hotwirednews Get GrokBot

Week 23

1–7 Jun 20262026-W23 · 19 filings · 18 companies · 7 net-new

The Hear and Speak layers led the week, with 5 filings each.

Hearspeech-to-text5

Ringg AIModel providermodel release2026-06-05net-new

Ringg ships Parrot STT for Hindi-English voice agents

Hindi-heavy, code-mixed phone audio is where generic STT still breaks agent actions — Ringg prices on transcript received and claims sub-100 ms compute latency on that path.

Details2 sources

Ringg AI released Parrot STT V1 for real-time voice agents, targeting noisy, code-mixed Hindi-English calls with vendor-claimed ~60 ms compute latency, normalised WER of 7.27 on open Hindi benches versus 8.94 ElevenLabs and 12.36 Deepgram, and transcript-based pricing.

Parrot STT V1

NVIDIAModel providermodel release2026-06-04

NVIDIA opens Nemotron 3.5 ASR for multilingual streaming transcription

Open multilingual streaming STT with an 80ms–1.12s latency dial, and Daily’s voice-agent bench placing it on the latency/accuracy frontier for builders who need self-hosted transcription.

Details2 sources

NVIDIA released open-weight Nemotron 3.5 ASR (0.6B), a multilingual streaming speech-to-text model covering about 40 language-locales with configurable chunk latency from 80ms.

Nemotron 3.5 ASR

TelnyxPlatformpartnership2026-06-03

Telnyx adds Soniox STT to its real-time voice AI workflows

Provider choice moves into Telnyx's existing assistant configuration path, so teams can test multilingual STT without rebuilding telephony or agent routing; no comparative quality result is published.

Also Connect

Details2 sources · 1 X post

Telnyx customers can select Soniox STT for real-time transcription, automatic language detection, interim results, and voice-agent endpointing.

Soniox STT for Telnyx Voice AI · vendor Soniox

Voice AI teams now have another STT option on Telnyx. We just added @soniox_ai STT for real-time transcription workflows. This matters because STT is one of those pieces you only notice when it gets things wrong. If the caller switches languages, says a product name, gives an…

https://x.com/telnyx/status/2062236990008107235
SonioxModel providerpartnership2026-06-02

Soniox brings its multilingual STT API into Tencent RTC

A native console integration puts Soniox's language-detection and code-switching path inside Tencent RTC; Tencent is the distribution surface, while Soniox remains the speech provider.

Also Connect

Details2 sources · 1 X post

Developers can select Soniox's multilingual STT in Tencent RTC for streaming transcription, automatic language detection, and voice-agent workflows.

Soniox STT in Tencent RTC · vendor Tencent RTC

We are pleased to announce the strategic partnership with @TencentRTC, bringing the world's most accurate speech-to-text natively into @tencentcloud. Starting today, developers can integrate the Soniox STT API directly within the Tencent RTC console. Build intelligent customer…

https://x.com/soniox_ai/status/2061742228012147165
MicrosoftModel providermodel release2026-06-02

Microsoft ships MAI-Transcribe-1.5 across 43 languages

Pairs with MAI-Voice-2 on the same day for a Foundry hear+speak stack — keyword biasing and long-file speed are the enterprise hooks, with streaming still on the roadmap.

Details1 source

Microsoft launched MAI-Transcribe-1.5 in Foundry with coverage expanded from 25 to 43 languages, vendor-claimed best-in-class FLEURS WER, 2.4% WER and #3 on Artificial Analysis, keyword biasing (up to 30% relative WER cut on FLEURS), and hour-of-audio transcription in under 15 seconds.

MAI-Transcribe-1.5

Seevision input1

EnvisionPlatformlaunch2026-06-01net-new

Envision ships on-device scene description and visual Q&A with Gemma 4

Moves assistive camera Q&A off the cloud path — Gemma 4 on Arm SME2 inside Envision — so descriptions keep working without signal, a practical gap versus cloud-only Seeing AI-class apps.

Also Use

Details1 source

Envision integrated Google's Gemma 4 with Arm SME2 so blind and low-vision users get scene description and follow-up visual questions on-device in the Envision app, including offline.

On-device scene description and visual Q&A (Gemma 4) · vendor Google · delivered by Arm

Speaktext-to-speech5

Boson AIModel providermodel release2026-06-04

Boson AI serves Higgs Audio v3 TTS on SGLang-Omni

Open serving path for a controllable multilingual agent TTS — tags in the text stream plus a shared multi-stage runtime, not another closed API-only drop.

Also Orchestrate

Details2 sources

Boson AI and the SGLang-Omni team announced end-to-end serving for Higgs Audio v3 TTS: a ~4B streaming conversational TTS covering 100 languages with inline emotion/style/prosody/SFX tags, zero-shot cloning, and open cookbook deploy via sgl-omni.

Higgs Audio v3 TTS on SGLang-Omni

VapiPlatformpartnership2026-06-03

Vapi makes Grok Voice the default for core platform voices

Distribution deal, not a new model card — Grok becomes the path of least resistance on one of the largest voice-agent platforms.

Also Orchestrate

Details1 source

xAI and Vapi announced that Grok Voice became the default TTS engine for Vapi's 12 core voices, following a Vapi blind evaluation that ranked Grok first, covering the installed base of voice agents on the platform.

Grok Voice as default TTS · vendor xAI

MicrosoftModel providermodel release2026-06-02

Microsoft ships MAI-Voice-2 multilingual expressive TTS

Puts in-house Microsoft TTS into Foundry and Contact Center with emotion tags and consent-gated cloning — 72% preference over Voice-1 on the same day as Transcribe-1.5.

Details1 source

Microsoft launched MAI-Voice-2 in Foundry, expanding from English-only to 15+ languages/locales with emotion tags, 5–60 s zero-shot voice prompting with consent guardrails, code-switching for Hindi–English and Spanish–English, and a 72% preference win over MAI-Voice-1.

MAI-Voice-2

EqualyzAIModel providerlaunch2026-06-02net-new

EqualyzAI VoiceMaker ships African-language TTS, STT, and developer SDKs

First dated 2026 Equalyz product ship in the corpus; prior Forbes Africa piece was profile-only.

Also Hear

Details4 sources

EqualyzAI's VoiceMaker platform for Nigerian-language TTS, multilingual transcription, and voice cloning became available with Terms effective 1 June 2026 and the first myvoicemaker Python SDK published 2 June 2026.

VoiceMaker

LokutorModel providerAPI2026-06-01

Lokutor TTS joins Pipecat as a community integration through pipecat-lokutor

A small TTS vendor reaches Pipecat builders with one pip install outside the core repo, but the post gives no latency, voice count, or pricing, so the integration says nothing yet about quality.

Details3 sources

Lokutor published pipecat-lokutor, a LokutorTTSService that drops Lokutor's text-to-speech into any Pipecat pipeline beside the builder's chosen STT and LLM.

pipecat-lokutor (LokutorTTSService)

Showavatars, video1

TavusPlatformlaunch2026-06-04

Tavus launches Solutions, built-to-order AI humans for enterprise workflows

Tavus is packaging its avatar, perception, and conversation stack as a managed vertical deployment service, with the Solutions page saying launches can happen in weeks rather than publishing an independent delivery measure.

Also ThinkUse

Details2 sources · 1 X post

Tavus Solutions offers production-ready AI humans that the Tavus team builds, tunes, and runs alongside enterprises in sales, healthcare, learning, and recruiting.

Tavus Solutions

Introducing Tavus Solutions. Complete, production-ready AI humans for the enterprise workflows where human-quality conversation changes the outcome. Built and run alongside you by the Tavus team.

https://x.com/tavus/status/2062586793187553338

ThinkLLMs, speech-to-speech2

ModulateModel providerAPI2026-06-03

Modulate opens its Velma audio-native model as an API

Audio-native analysis can expose conversational risk and affect without making transcription the only representation of a call; Modulate's 550M-plus-hour figure is a vendor claim, not an independent usage audit.

Details2 sources · 1 X post

Modulate made Velma available to developers as an audio-native API that analyses emotion, tone, intent, rhythm, vocal stress, and other conversational signals.

Velma API

The world's first audio-native AI model is now available as an API. Velma listens and understands like a human — emotions, tone, intent, rhythm, vocal stress. Already analyzed 550M+ hours of conversation for Fortune 500s. Now open to developers. 🧵

https://x.com/modulate_ai/status/2062170408950882357
GoogleModel providermodel release2026-06-03

Google opens Gemma 4 12B with native audio input for local multimodal agents

Native audio input makes Gemma 4 12B usable for local audio-aware agents without a separate audio encoder, although the launch is a general multimodal model rather than a dedicated speech model.

Also Hear

Details2 sources · 1 X post

Gemma 4 12B is an open model for laptops with native audio and vision inputs, agentic reasoning, and a memory footprint small enough to run with 16GB of VRAM or unified memory.

Gemma 4 12B

Today we’re introducing Gemma 4 12B — our latest open model that brings advanced agentic reasoning, vision and audio directly to your laptop. It delivers performance nearing our larger Gemma models with a much smaller total memory footprint, while being small enough to run…

https://x.com/Google/status/2062203526588088452

Orchestrateframeworks, evals1

CiscoPlatformlaunch2026-06-03net-new

Cisco previews Webex AI Concierge and workforce tools that score human and AI agents at Cisco Live

Dialog (2026-10-07-cisco-webex-dialog-voice, W41) later names AI Concierge as one of the agents it runs, so this June front door is the base the October harness extends; putting AI agents into the same QM and scheduling tools as staff is the operational half of that voice push.

Also Connect

Details1 source

Cisco announced AI Concierge, a pre-configured customer-facing Webex AI Agent with enhanced voice and persistent memory, and AI workforce engagement tools that forecast, schedule, and quality-score AI agents beside people.

Webex AI Concierge and AI WEM

Useagents in market4

FinPlatformlaunch2026-06-04net-new

Fin ships Voice 2 on Apex Flash for phone CX

CX agent vendor ships a voice-specialised model path with named resolution and latency deltas — phone automation framed as resolution rate, not demo fluency.

Also ThinkConnect

Details3 sources

Fin (formerly Intercom) launched Fin Voice 2 on its Apex Flash model with over 20 phone-ops updates, claiming 24.5% higher resolution and 0.43 s faster responses, plus identity verification, refunds, appointments, proactive outbound, and richer call insights.

Fin Voice 2 (Apex Flash)

VonagePlatformpartnership2026-06-03

Vonage embeds vertical AI agents for healthcare, FS, and retail in VCC

Vertical-trained agents land natively inside VCC on 3 June — Avaamo for healthcare and Syndeo for FS/retail — so operators skip custom telephony glue for those three industries.

Also Connect

Details1 source

Vonage launched industry-specific AI agents inside Vonage Contact Center through partnerships with Avaamo for healthcare and Syndeo for financial services and retail, with context-preserving handoff to live agents.

Vertical AI Agents (VCC + Avaamo / Syndeo) · vendor Avaamo / Syndeo

Loman AIPlatformpartnership2026-06-03net-new

Loman brings voice AI to Shift4 Dine restaurant operators

A POS distribution partnership puts a phone agent in the restaurant workflow, but Loman publishes no deployment count or independent measure for the claimed guest-experience improvement.

Also Connect

Details1 source · 1 X post

Shift4 Dine operators can use Loman to answer every restaurant call and handle guest conversations without adding headcount.

Loman Voice AI for Shift4 Dine · vendor Shift4

Big Partnership News - Loman has partnered up with shift4 to bring the industry's leading Voice AI platform to Shift4 Dine Operators! 🗞️ If you're running Shift4 POS, you can now leverage Loman to answer every call & deliver a better guest experience without adding headcount.

https://x.com/useloman/status/2062212857551299057
BroadvoicePlatformlaunch2026-06-02net-new

Broadvoice launches GoEngage S2S voice agent on GoContact

CCaaS vendor ships native S2S inbound handling plus a supervisor analyst agent on the same GoContact console — less bolt-on bot SKU, more platform AI.

Also ConnectThink

Details1 source

Broadvoice released GoEngage, a speech-to-speech inbound voice agent on GoContact that processes audio directly rather than a cascade IVR path, plus AI Analyst for plain-language contact-centre performance questions, both generally available to GoContact customers.

GoEngage and AI Analyst