hotwirednews Get GrokBot

Week 11

9–15 Mar 20262026-W11 · 17 filings · 14 companies · 6 net-new

The Speak, Orchestrate and Connect layers led the week, with 4 filings each.

Hearspeech-to-text1

Devnagri AIPlatformlaunch2026-03-09net-new

Devnagri launches Speech AI with Indic ASR and TTS for enterprise voice workflows

India-focused ASR+TTS platform launch for BFSI/government voice journeys — the stack Black Bird later claims to advance on Voice of India (2026-09-30-devnagri-black-bird).

Also SpeakUse

Details2 sources

Devnagri AI launched Speech AI, adding a Conversational Speech Layer (ASR) and Enterprise Voice Generation (TTS) for 15+ major Indian languages into its language infrastructure for contact centres, KYC, collections, and citizen helplines.

Speech AI

Speaktext-to-speech4

Bland AIModel providerlaunch2026-03-11net-new

Bland Speech opens Bland's voice library as a standalone text-to-speech product

Bland sold its in-house TTS on its own five months before Bland Speech v3 (2026-08-04-bland-ai-bland-speech-v3), so the phone-agent vendor also competes for plain TTS API spend; the March changelog states no price or latency figure to compare with ElevenLabs or Cartesia.

Details1 source

Developers can synthesise speech from any text with the full Bland voice library from the dashboard or API, outside Bland's phone-agent product.

Bland Speech (standalone TTS)

Daily / PipecatPlatformlaunch2026-03-10

Pipecat 0.0.105 adds concurrent Cartesia TTS audio contexts

Overlapping synthesis cuts inter-sentence gaps on Cartesia Sonic pipelines.

Details3 sources

pipecat-ai 0.0.105 lets CartesiaTTSService synthesise the next sentence while the previous still plays, and folds audio context management into TTSService.

pipecat-ai 0.0.105

Hume AIModel providermodel release2026-03-10net-new

Hume AI open-sources TADA for synchronised text and speech generation

Open 1B and 3B checkpoints plus a 2,048-token long-form context give builders a local TTS path, while the 27 March MLX update extends that path to Apple Silicon; speed and hallucination figures are Hume's own evaluations.

Details7 sources · 2 X posts

Hume AI released TADA, a 1B English and 3B multilingual open-source speech-language model that synchronises text and audio for expressive TTS.

TADA

Today we're releasing our first open source TTS model, TADA!

https://x.com/hume_ai/status/2031401003078062578

Today, we're shipping MLX support for TADA, our open-source text-to-speech model

https://x.com/hume_ai/status/2037592399682171019
Fish AudioModel providermodel release2026-03-09net-new

Fish Audio open-sources S2 TTS with inline natural-language control

Open weights plus SGLang streaming for instruction-tagged TTS land in March — the foundation the June free S2.1 Pro API row already assumes.

Details1 source

Fish Audio open-sourced S2, a dual-autoregressive TTS model trained on 10M+ audio hours across ~50 languages, with free-form inline prosody tags, weights, fine-tuning code, and an SGLang streaming engine.

Fish Audio S2 (open weights)

ThinkLLMs, speech-to-speech2

NVIDIAModel providermodel release2026-03-11

NVIDIA releases Nemotron 3 Super open hybrid MoE for agentic reasoning

Open 120B/12B-active long-context MoE that Daily already scores near GPT-4.1 on voice-agent instruction/tool benches — a self-host path while hybrid-stack inference tooling for TTFT is still catching up.

Details2 sources

NVIDIA launched Nemotron 3 Super, a 120B-total / 12B-active hybrid Mamba-Transformer MoE with a native 1M-token context and open weights, datasets, and recipes.

Nemotron 3 Super

LiveKitPlatformlaunch2026-03-11

LiveKit Agents 1.4.5 adds Phonic plugin, GPT-5.4, and modality-aware Instructions

Phonic and GPT-5.4 are the think headline.

Details1 source

livekit-agents 1.4.5 adds a Phonic realtime plugin with generate_reply/update_chat_ctx, GPT-5.4 model strings, and modality-aware Instructions.

livekit-agents 1.4.5

Orchestrateframeworks, evals4

WonderfulPlatformfunding2026-03-12

Wonderful raises $150M Series B at a $2B valuation

The round funds expansion of an agentic platform already deploying production-grade agents across a growing customer base, and provides an earlier financing marker before Wonderful's $550M Series C in W36.

Also UseSpeak

Details2 sources · 1 X post

Wonderful raised $150 million in Series B funding to accelerate its agentic platform and enterprise AI adoption across more than 30 markets.

Wonderful AI platform

We’ve raised our $150 Million Series B round at a $2 Billion valuation. Our investors @insightpartners , @IndexVentures , @IVP, @BessemerVP, and @VineVenturesLP are all tripling down on their conviction in Wonderful’s platform and hyper-local operating model, as we continue deploying production-grade agents across our growing customer base in 30+ countries. And, we’re hiring globally (a lot): https://t.co/ZxdxcQZeo8

https://x.com/wonderful_ai/status/2032102636103540939
Daily / PipecatPlatformAPI2026-03-12

Pipecat Context Hub 0.0.7 adds symbol-lookup filters and incremental refresh

The 90-second to 23-second figure applies to an unchanged index, so the saving shows mainly on scheduled refreshes rather than after upstream changes.

Details1 source

Pipecat Context Hub 0.0.7 adds module, class_name, and content_type filters to get_code_snippet and makes refresh incremental through content-hash tracking.

Pipecat Context Hub 0.0.7

AgoraPlatformlaunch2026-03-11net-new

Agora launches Conversational AI Agents for customer service and sales

Puts a listed RTC vendor (NASDAQ: API) into the voice-agent orchestration lane beside Twilio/LiveKit-class connect stacks; backfill for roster seed AgoraIO.

Also Connect

Details1 source

Agora unveiled Conversational AI Agent solutions for customer service and sales & marketing on its SDRTN stack, combining Agent Studio, a Conversational AI Engine, and realtime network infrastructure.

Conversational AI Agents

Daily / PipecatPlatformlaunch2026-03-10

Pipecat 0.0.105 adds ServiceSwitcher failover strategy

STT/TTS/LLM redundancy becomes a strategy instead of a hand-rolled error handler.

Details3 sources

pipecat-ai 0.0.105 adds ServiceSwitcherStrategyFailover, which advances to the next service on non-fatal errors from the active one.

pipecat-ai 0.0.105

Connecttelephony, channels4

SalesforcePlatformlaunch2026-03-10

Salesforce launches Agentforce Contact Center with voice, CRM, and AI agents

The launch makes voice a first-class surface inside Salesforce's CRM rather than a telephony integration beside it, setting up the later GA milestone and the existing AWS A2A handoff.

Also UseOrchestrate

Details1 source

Agentforce Contact Center unifies voice, digital channels, CRM data, and AI agents in one service system for self-service and contextual human handoffs.

Agentforce Contact Center

RingCentralPlatformlaunch2026-03-10

RingCentral launches AIR Pro agentic voice AI for contact centres

UCaaS/CCaaS vendor ships a no-code agentic voice builder into its own stack on 10 March — controlled U.S. availability only, so treat GA timing as later.

Also OrchestrateUse

Details3 sources

RingCentral announced AIR Pro, a voice-first omnichannel AI agent platform with no-code AIR Pro Studio, embedded in its business communications and contact centre stack, starting in controlled availability for selected U.S. businesses.

AIR Pro

Daily / PipecatPlatformAPI2026-03-10

Pipecat 0.0.105 adds Daily custom video tracks and cloud-audio-only recording

Multi-track video and audio-only cloud recording cover avatar-plus-camera and compliance recording cases.

Details3 sources

pipecat-ai 0.0.105 adds video_out_destinations for multiple Daily video tracks and a cloud-audio-only recording option.

pipecat-ai 0.0.105

BolnaPlatformlaunch2026-03-10

Bolna adds TRAI-compliant outbound voice calling in India

The 140/160 split gives Indian teams a documented route for regulated outbound traffic, while Bolna's docs require separate DLT registration and list Vobiz and Plivo as providers.

Details2 sources · 1 X post

Bolna voice agents can place telemarketing and transactional calls through regulated 140- and 160-series numbers in India.

TRAI-compliant outbound voice calling · vendor Vobiz / Plivo

TRAI-compliant Outbound Voice AI Calling in India

https://x.com/bolna_dev/status/2031348900385444330

Useagents in market2

TailorCareEnterprisepartnership2026-03-10net-new

TailorCare partners with Cresta to standardise patient conversations

A named specialty-care provider gives Cresta a healthcare deployment in which call-level insights and real-time agent support are the first scale controls; the 10 March customer story says the partnership is early and reports no patient-outcome metric.

Details2 sources · 1 X post

TailorCare is using Cresta Agent Assist and Insights to give Care Advocates real-time support and analyse tens of thousands of patient conversations.

Cresta Agent Assist and Insights · vendor Cresta

Tailor_Care partnered with Cresta to standardize patient conversations as they grow

https://x.com/cresta/status/2031392202367156596
DecagonPlatformlaunch2026-03-09

Decagon ships Proactive Agents with user memory and outbound voice

Outbound calling plus durable user memory moves Decagon from reactive chat concierge into proactive voice outreach — a March ship that was still missing after the week-by-week backfill.

Also ConnectOrchestrate

Details2 sources · 1 X post

Decagon launched Proactive Agents, combining cross-channel user memory with outbound voice so AI concierge agents can initiate calls, retain customer context, and run Agent Operating Procedures without engineering tickets for every change.

Proactive Agents (User Memory + Outbound Voice)