hotwirednews Get GrokBot

Week 29

13–19 Jul 20262026-W29 · 22 filings · 21 companies · 6 net-new

The Orchestrate layer led the week with 7 of 22 filings, from Bolna, LiveKit, PolyAI, Talkative and 3 more.

Hearspeech-to-text2

NineNineSixModel providermodel release2026-07-17

NineNineSix opens Diamond 1.0, a generative model that restores degraded speech to 44.1 kHz

The founder says NineNineSix used it to prepare training data for its TTS (reply, 18 July), so it doubles as data tooling behind Gepard (2026-07-07-nineninesix-gepard-1-0, W28); the DNSMOS ranking is self-reported against six open restorers.

Details4 sources · 1 X post

Apache 2.0 speech-restoration model that regenerates band-limited, clipped, or packet-damaged recordings as full-band audio instead of masking noise.

Diamond 1.0

Our new open speech restoration model turns degraded audio into near-studio 44.1 kHz speech.

https://x.com/defoemark/status/2077955078758248935
WillowPlatformlaunch2026-07-14

Willow launches free unlimited AI dictation on iOS

The custom keyboard puts Willow's dictation product inside every iOS text field and changes the free tier to unlimited use; its comparative speed and accuracy claims are vendor claims.

Also Use

Details2 sources · 1 X post

Willow launched an iOS custom keyboard that lets users dictate into any text field with free, unlimited AI transcription.

Free unlimited AI dictation on iOS

Today, Willow is launching free, unlimited AI dictation on iOS.

https://x.com/WillowVoiceAI/status/2077117335673155814

Speaktext-to-speech4

LiveKitPlatformlaunch2026-07-18

LiveKit Agents 1.6.6 expands Fish Audio and SmallestAI TTS streaming

TTS streaming protocol upgrades.

Details1 source

livekit-agents 1.6.6 exposes Fish Audio temperature/top_p/bitrate/normalize options and uses Waves continuous WebSocket streaming for SmallestAI TTS.

livekit-agents 1.6.6

RimeModel providerfunding2026-07-15

Rime raises $24M Series A led by M13 to build voice interaction models

Twilio Ventures joins the cap table, and the round funds a move from TTS (Coda, filed 19 May) toward speech-to-speech; no Rime speech-to-speech model has a ship date yet.

Also Think

Details3 sources · 1 X post

Rime announced a $24 million Series A led by M13, with Corazon Capital, Unusual Ventures, Cadenza Ventures, and Twilio Ventures participating, to build speech-to-speech voice interaction models for enterprise conversations.

Today we announced our $24M Series A, led by @M13Company

https://x.com/rimelabs/status/2077395939195490392
Alibaba / QwenModel providermodel release2026-07-14

Alibaba ships Qwen-Audio-3.0-TTS in Flash and Plus on Model Studio, covering 16 languages

Q3 TTS comparisons that stop at Cartesia, ElevenLabs, and Inworld miss a multilingual, promptable option already on Alibaba Cloud.

Details6 sources · 2 X posts

Alibaba Cloud shipped Qwen-Audio-3.0-TTS in Flash (about 300ms first packet) and Plus variants on Model Studio, covering 16 languages with natural-language style control and inline non-verbal tags.

Qwen-Audio-3.0-TTS

Qwen-Audio-3.0-TTS is here. 🎙 Our latest text-to-speech model, in two flavors: • Flash: real-time interaction • Plus: high-quality generation https://t.co/O86LJfwKPu

https://x.com/TONGYI_SpeechAI/status/2079389657952113086

Production-grade TTS needs to do more than just read clean text—it has to handle messy real-world inputs like noisy references, abbreviations, and long formats. Meet Qwen-Audio-3.0-TTS: a new multi-lingual speech synthesis system built for actual production environments.

https://x.com/TONGYI_SpeechAI/status/2085645698662572518
TelnyxPlatformpartnership2026-07-13

Telnyx adds Fish Audio TTS to Voice AI Mission Control

A CPaaS voice-agent stack adds a creator-focused TTS provider without requiring builders to leave Mission Control or assemble the provider integration themselves.

Also Orchestrate

Details2 sources · 1 X post

Telnyx added Fish Audio's expressive multilingual voices to its Voice AI provider catalog, with three model tiers and ten voices across seven languages.

Fish Audio TTS in Telnyx Voice AI · vendor Fish Audio

Fish Audio TTS is now available for Telnyx Voice AI.

https://x.com/telnyx/status/2077426715048264103

Showavatars, video4

LemonSliceModel providermodel release2026-07-18

LemonSlice Lite targets high-volume consumer avatar apps

Keeps the July consumer Lite SKU distinct from October’s CWM-1 Lite so the two are not merged into one price story. An undated docs page citing $0.039/min was not used as this row’s price.

Details1 source

LemonSlice Lite is a lower-cost interactive avatar model for consumer apps with large audiences, announced in the Week 29 shipping log along with an ElevenLabs voice-agent import.

LemonSlice Lite

Daily / PipecatPlatformdemo2026-07-16

Pipecat Visemes turns any TTS stream into RTVI mouth keyframes

Provider-agnostic phonetic visemes close the gap between amplitude-only orb UIs and high-end avatars without locking clients to TTS-vendor viseme APIs — a show-layer path that stays on the same RTVI wire as the later W38 timing fixes.

Also SpeakOrchestrate

Details2 sources · 1 X post

Jon Taylor (@JonPTaylor) demos pipecat-visemes, a drop-in DSP lipsync processor that runs LPC formant analysis on any Pipecat TTS audio and ships playout-timed mouth keyframes to client-rendered avatars over RTVI.

pipecat-visemes lipsync processor

Voice agent UI feels stuck in the wiggly-waveform / glowing-orb phase. High-end avatars are exciting, but simple client-rendered characters need richer data than just amplitude to convey speech. An RMS-driven mouth flap gets you pretty far, but believable mouths need phonetic visemes — which usually means provider-specific viseme/timestamp APIs that only some TTS services offer. pipecat-visemes is a drop-in for Pipecat bots that works with any TTS provider: it runs lightweight formant analysis (LPC, numpy only — no ML) on the TTS audio stream itself and ships mouth keyframes to the client as standard RTVI messages. Keyframes ride the output transport's clock, so they land ~200ms ahead of the matching audio, stay in sync, and vanish on interruption. The browser just renders. Benchmark harness (in repo) scores the analyzer against Praat reference formant tracks and a phonetically-designed corpus ("Father was calm as he walked past the palm trees…"), checking raw formant error, mouth-trajectory shape, and articulation events. Tuning runs re-score offline against a committed baseline, so every DSP tweak shows up as a number. The formant analyzer is Tier 0 of a planned ladder: provider viseme events, timestamp+G2P, and an ONNX phoneme model can all slot in behind the same wire format — clients never change, accuracy just goes up.

https://x.com/JonPTaylor/status/2077765312687046703
TavusPlatformlaunch2026-07-15

Tavus Magic Canvas lets PALs surface interactive components during conversations

PALs gain a structured visual/action surface instead of relying on speech and video alone; Tavus publishes no independent adoption or task-completion measure for the launch.

Also Orchestrate

Details2 sources · 1 X post

Tavus launched Magic Canvas, a shared interactive surface where a PAL can render charts, calendars, questions, text, inputs, or alerts at the right moment in a conversation.

Magic Canvas

Now a PAL can bring anything to life inside a shared canvas at just the right time.

https://x.com/tavus/status/2077453632338030661
SynthesiaPlatformmodel release2026-07-15

Synthesia launches the Express-3 avatar model

Express-3 is a show-layer avatar ship with voice pairing and improved lip sync, not a speech-to-speech or voice-agent model; its availability belongs to the July primary date rather than the later X amplification.

Also Speak

Details2 sources · 1 X post

Synthesia launched Express-3, an avatar model with sentiment-aware performances, more natural lip sync and body motion, and up to twice-faster video generation.

Express-3

Introducing Express-3, our most advanced Avatar model yet.

https://x.com/synthesiaIO/status/2098023689669300281

Orchestrateframeworks, evals7

BolnaPlatformlaunch2026-07-19

Bolna opens an MCP server and a CLI beta for managing voice agents

Outbound calls become an action a coding assistant can take, gated only by the client's confirmation prompt; the CLI's balance check is the one built-in spend guard Bolna describes.

Details3 sources

Developers can create agents, place calls, and read transcripts from Claude, Cursor, or other MCP clients, or from a new `bolna` terminal tool.

Bolna MCP server and CLI beta

LiveKitPlatformlaunch2026-07-18

LiveKit Agents 1.6.6 adds update_models and MCP per-tool options

Runtime model swapping is a major orchestration affordance.

Details1 source

livekit-agents 1.6.6 adds Agent.update_models() to swap STT/VAD/LLM/TTS at runtime and per-tool options with progress updates on MCPToolset.

livekit-agents 1.6.6

PolyAIPlatformlaunch2026-07-17

PolyAI opens Builder MCP for natural-language agent management

Builder MCP moves PolyAI agent configuration, testing, and deployment into Claude Code, Cursor, and Codex, so developers can manage voice agents without opening PolyAI's studio.

Also Connect

Details1 source

The Model Context Protocol server wraps PolyAI's REST API so coding agents can create, configure, test, and deploy dialog agents in natural language.

PolyAI Builder MCP

TalkativePlatformlaunch2026-07-16

Talkative opens Voice AI Test Runs in beta to replay simulated customer scenarios against voice agents

Simulated-caller testing reached a second vendor's product the same week as Vattara's Evals Agent GA (2026-07-14-vattara-evals-agent); Talkative's beta runs only inside its own contact-centre platform and publishes no pass-rate figures.

Details2 sources

Talkative customers can write or auto-generate caller scenarios with success criteria and rerun them after each configuration change to catch regressions in their voice agents.

Voice AI Test Runs (beta)

SierraPlatformlaunch2026-07-16

Sierra Horizon: long-horizon agents billed on outcomes

Sierra is selling work that finishes over weeks (loans, prior auths), beside the six-minute phone call it already ran.

Details1 source · 1 X post

Sierra announced Horizon for agents that pursue multi-day, week, or month outcomes such as loan origination and healthcare prior authorisation, with a context engine and outcome-based pricing.

Horizon

Two topics have dominated our recent discussions with CEO - differentiation and tokenomics. Horizon is Sierra’s answer to both. @btaylor on tokenomics with @Kr00ney.

https://x.com/SierraPlatform/status/2077794548177391690
Druid AIPlatformlaunch2026-07-16net-new

Druid Voice reaches GA as a native enterprise STT/TTS voice stack

Enterprise agent-platform vendor ships first-party voice GA rather than gateway glue — first dated 2026 Druid product row.

Also SpeakHearConnect

Details1 source

Druid 9.27 made Druid Voice generally available as a unified conversational voice stack with proprietary STT/TTS fine-tuned for banking, insurance, healthcare, higher education, and retail.

Druid Voice

Vattara AIPlatformlaunch2026-07-14net-new

Vattara AI launches Evals Agent for voice-agent reliability testing

Dated GA for a Voice-Ops eval layer; prior pass dismissed IssueWire as undated/low-signal — the wire carries 14 July 2026.

Details2 sources

Vattara AI announced general availability of Evals Agent, a vendor-neutral platform that runs synthetic voice conversations and scores agents on its CLEAR framework before production.

Evals Agent

Useagents in market5

DeepLPlatformlaunch2026-07-17

DeepL Voice adds real-time translated subtitles to Google Meet

A translation vendor puts live meeting subtitles behind a simple Google Meet join flow, extending DeepL's Voice-to-Voice and Voice API lines into a widely used meeting surface.

Also HearSpeak

Details2 sources · 1 X post

DeepL made Voice for Meetings available in Google Meet for real-time translated subtitles, launching with five languages and a paste-a-link setup.

DeepL Voice for Google Meet

DeepL Voice is now available for Google Meet, giving you real-time translated subtitles.

https://x.com/DeepLcom/status/2078058843699814685
Raintree SystemsPlatformacquisition2026-07-15net-new

Raintree acquires Spike Technologies for agentic voice RCM inside its rehab EHR

Fills the Tarush-roster Raintree stub with a dated 2026 voice-RCM acquisition — agents inside the EMR, not a bolted-on phone tree.

Details2 sources

Raintree announced it acquired Spike Technologies so agentic voice agents for payer calls, claim follow-up, eligibility, prior authorisation and patient outreach run native to its rehabilitation EHR.

Spike Technologies acquisition

PwCEnablerlaunch2026-07-15net-new

PwC launches OpenAI-built agentic contact and service solutions with a voice agent CoE

Dated SI productisation of OpenAI multimodal voice for contact centres with unnamed clients — file as enabler, not an invented operator row.

Also OrchestrateSpeak

Details2 sources

PwC US launched agentic contact and service solutions developed with OpenAI, centred on an AI-powered voice and digital agent capability plus a joint Center of Excellence for enterprise deployments.

Agentic contact and service solutions (with OpenAI) · vendor OpenAI

SoftBankEnterprisepartnership2026-07-14net-new

SoftBank becomes Sierra’s exclusive partner in Japan

A telco with live LINEMO numbers taking exclusive Japan distribution is a stronger enterprise signal than another lab blog post.

Details1 source

SoftBank and Sierra formed a strategic partnership on 13 Jul 2026; from 14 Jul SoftBank is Sierra’s exclusive partner in Japan, selling Agent OS (voice and text) to Japanese enterprises.

Sierra Agent OS (Japan exclusive) · vendor Sierra

LiberatePlatformlaunch2026-07-14net-new

Liberate launches Voice FNOL Accelerator on Guidewire Marketplace

Puts insurance voice AI on the Guidewire distribution path with straight-through ClaimCenter write-back, which is the difference between a demo IVR and a claims system of record.

Details2 sources

Liberate shipped a Voice First Notice of Loss accelerator that takes insurance claim calls with AI voice agents, validates policies in PolicyCenter, and submits claims into Guidewire ClaimCenter.

Voice FNOL Accelerator for Guidewire ClaimCenter