 0.0.104

Speaktext-to-speech3

MicrosoftPlatformlaunch2026-03-06net-new

Microsoft adds Custom Neural Voices to Dynamics 365 Contact Center

Branded TTS inside Dynamics contact centre from 50 samples — limited-access / consent-gated, so not a open Foundry SKU like later MAI-Voice-1.

Also ConnectUse

Details1 source

Microsoft announced Custom Voice for Dynamics 365 Contact Center, letting verified customers clone branded neural voices from as few as 50 samples for virtual agents and intelligent IVR under its responsible AI limited-access controls.

Dynamics 365 Contact Center Custom Voice

LiveKitPlatformlaunch2026-03-03

LiveKit Agents 1.4.4 defaults Cartesia to Sonic 3 and adds Telnyx TTS

TTS defaults and new TTS plugins in one speak row.

Details1 source

livekit-agents 1.4.4 upgrades the Cartesia TTS default to Sonic 3 and adds Telnyx TTS beside Gradium pronunciation ids.

livekit-agents 1.4.4

Daily / PipecatPlatformAPI2026-03-02

Pipecat 0.0.104 unifies TTS text aggregation and word timestamps

One aggregation and timestamp contract across providers reduces per-service TTS forks.

Details3 sources

pipecat-ai 0.0.104 adds TextAggregationMode on TTSService, moves word timestamps into TTSService via supports_word_timestamps, and adds TextAggregationMetricsData.

pipecat-ai 0.0.104

Showavatars, video2

Daily / PipecatPlatformlaunch2026-03-02

Pipecat 0.0.104 adds LemonSlice realtime avatars on Daily

LemonSlice joins HeyGen LiveAvatar as a Daily-room avatar transport option.

Details3 sources

pipecat-ai 0.0.104 adds LemonSliceTransport and LemonSliceApi for realtime LemonSlice avatars in any Daily room.

pipecat-ai 0.0.104

falPlatformAPI2026-03-02net-new

fal hosts HeyGen avatar, Video Agent, and Video Translate APIs

Extends fal's voice-adjacent hosting beyond TTS/STT into show-layer talking avatars; pairs with the later Inworld TTS host note in W13.

Also Speak

Details1 source

fal made HeyGen's Video Agent, Avatar IV, Video Translate, and Avatar Video APIs available on its inference platform for talking avatars and multilingual lip-sync video.

HeyGen models on fal · vendor HeyGen

ThinkLLMs, speech-to-speech2

KakaoModel providerAPI2026-03-04net-new

Kakao opens its Kanana-1.5-o 9.8B omni model as an API for Korean and English speech in and out

Kakao's speech-in, speech-out model ships as an API in March, five months before the instructable-voice upgrade (2026-08-04-kakao-kanana-o-instruct-tts, W32) and seven before Kanana Space puts the stack in front of KakaoTalk users (2026-10-13-kakao-kanana-space); the Korean benchmark claims on the card are self-reported.

Also HearSpeak

Details2 sources

Kakao's Kanana-1.5-o-9.8B-instruct-2602 takes text, image, and audio input and answers in text or natural speech through an API, covering speech recognition, text-to-speech, speech translation, and multi-turn spoken conversation.

Kanana-1.5-o-9.8B-instruct-2602 API

LiveKitPlatformlaunch2026-03-03

LiveKit Agents 1.4.4 adds OpenAI Responses websocket mode and gpt-realtime-1.5

Realtime/LLM surface distinct from STT/TTS.

Details1 source

livekit-agents 1.4.4 supports OpenAI Responses websocket mode and adds gpt-realtime-1.5 to RealtimeModels, plus SambaNova LLM.

livekit-agents 1.4.4

Orchestrateframeworks, evals3

Daily / PipecatPlatformAPI2026-03-03

Pipecat Context Hub 0.0.6 indexes source for every configured repository

Community repositories added through PIPECAT_HUB_EXTRA_REPOS (v0.0.2, folded into 2026-02-20-pipecat-context-hub-0-0-1) now get API-level search; the 19% rise in source records is the maintainers' count for three repositories.

Details1 source

Pipecat Context Hub 0.0.6 runs the AST source ingester across all configured repositories so search_api covers more than the framework, and batches ChromaDB operations in groups of 5,000.

Pipecat Context Hub 0.0.6

NVIDIAPlatformlaunch2026-03-03

NVIDIA releases Nemotron Voice Agent, an open voice-agent blueprint on Pipecat and Nemotron speech models

The v1.0.0 reference stack puts Pipecat, NIM-hosted Nemotron LLMs, WebRTC, and Jetson Thor in one deployable blueprint, but its 20-turn history cap is a practical constraint for long calls. The later 2026-07-09-nvidia-nemotron-voice-agent-2-0-0 removes speculative speech processing and rebuilds the project around upstream Pipecat.

Details2 sources

The open blueprint combines NVIDIA speech models, Nemotron LLMs, WebRTC, and deployment tooling for production-ready voice-agent examples.

Nemotron Voice Agent

Daily / PipecatPlatformlaunch2026-03-02

Pipecat 0.0.104 adds StartupTimingObserver and latency breakdown events

Startup and turn latency become observable without custom observers in every bot.

Details3 sources

pipecat-ai 0.0.104 adds StartupTimingObserver, TurnMetricsData, and UserBotLatencyObserver events for per-service TTFB and first-bot-speech latency.

pipecat-ai 0.0.104

Connecttelephony, channels1

HuaweiPlatformlaunch2026-03-02net-new

Huawei launches next-gen AICC voice virtual agents at MWC

Carrier/enterprise CCaaS stack from Huawei adds agentic closed-loop voice with a claimed 20% self-service lift — MOS >4.5 and resolution gains are Huawei-stated only.

Also ThinkUse

Details1 source

Huawei announced next-generation voice virtual agents for its Artificial Intelligence Contact Center at MWC Barcelona, pairing domain-tuned LLMs with a Conversational Agent Engine and no-code SOP orchestration for carrier and enterprise contact centres.

AICC Voice Virtual Agents

Useagents in market5

SalesforcePlatformlaunch2026-03-05

Salesforce makes Agentforce Voice HIPAA-ready for healthcare support

HIPAA readiness opens the voice channel to a higher-consequence healthcare workflow, where contextual handoff matters as much as automated answers; the source does not publish a named customer result for the voice capability.

Also Connect

Details1 source

Agentforce Voice is HIPAA-ready for healthcare use cases, answering copay, benefits, and eligibility questions or routing callers to people with context.

Agentforce Voice for healthcare

SuperDialPlatformpartnership2026-03-04net-new

SuperDial partners with Omega Healthcare to scale payer voice AI

Back-office payer telephony via an SI/BPO channel — SuperDial voice agents land inside Omega’s agentic RCM stack rather than only direct provider sales.

Also Connect

Details3 sources

SuperDial and Omega Healthcare formed a commercial partnership so Omega’s agentic platform can run SuperDial voice AI agents on outbound payer calls for benefits verification, prior auth, claim follow-up, and credentialing outreach.

Voice AI × Omega Healthcare · delivered by Omega Healthcare

SingtelEnterprisepartnership2026-03-04net-new

Singtel puts Sierra behind its Shirley assistant and plans voice agents for outbound sales calls

The 70,000-case, 73%, and 76% figures describe the chat-led Shirley pilot; outbound voice sales calls were a plan in March, and Sierra's 21 October 2026 customer video does not say whether they went live. This precedes SoftBank's Sierra deal (2026-07-14-softbank-sierra-japan) by four months.

Details3 sources · 1 X post

Singtel's Shirley assistant uses Sierra's conversational AI for chat and voice customer-service journeys, with outbound sales agents planned.

Shirley AI agent on Sierra · vendor Sierra

For more than 145 years, Singtel has evolved alongside its customers, who now expect every interaction to be convenient, reliable, and fast. Its AI agent responds in a tone that feels natural and locally relevant. The agent is part of a company-wide commitment to AI that includes training 13,000 employees and governance that keeps pace with innovation. Sierra is proud to partner with Singtel as they work to become the world’s most intelligent connectivity company. Read more about our partnership with Singtel: https://sierra.ai/customers/singtel

https://x.com/SierraPlatform/status/2112750843657396366
TennrPlatformlaunch2026-03-02net-new

Tennr embeds voice AI phone calling into referral workflows

Referral orchestration vendor adding concurrent payer/patient voice calls inside the same order workflow — voice as RCM/front-office plumbing, not a greenfield CC agent.

Also Connect

Details2 sources

Tennr launched an embedded phone-calling feature so specialty front offices can run concurrent voice AI calls for payer benefits checks, documentation follow-up, patient welcome, and order confirmation without leaving the referral platform.

Phone Calling (voice AI)

Deutsche TelekomEnterprisepartnership2026-03-02

Deutsche Telekom embeds ElevenLabs Magenta AI Call Assistant in-network

Carrier-grade in-network voice agents (no app) with live translation — Germany GA is still "later this year" as of 2 March, so treat as MWC unveil not fleet-wide ship.

Also ConnectThink

Details1 source

Deutsche Telekom and ElevenLabs unveiled Magenta AI Call Assistant at MWC Barcelona: network-embedded voice AI for live translation, in-call agentic help, and post-call summaries on any phone, without a separate app.

Magenta AI Call Assistant · vendor ElevenLabs