hotwirednews Get GrokBot

Week 33

10–16 Aug 20262026-W33 · 24 filings · 22 companies · 8 net-new

The Use layer led the week with 10 of 24 filings, from Sarvam AI, Hippocratic AI, TCN, Currys and 6 more.

Hearspeech-to-text2

KrispPlatformmodel release2026-08-12

Krisp Voice Isolation 2.5 cuts competing-speaker WER from 36% to 11% across ten STT engines

Krisp's own table shows the trade-off: a 70% error cut when a second voice is in the room, but higher WER than raw audio on hard single-speaker acoustics (11.61% against 8.02%), so teams should gate isolation rather than run it everywhere; the October FP16 mode (2026-10-07-krisp-voice-isolation-fp16) roughly doubles how many streams it can clean per CPU.

Details2 sources

The VIVA voice-isolation model, now tuned for speech-to-text input rather than human listeners, ships in a full size and a lite size for CPU-constrained and edge deployments.

Krisp Voice Isolation 2.5 and 2.5 lite (VIVA)

FluidVoicePlatformlaunch2026-08-10

FluidVoice opens an early Windows release with local dictation and Fluid-1 mini

A local dictation product expands from its earlier platform into Windows and exposes a small research-preview model at the edge, keeping audio processing closer to the user rather than requiring a hosted transcription service.

Also Use

Details2 sources · 1 X post

FluidVoice launched an early Windows release of its free local dictation app, with an exclusive research preview of the 1.5 GB Fluid-1 mini model.

FluidVoice for Windows

The fastest free and local dictation app is on Windows now.

https://x.com/FluidVoiceApp/status/2086845815671451734

Speaktext-to-speech4

Resemble AIModel providereol2026-08-13

Resemble AI stops taking new paid voice-generation customers to focus on deepfake detection

The filing records a commercial exit from new paid voice-generation sales, not the disappearance of the underlying models; existing customers and open models remain in scope.

Details1 source

Resemble AI said it will support existing voice-generation customers but will not take new paid voice-generation customers as it shifts its commercial focus to deepfake detection.

Paid voice generation

Daily / PipecatPlatformdemo2026-08-12

Pipecat demo showcases Deepgram Flux TTS conversation-native voice in-pipeline

An employee video at partner GA is a clearer speak-layer proof of Flux-in-Pipecat than the July changelog row alone, and it sits beside Deepgram's own Flux TTS launch without replacing it.

Also Orchestrate

Details2 sources · 1 X post

@pipecat_ai marks Deepgram Flux TTS GA as live in Pipecat, with Jon Taylor (@JonPTaylor) walking through conversation-context tone, pronunciation consistency, native interruption handling, and sub-200 ms latency without SSML or style tags.

Deepgram Flux TTS in Pipecat demo · vendor Deepgram

Congrats to @DeepgramAI on the GA launch of Flux TTS, now live in Pipecat 🎉 @JonPTaylor looks at how Flux TTS delivers a more consistent voice experience. Flux reads the whole conversation, not just the next line — adaptive tone, consistent pronunciation, clean interruption handling natively (no SSML markup, no style tags!) all at sub-200ms latency. ➡️ https://docs.pipecat.ai/api-reference/server/services/tts/deepgram#deepgram-text-to-speech

https://x.com/pipecat_ai/status/2087575984102900191
DeepgramModel providerlaunch2026-08-12

Deepgram launches Flux TTS for conversation agents

STT, TTS, and Agent API from one vendor. Promo ended 12 Sep; standard pricing from 13 Sep. Users cited: Decagon, Sierra, Vapi, and Granola.

Details2 sources · 1 X post

Deepgram launched Flux TTS, a conversation TTS for enterprise voice agents, generally available, with a free promo through 12 Sep 2026 and standard pricing from 13 Sep.

Flux TTS

Today, we're launching Flux TTS: expressive voices built for live conversation. We asked our very own Lauren O’Connor to put Flux TTS to the test. Most TTS handles one line at a time. Flux TTS carries context across the whole conversation.

https://x.com/DeepgramAI/status/2087539784289849423
SonioxModel providermodel release2026-08-11

Soniox ships TTS v2 with audio tags across 60+ languages

Puts Soniox on the TTS side of the stack at a published $0.70/hour price, after the roster already tracked it mainly as an STT name.

Details1 source

Soniox released TTS v2 (`tts-rt-v2`) for realtime synthesis with programmable audio tags, voice cloning from short clips, character-level timestamps, and coverage across more than 60 languages at $0.70 per generated hour.

Soniox TTS v2

Showavatars, video1

LemonSlicePlatformpartnership2026-08-13

LemonSlice avatars become available directly inside Agora

An avatar vendor gains a distribution and transport path through a real-time communications platform, reducing the integration work for builders who want interactive visual agents on Agora.

Also ConnectSpeak

Details2 sources · 1 X post

LemonSlice and Agora partnered so developers can use LemonSlice interactive avatars with Agora's real-time voice, video, and streaming infrastructure.

LemonSlice avatars in Agora · vendor Agora

You can now use LemonSlice avatars directly within Agora.

https://x.com/LemonSliceAI/status/2087947641355125066

Orchestrateframeworks, evals5

Sarvam AIPlatformlaunch2026-08-13

Sarvam opens Voice Agents to everyone at ₹3.50 a minute

₹3.50 a minute with telephony and evals bundled gives Indian buyers a public per-minute benchmark, where Credgenics Prix AI (2026-10-03-credgenics-prix-ai, W40) prices on collection outcomes instead; the 3.5× conversion claim is Sarvam's, and its own posts disagree on volume (325 million minutes at Epoch, more than 350 million conversations at GA).

Also Connect

Details5 sources · 1 X post

Sarvam made its Voice Agents platform self-serve for any business or developer, bundling agent authoring, call simulation, phone numbers, evaluation, and analytics in one per-minute price.

Voice Agents (Indus)

Today, we're making Voice Agents available to everyone. With Voice Agents, you can build and scale intelligent, human-like agents that understand context, remember past conversations, and get better at achieving your business outcomes with every interaction. Our Voice Agents have already powered more than 350 million conversations across enterprise deployments. Operating at this scale has helped us refine the platform, deepen its capabilities, and build out the complete stack needed to handle the full complexity of voice interactions. The platform brings together the tools and systems to personalise, orchestrate, govern, and launch Voice Agents at scale. Start building your Voice Agent today: https://t.co/d5OFtxYqNp

https://x.com/SarvamAI/status/2087864104799314216
NewoPlatformlaunch2026-08-13

Newo adds Health Check to Vibe Mode for voice-agent configuration

A dated reliability feature for voice-agent operations turns configuration review into a repeatable, conversational check before a customer encounters a failed transfer or disconnected integration.

Also Use

Details1 source

Newo launched Health Check, a Vibe Mode capability that reviews phone routing, integrations, business hours, workflows, and transfer rules and explains what needs attention.

Health Check

LiveKitPlatformlaunch2026-08-12

LiveKit expressive mode: one `expressive` flag, provider-specific TTS markup

Emotion tags stop being a per-vendor prompt hack on those four TTS backends and become a flag on the agent runtime.

Also Speak

Details1 source · 1 X post

LiveKit announced a single `expressive` flag on AgentSession that injects provider-specific TTS markup so agents can match emotional register, working with Fish Audio, Inworld, Cartesia, and SpaceXAI Inference TTS at launch.

expressive mode

1/ Today we're launching Expressive mode in LiveKit Agents. Most voice agents deliver every sentence in that same, upbeat register we all know and hate (including bad news). Expressive mode makes your agent sound like it registers what the caller is actually feeling.

https://x.com/livekit/status/2087610892456546495
BlackBeltHelpPlatformlaunch2026-08-12net-new

BlackBeltHelp launches fully bilingual Spanish voice agents for college helpdesks

Broward College is the named production customer, while the 645-institution HACU figure frames the market rather than a BlackBeltHelp deployment; the company's 130-plus institutions and 8 million annual interactions are self-reported.

Also HearSpeak

Details1 source

The agents handle whole calls natively in Spanish, from identity verification to SMS follow-up, and run in production on Broward College's Academic Advising (Español) line.

Spanish AI Voice Agents

Daily / PipecatPlatformdemo2026-08-11

Pipecat demo shows UIWorker subagents steering async tasks off the voice loop

A dated employee showcase of UIWorker + async task steering makes the multi-worker abstractions concrete beyond changelog entries, and it is the pattern later demos compose for macOS assistants and distributed chess agents.

Details4 sources · 2 X posts

Kwindla (@kwindla) walks through Pipecat distributed agents / multi-worker support, with Mark Backman (@mark_backman) demonstrating a UIWorker that reads client-js screen hooks and steers async task agents without blocking the main conversation loop.

Pipecat distributed agents / UIWorker demo

We've spent a lot of time this year working on subagents support in Pipecat. Why would you want to use subagents? Well, in a voice agent, you can never block the main conversation loop. But your agent will often need to do various kinds of complex work that takes time to complete. - asynchronous tool calls - multi-step searches/research - session content analysis (guardrails, steering processes) - delegating a subset of work to an LLM configured differently from the voice LLM (for example, with reasoning effort set to high) - delegating work to a different LLM entirely (for example, an LLM fine-tuned to do a specific job) I think about these use cases as "subagents," but the abstractions in Pipecat are very flexible and general. So the Pipecat docs call these building blocks "distributed agents" and "multi-worker." Here's @mark_backman demonstrating one of the multi-worker examples. Mark's example code shows how to build a `UIWorker` subagent that communicates with hooks in the Pipecat client-js library to see what's on the screen and push UI updates. This example also shows how to start, stop, and steer async task agents.

https://x.com/kwindla/status/2087319311014904218
https://x.com/kwindla/status/2087319312751284647

Connecttelephony, channels2

TavusPlatformlaunch2026-08-13

Tavus launches Deployments for no-code PAL widgets, embeds, and landing pages

Tavus closes the path from no-code PAL creation to no-code distribution, letting a user publish an agent without standing up a backend or handling keys.

Also ShowOrchestrate

Details2 sources · 1 X post

Tavus launched Deployments so a PAL can be shipped as a website widget, an embed inside a product, or a standalone landing page without a backend, code, or API keys.

Deployments

RegalPlatformpartnership2026-08-13net-new

Regal connects voice AI agents to Five9 through AI Agent Connect

This makes a voice-agent vendor an event-driven follow-up layer for an installed CCaaS platform, rather than a separate dialler that operators must coordinate manually.

Also OrchestrateUse

Details2 sources

Regal's Five9 integration receives call outcomes and events so businesses can trigger follow-up AI calls or texts from the same contact-centre workflow.

Regal AI Agents × Five9 · vendor Five9

Useagents in market10

Sarvam AIPlatformpartnership2026-08-15

HP and Sarvam sign MoU to pre-install the Kivi voice app on HP laptops

An OEM pre-install would put Indian-language dictation on PCs before users go looking for it, but the MoU names no models, volumes, or date, and Kivi had no public Windows release as of 5 October 2026, so this is distribution intent rather than units shipped.

Also Hear

Details3 sources · 1 X post

Sarvam said Kivi, its voice app for dictation, drafting, and search across PC applications, will come pre-installed on HP laptops under an MoU that starts with voice and is meant to widen to other AI features.

Kivi

We're partnering with @HPIndia to bring voice to the centre of how people work on their PCs. This Independence Day, we're proud to share that Kivi, powered by Sarvam, will come pre-installed on HP laptops. Kivi lets people dictate instead of type, draft or rewrite content, search for information, and get help across the applications they already use. By bringing together HP's device ecosystem and reach with Sarvam's voice and language capabilities, we're making everyday work faster, simpler, and more accessible to millions of people. This is the beginning of a broader partnership between HP and Sarvam, with more to come. Read more: https://t.co/w9NOc2Cxlm

https://x.com/SarvamAI/status/2088591914438385959
Hippocratic AIPlatformlaunch2026-08-13net-new

Hippocratic AI ships Agentic Orchestrators for multi-agent clinical outcomes

Frames healthcare voice AI as supervised multi-agent programmes tied to population metrics, not one-off task bots for scheduling alone.

Also Orchestrate

Details2 sources

Hippocratic AI launched Agentic Orchestrators that coordinate teams of conversational voice agents around outcomes such as readmission reduction, Star Ratings/HEDIS, chronic care, and trial enrolment.

Agentic Orchestrators

TCNPlatformlaunch2026-08-12net-new

TCN expands AI Smart Node to full voice interactions with accent-capable TTS

Marks TCN Operator's voice-agent path in the corpus before the Abstrakt assist/QA buy — Smart Node is the self-serve voice automation node; Abstrakt is the live-agent assist layer.

Also OrchestrateSpeak

Details2 sources

TCN's Q3 2026 Operator highlights say AI Smart Node now fully supports voice calls with premium, accent-capable voices for intent, KYC, and CRM or Journey lookups.

AI Smart Node for voice

CurrysEnterpriselaunch2026-08-12net-new

Currys unifies voice, email, and social on NiCE CXone with Concentrix

Named UK retailer + Concentrix enabler wire on CXone (~6M interactions/year, ~80% FCR) — platform migration with transcription, not a standalone AI voice-agent SKU.

Also Connect

Details1 source

UK retailer Currys moved customer service onto NiCE CXone with Concentrix, covering voice, email, and social across purchase, delivery, repair, and support for nearly 6 million interactions a year.

NiCE CXone (with Concentrix) · vendor NiCE · delivered by Concentrix

Assort HealthPlatformlaunch2026-08-12

Assort Health launches Referrals voice-and-omnichannel AI agent

Puts outbound voice on the critical path of specialty referral leakage, with EHR booking in the same conversation rather than a separate dialler.

Details2 sources

Assort Health generally available Referrals, an AI agent that processes inbound referrals, matches providers, then reaches patients by voice, text, or email and books in the EHR.

Referrals

ReferralMDPlatformlaunch2026-08-11net-new

ReferralMD launches a voice agent that books patient appointments straight into the EHR

The call can start as soon as a referral lands in the platform that already holds it; ReferralMD claims bookings within minutes rather than days but publishes no call volumes or booking accuracy, and the October outbound agent (2026-10-14-referralmd-outbound-referral-voice-agent) extends it to stalled referrals.

Also Connect

Details2 sources

The Voice AI Scheduling Agent places and answers calls with patients, guardians, and providers, verifies identity, and books against live provider availability under each organisation's own scheduling rules.

Voice AI Scheduling Agent

CrestaEnablerpartnership2026-08-11

Cresta and Firstsource partner to scale AI-powered customer experiences

A second dated BPO/SI channel partnership for Cresta in this batch shows the company building distribution and operational capacity around its customer-service AI, rather than announcing a duplicate product release.

Also Orchestrate

Details2 sources · 1 X post

Cresta and Firstsource formed a strategic partnership combining Cresta's AI customer-service platform with Firstsource's operational expertise for enterprise deployments.

Cresta and Firstsource strategic partnership · vendor Cresta · delivered by Firstsource

We're excited to announce a new strategic partnership with Firstsource.

https://x.com/cresta/status/2087177108069958090
KotobaModel providerpartnership2026-08-10

Kotoba's Koto powers Japanese Agentforce Voice on Salesforce

Named enterprise telephony deployment for Koto after the May API alpha and June seed; Salesforce is vendor, Kotoba remains the primary actor.

Also HearSpeak

Details1 source

Kotoba said its Koto speech foundation model was selected to power Japanese-language conversations in Salesforce Agentforce Voice after the product became available in Japan on 7 August 2026.

Koto for Agentforce Voice · vendor Salesforce

American Residential ServicesEnterprisepartnership2026-08-10net-new

American Residential Services deploys Netic AI agents nationwide on ARS Connect

One of the largest US HVAC/plumbing roll-ups puts a single AI vendor inside its proprietary Connect stack nationwide, beside Allstar's Revin roofing roll-out (2026-10-02-allstar-services-revin-voice-agents).

Details2 sources · 1 X post

ARS/Rescue Rooter is rolling Netic's AI agents across its national home-services network, integrated into the proprietary ARS Connect platform to handle routine calls and scheduling beside human specialists.

Netic AI agents on ARS Connect · vendor Netic

American Residential Services is one of America's largest home services providers. Netic now runs inside American Residential Services' systems. Proud to roll out our partnership nationwide, helping @ARSRescueRooter serve more homeowners faster and better!

https://x.com/Netic_AI/status/2086864816266764339
AdmiralEnterprisepartnership2026-08-10net-new

Admiral runs an ElevenAgents voice agent on Admiral Money loan settlement-quote calls

A regulated UK lender keeps vulnerable and arrears callers away from the agent by design, so the 2.5-minute against 5-minute comparison covers only screened, simple-path calls, and both figures come from ElevenLabs webinars rather than an Admiral release.

Details3 sources

Admiral put an ElevenLabs-built voice agent into production for UK loan settlement quotes, with an authentication sub-agent first and a sentry sub-agent that sends vulnerable or arrears callers to a human priority queue.

Settlement-quote voice agent (Admiral Money) on ElevenAgents · vendor ElevenLabs