hotwirednews Get GrokBot

Week 20

11–17 May 20262026-W20 · 19 filings · 15 companies · 5 net-new

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

Hearspeech-to-text2

Daily / PipecatPlatformAPI2026-05-14

Pipecat 1.2.0 expands ElevenLabs STT keyterm steering

Domain vocabulary steering reduces proper-noun errors on Scribe pipelines.

Details3 sources

pipecat-ai 1.2.0 improves ElevenLabs STT services so Scribe callers can bias transcription with keyterms.

pipecat-ai 1.2.0

DeepgramModel providermodel release2026-05-14

Deepgram expands Nova-3 speech-to-text support across Asia-Pacific

Nova-3's language expansion gives one STT product a broader Asia-Pacific coverage path; the same-day Flux Multilingual restaurant post is folded into the earlier 10-language Flux filing rather than treated as a second model launch.

Details1 source · 1 X post

Nova-3 adds Thai, Traditional Cantonese, Simplified and Traditional Mandarin, and Gujarati support alongside accuracy improvements for Bengali, Marathi, Tamil, and Telugu.

Nova-3 Asia-Pacific languages

Nova-3 expands speech-to-text support across Asia-Pacific. 🌏 New support now available for: 🔹 Thai 🔹 Cantonese Traditional 🔹 Mandarin Simplified 🔹 Mandarin Traditional Plus: ✔️ Accuracy improvements for Bengali, Marathi, Tamil, and Telugu ✔️ New Gujarati support Built for production-ready voice AI across streaming and batch workflows. Learn more: https://t.co/uKfHDgrYOq

https://x.com/DeepgramAI/status/2055000466082177467

Speaktext-to-speech5

Daily / PipecatPlatformAPI2026-05-14

Pipecat 1.2.0 adds Cartesia buffer delay and Deepgram MIP opt-out

Latency versus prosody trade-offs become per-bot settings rather than provider defaults only.

Details3 sources

pipecat-ai 1.2.0 adds max_buffer_delay_ms on CartesiaTTSService and mip_opt_out on Deepgram TTS services.

pipecat-ai 1.2.0

Resemble AIModel providermodel release2026-05-13net-new

Resemble ships DramaBox prompt-directed expressive TTS

Screenplay-style direction plus default watermarking targets game/audiobook/agent delivery that flat TTS still misses — English-only by design, not a multilingual race.

Details1 source

Resemble released DramaBox, an English-only 3.3B diffusion-transformer TTS directed by plain-language screenplay prompts, with optional 10-second voice cloning, ~2.5 s generation on a warm H100, and Resemble Watermarker embedded on every output.

DramaBox

TelnyxPlatformAPI2026-05-12

Telnyx makes more natural Ultra Voices live for AI Assistants

The update positions voice quality and prosody as a production usability change for live agents, but Telnyx supplies no public benchmark for the improvement in this announcement.

Also Connect

Details1 source · 1 X post

Telnyx upgraded its Ultra Voices for AI Assistants with more natural pauses, pacing, and handling of numbers, codes, and names.

Telnyx Ultra Voices

More natural Ultra Voices are now live on Telnyx. This is one of those voice AI upgrades that sounds small until you actually hear it. Because in production voice agents, better voice quality is not just about sounding nicer. It changes how usable the agent feels. A few things matter a lot on real calls: ➡️ Does the voice pause naturally? ➡️ Does it handle numbers, codes, and names without sounding awkward? ➡️ Does it keep sentence flow when the prompt gets more complex? ➡️ Does it feel like something you would actually put in front of customers? That is what this Ultra Voices upgrade is focused on. Cleaner delivery. More natural pacing. Better handling for the messy parts of live conversations. If you are building voice agents on Telnyx AI Assistants, upgraded Ultra Voices are available now. Read more about the launch here:

https://x.com/telnyx/status/2054215321163358624
InworldModel providermodel release2026-05-12

Inworld brings Realtime TTS to device-side private preview

Local TTS removes network dependence and marginal cloud inference from consumer voice experiences, but private-preview availability means production access is not yet self-serve.

Details1 source · 1 X post

Inworld's realtime TTS can run locally on mobile, PC, and other consumer hardware without cloud inference, with private-preview access.

Inworld on-device TTS

The #1 TTS is now available on-device Running our TTS locally is something we've been working toward for a while. Today it's in private preview. Leaderboard-topping quality, no cloud inference costs, no network dependency. Just the model on your hardware running locally.

https://x.com/inworld/status/2054238124914659449
SalesforcePlatformlaunch2026-05-11

Salesforce launches Briefings, an audio summary feed for field teams

An audio-first account surface turns Salesforce context into a recurring spoken workflow for mobile teams, distinct from voice input and phone agents because the primary interaction is listening to generated briefings.

Also Use

Details1 source

Salesforce Briefings gives field sales and life-sciences teams daily, context-curated audio summaries of healthcare-provider accounts in an offline-ready playlist.

Briefings

Showavatars, video2

TavusPlatformlaunch2026-05-14

Tavus launches Image-to-Replica for conversational AI humans

Image-to-Replica expands conversational avatar creation beyond people available for a recording session; it qualifies here because the output is a speaking, real-time AI human rather than a standalone image-generation feature.

Also SpeakThink

Details1 source · 1 X post

Tavus can create a usable Phoenix-4.5 AI human from a single image, including an illustrated character, brand mascot, or AI-generated face that can hold a real conversation.

Image-to-Replica

Introducing Image-to-Replica. Any face that can exist as a single image, whether it's a brand mascot that has lived in a logo for thirty years, a historical figure, or an AI character designed from a single prompt, can now hold a real conversation.

https://x.com/tavus/status/2054990954118152627
Boson AIModel providermodel release2026-05-13

Boson AI launches Higgs Avatar for real-time talking faces

Adds a face layer to Boson's own STT/TTS stack for production voice agents — private preview only, eight sessions per H100 is the cost claim to watch.

Also Speak

Details1 source

Boson AI introduced Higgs Avatar, a real-time avatar foundation model that generates an expressive speaking face from a single still image, co-designed with Higgs TTS, streaming frames at ~16 ms on one H100 hosting up to eight concurrent conversations.

Higgs Avatar

ThinkLLMs, speech-to-speech1

Shunya LabsModel providermodel release2026-05-11

Shunya Labs releases Vak, an open-weight real-time speech translation system for 55 Indian languages

Open weights for an Indic speech-to-speech translation stack let state and regulated buyers keep audio on-premises, a month before Shunya's 216-language voice AI launch (2026-06-15-shunya-labs-multilingual-voice-ai, W25); latency and voice preservation are vendor claims with no third-party test, and the 38.5 BLEU covers text translation only.

Also HearSpeak

Details5 sources · 1 X post

Vak translates speech any-to-any across 55 Indian languages while keeping the speaker's voice and emotional tone, with weights published so government, healthcare, and BFSI users can run it on their own infrastructure.

Vāķ

🚀 Launching Vāķ: Open-Weight Voice AI for 55 Indian Languages Jointly launched by @ShunyaLabsAI & @nasscom at #IndiaAIImpactSummit2026 📍 Join us: Feb 18 | 11:30 AM | Shakuntulam Banquet 🔗 https://t.co/nxIqyk7zfQ 🔗 https://t.co/aWhKWcXyXR #ShunyaLabs #VAK #Nasscom #ASR

https://x.com/ShunyaLabsAI/status/2024004177416921224

Orchestrateframeworks, evals3

Daily / PipecatPlatformlaunch2026-05-14

Pipecat 1.2.0 adds incomplete-turn and deferred stop strategies

Incomplete-turn handling moves from an LLM mixin into composable turn strategies. Class-2: folded quiet patch(es) 1.2.1 into this row.

Details5 sources

pipecat-ai 1.2.0 adds FilterIncompleteUserTurnStrategies, DeferredUserTurnStopStrategy, ExternalUserTurnCompletionStopStrategy, and LLMTurnCompletionUserTurnStopStrategy.

pipecat-ai 1.2.0

LiveKitPlatformlaunch2026-05-13

LiveKit Agents 1.5.9 introduces Answering Machine Detection

Outbound call classification is a new orchestration capability.

Details1 source

livekit-agents 1.5.9 introduces Answering Machine Detection so outbound calls can classify person, voicemail, IVR, or unreachable numbers.

livekit-agents 1.5.9

ParloaPlatformlaunch2026-05-11

Parloa adds Subtask Agents, a multi-agent orchestration model, to its Agent Management Platform

The 24% and 47% figures come from one unnamed customer's live tests, so treat them as a vendor claim. The pre-LLM gate is the compliance feature for insurers, banks, and healthcare; Pipecat shipped its own multi-agent workers two weeks later (2026-05-28-pipecat-1-3-0-multi-agent-workers), with no documented link between the two.

Details1 source

Subtask Agents split one long agent prompt into task-focused subagents, such as greeting and triage, authentication, billing, and order management, with deterministic gates deciding which one the model can see.

Subtask Agents

Connecttelephony, channels1

Daily / PipecatPlatformAPI2026-05-14

Pipecat 1.2.0 adds RTVI UI Agent Protocol messages

Voice bots can observe and cancel UI jobs over the same RTVI channel as audio.

Details3 sources

pipecat-ai 1.2.0 adds first-class RTVI ui-event, ui-snapshot, and ui-cancel-task client-to-server messages for UI-driven agents.

pipecat-ai 1.2.0

Useagents in market5

KolsetuPlatformlaunch2026-05-13net-new

Kolsetu opens Elba Self-Serve voice AI to EU builders

EU self-serve voice orchestration with compliance as the opener — lowers the bar for agencies and SMBs that refuse US-only stacks, while keeping an enterprise services lane.

Also OrchestrateConnect

Details1 source

Kolsetu opened Elba Self-Serve on 13 May for European builders: GDPR-oriented voice-agent platform with provider-flexible models, MCP server connections, no credit card to start, and an enterprise path for mission-critical deployments.

Elba Self-Serve

VapiPlatformfunding2026-05-12net-new

Vapi raises $50M Series B after one billion calls

The round funds a voice-agent platform with a stated billion-call production base, rather than a model lab or an unlaunched prototype; operating and growth figures are company-reported.

Also Orchestrate

Details1 source · 1 X post

Vapi announced a $50 million Series B led by Peak XV, bringing total funding to $72 million after its agents crossed one billion handled calls and enterprise revenue grew tenfold.

Vapi voice-agent platform (Series B)

We've raised a $50M Series B to keep making everyday calls extraordinary. After 1,000,000,000 calls and 10x enterprise ARR growth, Vapi's platform approach to voice agents is becoming the standard for businesses building human interfaces.

https://x.com/Vapi_AI/status/2054215176862183757
ParloaPlatformpartnership2026-05-12

Parloa and SAP partner to bring Parloa agents into SAP Service Cloud

The release calls this the first step of a broader collaboration, with deeper product integration and go-to-market work still to come, and it states no investment size or customer metric. SAP appears as a Parloa customer in the Series D release (2026-01-15-parloa-series-d), four months earlier.

Also Connect

Details1 source

Parloa's AI agents now work with SAP Service Cloud AI agents for voice and digital service, with the Parloa solution listed on the SAP Store.

Parloa for SAP Service Cloud

Yellow.aiPlatformlaunch2026-05-11net-new

Yellow.ai launches Nexus Vox zero-hop enterprise voice

Enterprise CX platform ships voice as a native runtime layer rather than another bolted STT/TTS pair — 500+ language claim and sub-400 ms e2e are vendor figures to pressure-test.

Also ThinkSpeakConnect

Details1 source

Yellow.ai launched Nexus Vox inside its Nexus agentic platform: a single-runtime voice stack claiming sub-400 ms end-to-end latency, 500+ languages/dialects, 10-second voice cloning, real-time sentiment, and direct CRM/ticketing orchestration under one SLA.

Nexus Vox

Mahindra & MahindraEnterprisepartnership2026-05-11net-new

Mahindra deploys nAIna voice agents with ElevenLabs for XUV 7XO demand

A named Indian automotive deployment puts voice agents in launch-scale outbound demand generation; the conversion uplift and contact-rate improvement are claims in ElevenLabs' customer story.

Also SpeakConnect

Details1 source · 1 X post

Mahindra AI used nAIna, an AI conversational voice agent with ElevenLabs Text to Speech, for outbound sales and enquiries during the XUV 7XO launch.

nAIna voice agent for XUV 7XO launch · vendor ElevenLabs

@Mahindra_Auto, one of the world's largest automotive manufacturers operating across 100+ countries, deployed AI voice agents powered by ElevenLabs to manage peak demand during the XUV 7XO launch - achieving higher contact rates and an ~8% conversion uplift.

https://x.com/ElevenLabs/status/2053765427239506432