hotwirednews Get GrokBot

Week 31

27 Jul – 2 Aug 20262026-W31 · 28 filings · 20 companies · 7 net-new

The Hear layer led the week with 8 of 28 filings, from Daily / Pipecat, Gladia, Sarvam AI, Edge0 and 2 more.

Hearspeech-to-text8

Daily / PipecatPlatformAPI2026-08-01

Pipecat 1.7.0 adds STT usage metrics and ElevenLabs background-audio filter

Usage metering and cleaner Scribe transcripts matter for production cost and IVR noise.

Details3 sources

pipecat-ai 1.7.0 expands speech-to-text usage metrics and adds filter_background_audio on ElevenLabsRealtimeSTTService, plus shared http_client for OpenAI TTS and Whisper STT.

pipecat-ai 1.7.0

GladiaModel provideracquisition2026-07-31

OVH Groupe completes acquisition of Gladia speech-to-text

A European sovereign-cloud operator internalises a multilingual STT API rather than renting US speech models — Gladia stays branded, but compute and residency now sit on OVH.

Details3 sources

OVH Groupe announced it completed the acquisition of Paris speech-to-text company Gladia, following exclusive negotiations disclosed on 11 June 2026.

Gladia STT

Sarvam AIModel providermodel release2026-07-30

Sarvam ships Saaras V4 Multi-speaker for overlapping-speech transcription

Joint transcription and speaker attribution for 22 Indian languages lands eight weeks before NVIDIA's open, diarisation-only Nemotron 3 Diarization (2026-09-23-nvidia-nemotron-3-diarization, W39); the wins over Scribe v2 and Deepgram Diarization v2 are Sarvam's own, measured on a benchmark Sarvam built and released.

Details5 sources · 1 X post

Sarvam introduced Saaras V4 Multi-speaker at Epoch, a speech recognition model that transcribes and attributes overlapping speakers in one pass instead of running diarisation before ASR, available through its API and playground.

Saaras V4 Multi-speaker

Introducing Saaras V4 Multi-speaker Our best-in-class speech recognition model that accurately captures overlapping conversations and multi-speaker interactions, while delivering state-of-the-art performance in English across accents. https://t.co/fUXgnMQiet

https://x.com/SarvamAI/status/2082816318764618061
Sarvam AIPlatformlaunch2026-07-30

Sarvam launches Kivi, a Mac voice app for dictation and editing in 22 Indian languages

A model lab now ships its own desktop client for Hindi–English code-mixed dictation, where Nabla's same-week Mac dictation (2026-07-28-nabla-dictation-mac) is clinical and Fireflies Talk (2026-09-29-fireflies-talk, W40) is a bundled extra; Sarvam has published no accuracy figures for the app, and pricing after the free alpha is unannounced.

Also Use

Details4 sources · 2 X posts

Sarvam unveiled Kivi at its Epoch event in Bengaluru, a macOS app where holding Fn dictates cleaned text into any application and Fn plus Control lets users rewrite, format, or draft by voice, in English and 22 Indian languages with mid-sentence switching.

Kivi

kivi for Mac is here. Typing is officially optional on your Mac. Talk in any app, and kivi turns what you say into clean, ready-to-send text. No more typing out long emails, messages or prompts. Here's everything kivi can do 🧵

https://x.com/heykiviai/status/2107076189714305429

Kivi is now live on Mac. Use your voice to get more done across your Mac. Powered by Saaras V4, our latest Speech to Text model. Try it at https://t.co/lTk84KBc7P https://t.co/rrVyDyUi6Q

https://x.com/SarvamAI/status/2107090415417786760
GladiaPlatformlaunch2026-07-30

Gladia releases GladiaFlow, an open-source desktop dictation tool

An STT vendor ships a local desktop workflow that turns its speech API into a general dictation surface, making the product useful beyond developers integrating a transcription endpoint.

Also Use

Details2 sources · 1 X post

Gladia released GladiaFlow, an open-source desktop voice-dictation tool that works in any text field with one hotkey and uses Gladia STT.

GladiaFlow

Today we're releasing GladiaFlow: an open-source desktop voice dictation tool that works in any text field, on any app.

https://x.com/gladia_io/status/2082825497134670090
Edge0Model providermodel release2026-07-30net-new

Edge0 open-sources ARK-ASR-3B multilingual speech recognition

Largest Edge0 ASR open weight before Audio8-ASR-0.1B and Infinite — vendor Open ASR English average WER 5.04% on the 3B card, Apache-2.0, nineteen languages.

Details5 sources · 1 X post

Edge0 released ARK-ASR-3B, an Apache-2.0 multilingual ASR model claiming a 5.04% average English WER on the Hugging Face Open ASR Leaderboard short-form suite across nineteen languages, alongside a smaller ARK-ASR-0.6B sibling.

ARK-ASR-3B

ARK-ASR-3B: #1 overall; ARK-ASR-0.6B: #5 overall. Both support 19 languages and are released under Apache 2.0.

https://x.com/SamuelZengML/status/2082800506003767590
OpenAIModel providermodel release2026-07-29

GPT-Transcribe and GPT-Live-Transcribe: context-aware ASR

Promptable ASR. Community cite for the Live variant: context-aware semantic accuracy 38.5% to 44.6% with free-form context.

Details2 sources

OpenAI introduced two context-aware transcription models, streaming GPT-Live-Transcribe and async GPT-Transcribe, that take prompts, keywords, and language hints.

GPT-Transcribe

NablaPlatformlaunch2026-07-28net-new

Nabla launches on-device medical dictation for Mac with Epic

On-device clinical dictation on Mac with Epic — privacy and latency pitch versus cloud STT, bundled with ambient rather than a separate vendor.

Also Use

Details3 sources

Nabla shipped Nabla Dictation for Mac, a fully on-device clinical dictation app built with Core ML and Apple silicon that pairs with Nabla ambient documentation and Epic workflows without sending audio to the cloud.

Nabla Dictation for Mac

Speaktext-to-speech7

Daily / PipecatPlatformAPI2026-08-01

Pipecat 1.7.0 adds Azure TTS force_locale

Forces Azure locale when auto language detection would pick the wrong voice pack.

Details3 sources

pipecat-ai 1.7.0 adds opt-in force_locale on AzureTTSService and AzureHttpTTSService, and reach_inactive_services on settings update frames.

pipecat-ai 1.7.0

Sarvam AIModel providermodel release2026-07-30

Sarvam previews Bulbul V4 expressive TTS with emotion and emphasis controls

Inline emotion and emphasis control reaches Indian-language TTS six months after Bulbul V3 (2026-02-05-sarvam-bulbul-v3, W06), but only as a preview with no model ID or price at 14 August; Eleven v4 inline tags (2026-09-28-elevenlabs-v4, W40) are the generally available comparison.

Details4 sources · 2 X posts

Sarvam presented Bulbul V4 at Epoch as a research preview of its text-to-speech model, letting developers steer emotion, emphasis, pacing, and cues such as laughter within a line across Indian languages.

Bulbul V4

Bulbul V4 is here! Our Text-to-speech model with richer emotion, natural expression and greater vocal range. Give it a listen: https://t.co/rtEJSwj5SP

https://x.com/SarvamAI/status/2082753872100299024

Presenting Bulbul V4 Our Text-to-speech model with richer emotion, natural expression and greater vocal range. Give it a listen: https://t.co/6zQzbTpfAi

https://x.com/SarvamAI/status/2082757345281782252
ElevenLabsModel providerlaunch2026-07-30

ElevenLabs adds Character Casting to ElevenCreative Audiobooks

Voice selection becomes a manuscript-aware casting workflow rather than a one-voice audiobook setting, increasing the practical usefulness of a large voice catalogue for multi-character books.

Also Use

Details2 sources · 1 X post

ElevenLabs launched Character Casting so audiobook creators can upload a manuscript, preview voices against real dialogue, and choose voices for every character.

Character Casting in ElevenCreative Audiobooks

Introducing Character Casting in ElevenCreative Audiobooks.

https://x.com/ElevenLabs/status/2082850828340064408
Edge0Model providermodel release2026-07-29net-new

Edge0 open-sources Audio8 TTS Preview 0.6B with zero-shot cloning

Compact DualAR zero-shot TTS under Apache 2.0 with a bundled 44.1 kHz codec — vendor Seed-TTS EN WER 1.506 / ZH CER 0.950 on the launch thread — and the speak-layer sibling to Edge0's later ASR Infinite filing (2026-09-23-edge0-audio8-asr-infinite).

Details5 sources · 1 X post

Edge0 released Audio8 TTS Preview 0.6B, an Apache-2.0 multilingual text-to-speech model with zero-shot voice cloning, a DualAR architecture, and a bundled 44.1 kHz neural codec across eleven recommended languages.

Audio8 TTS Preview 0.6B

Today, we're open-sourcing Audio8-TTS Preview: 11 languages. Zero-shot voice cloning. A bundled 44.1 kHz codec. Apache 2.0.

https://x.com/SamuelZengML/status/2082430003460166142
Fish AudioModel providerfunding2026-07-28

Fish Audio raises $52M seed led by Coreline Ventures and Capital Today

Dates the seed behind the already-filed S2 open-source and S2.1 Pro free-API ships — a creator-to-enterprise TTS lab with a named LiveKit/Retell integration plan, not a new model drop.

Details2 sources

Fish Audio announced $52 million in seed funding led by Coreline Ventures and Capital Today as it marked its first anniversary, citing $21 million ARR and more than 8 million users.

Fish Audio

Ringg AIPlatformlaunch2026-07-27

Ringg launches voice cloning for its agents

The feature adds a branded-voice path to Ringg's agent platform alongside its Parrot STT model and later multi-channel customer deployments; no quality or consent details are published in the launch post.

Also Use

Details1 source · 1 X post

Ringg launched voice cloning that lets teams upload a recording and create a clone for Ringg agents.

Ringg Voice Cloning

Introducing Voice Cloning on Ringg. Upload a recording to create a voice clone for your Ringg agents and deliver conversations that feel familiar from the very first hello! DM us or reach out at admin@ringg.ai to get started. https://t.co/czEHNTH1Fn

https://x.com/ringg_ai/status/2081783208505708963
KRAFTONModel providermodel release2026-07-27net-new

KRAFTON publishes the A.X K2 Raon-Speech speech model

The card confirms a speech-model release rather than only a company research claim, with both recognition and direct speech-generation paths in one checkpoint.

Also HearThink

Details1 source

KRAFTON’s A.X K2 Raon-Speech is a 21B-A3B speech model that handles speech understanding and direct speech generation.

A.X K2 Raon-Speech 21B-A3B

Showavatars, video1

Daily / PipecatPlatformAPI2026-08-01

Pipecat 1.7.0 adds LemonSlice local background images and faster Tavus audio

Avatar sessions start faster and look branded without remote background downloads.

Details3 sources

pipecat-ai 1.7.0 lets LemonSlice transport take a local image as the avatar background and defaults TavusParams.audio_out_faster_than_realtime to True.

pipecat-ai 1.7.0

ThinkLLMs, speech-to-speech3

Thinking Machines LabModel providermodel release2026-07-30net-new

Thinking Machines releases Inkling-Small, an open-weight model with native audio and image reasoning

An open-weight reasoning model brings audio into the reasoning loop and offers a fine-tuning path, giving voice builders another architecture besides a cascaded speech pipeline; no independent benchmark is added here.

Also HearSpeak

Details2 sources · 1 X post

Thinking Machines released Inkling-Small as an open-weight model with native reasoning over audio and images, variable thinking effort, and a checkpoint for fine-tuning.

Inkling-Small

PolyAIModel providermodel release2026-07-30

PolyAI Dialog-RSN-1: audio-native dialog, TTS left to a separate model

Contact-centre stacks that need a branded TTS can take audio-aware input without moving the call onto a speech-to-speech model.

Also Hear

Details1 source · 1 X post

PolyAI published Dialog-RSN-1, a request-based model that does turn-taking, ASR, function calling, and response, and leaves speech generation to a separate TTS.

Dialog-RSN-1

The traditional cascade loses the audio. Speech-to-speech loses control. Dialog-RSN-1 loses neither. We're introducing an entirely new approach to building voice agents, one that hears every caller's audio directly.

https://x.com/polyaivoice/status/2082819930383130879
xAIModel providermodel release2026-07-29

Grok Voice Think Fast 2.0: speech-to-speech at $0.08/min

Full-duplex STS at $0.08/min with an A/B test on the Starlink support line. Became grok-voice-latest on 5 Aug.

Details2 sources

xAI released Grok Voice Think Fast 2.0, a speech-to-speech model with 82.9% on AA STS, 0.70s time-to-first-audio, and $0.08/min pricing.

Grok Voice Think Fast 2.0

Orchestrateframeworks, evals4

Daily / PipecatPlatformAPI2026-08-02

Pipecat Context Hub 0.4.0 renames its Pipecat CLI bridge to pipecat context-hub

Anyone who scripted pipecat mcp from 0.3.0 has to change the command; the notes argue nothing depended on it yet and mention no compatibility alias for the old name.

Details1 source

Pipecat Context Hub 0.4.0 renames the Pipecat CLI bridge from pipecat mcp to pipecat context-hub, with pipecat ch as a shorter alias hidden from pipecat --help.

Pipecat Context Hub 0.4.0

TwinMindPlatformlaunch2026-07-30

TwinMind MCP opens meeting notes, transcripts, and tasks to ChatGPT and Claude

Read-only MCP access makes captured speech a context source for third-party assistants instead of a silo inside TwinMind's app, but only on the top Max tier. BotPhone MCP (2026-10-20-botphone-mcp-agent-phone-line) uses the same protocol to give agents a phone line, the opposite direction from TwinMind's retrieval.

Also Hear

Details1 source

TwinMind Max subscribers can connect ChatGPT or Claude to their TwinMind account and have the assistant search notes, pull full transcripts, and check action items, decisions, and deadlines.

TwinMind MCP

TavusPlatformlaunch2026-07-29

Tavus launches PAL Maker, a no-code builder for personal AI agents

Tavus lowers the construction boundary for its multimodal PALs from SDK or developer configuration to natural-language creation, broadening who can prototype an interactive voice-and-video agent.

Also ShowUse

Details1 source · 1 X post

Tavus launched PAL Maker, a no-code builder where anyone can describe a PAL and create an agent that sees, hears, acts, remembers, and emotionally understands.

PAL Maker

Alibaba / QwenPlatformlaunch2026-07-28

Alibaba's Tongyi speech team open-sources Qwen Audio Agent, a full-duplex voice layer for existing coding and task agents

Voice reaches Claude Code, Codex, and other coding agents here through ACP adapters with the agent kept in the background, and the default path is billed through a DashScope key for Qwen Audio Realtime, two weeks after Alibaba's Qwen-Audio-3.0 TTS (2026-07-14-alibaba-qwen-audio-3-0-tts, W29).

Also Think

Details4 sources · 1 X post

An Apache 2.0 Node.js runtime keeps a realtime voice conversation going while a backend agent such as OpenCode, Claude Code, Codex, or Hermes runs delegated tasks over ACP, with a macOS desktop app from version 1.0.0.

Qwen Audio Agent

Why do voice assistants still freeze the second they need to search or call a tool? Qwen-Audio-Agent was built to end that awkward wait—conversation and execution finally run side by side.

https://x.com/TONGYI_SpeechAI/status/2098339191897915645

Connecttelephony, channels1

Daily / PipecatPlatformAPI2026-08-01

Pipecat 1.7.0 adds LiveKit inbound SIP DTMF

LiveKit SIP calls can collect digits the same way Daily and RTVI already could.

Details3 sources

pipecat-ai 1.7.0 adds inbound SIP DTMF on LiveKitTransport via on_dtmf_event and InputDTMFFrame.

pipecat-ai 1.7.0

Useagents in market4

InfinitusPlatformlaunch2026-07-30

Infinitus launches AI-first hub for specialty pharma patient support

Packages Infinitus voice agents into a specialty-pharma hub SKU billed on outcomes and work performed, competing with legacy hub call centres rather than only selling payor diallers.

Also ThinkConnect

Details3 sources

Infinitus launched an AI-first, human-backed hub that uses voice and digital AI agents for specialty-therapy enrollment, benefits verification, prior authorisation, adherence, and after-hours support, with humans kept on judgment calls.

AI-first hub solution

YelpPlatformlaunch2026-07-28net-new

Yelp Host passes 1 million calls and adds OpenTable bookings and pickup ordering

Pickup ordering through Stream into Toast, Square, and Clover opens Yelp Host to counter-service restaurants for the first time; the 38% monthly growth and 80 saved weekly hours are Yelp and customer figures rather than independent measures.

Also Connect

Details2 sources

Restaurants in the US and Canada can use Yelp Host for OpenTable reservations and pickup orders, alongside its phone-answering service.

Yelp Host

Tata CommunicationsPlatformpartnership2026-07-28net-new

Tata Communications and Tata Tele Business Services bring unified AI infrastructure to Indian SMBs

The dated company release confirms the conditional product launch and puts speech-to-speech Voice AI inside an SMB distribution partnership.

Also SpeakOrchestrate

Details1 source

Tata Communications and Tata Tele Business Services joined forces to bring unified AI infrastructure and a speech-to-speech Voice AI platform to India’s SMB economy.

Unified AI infrastructure and Voice AI for India’s SMB economy · vendor Tata Tele Business Services

IAG (Insurance Australia Group)Enterprisepartnership2026-07-27net-new

IAG signs up for OpenAI Presence to answer claim calls during natural-disaster surges

IAG is the first named Presence customer filed after the W30 launch (2026-07-22-openai-presence), but in late July its July–December 2026 go-live was still at solution design, so this records a buying decision rather than a live claims line; the sources are trade press quoting IAG executives, not an IAG release. Insurance News ties the short rollout to Travelers, whose OpenAI-based claim assistant is filed as 2026-02-18-travelers-ai-claim-assistant.

Details3 sources

Insurance Australia Group will use OpenAI Presence for a customer-facing voice agent that takes claim lodgements, starting with high-volume natural-peril claims.

OpenAI Presence claims voice agent · vendor OpenAI