hotwirednews Install bot

Week 32

3–9 Aug 20262026-W32 · 11 filings · 9 companies · 2 net-new

The Speak layer led the week with 3 of 11 filings, from LiveKit, ElevenLabs and Bland AI.

Hearspeech-to-text2

LiveKitPlatformlaunch2026-08-07

LiveKit Agents 1.6.9 adds Baseten Qwen3-ASR and OpenAI STT context hints

Qwen3-ASR and STT context hints. Class-2: folded quiet patch(es) 1.6.2, 1.6.5, 1.6.6 into this row.

Details4 sources

livekit-agents 1.6.9 adds Baseten Qwen3-ASR STT, OpenAI STT context hints, and Google credentials objects on STT/TTS.

livekit-agents 1.6.9

Edge0Model providermodel release2026-08-06

Edge0 open-sources Audio8-ASR-0.1B multilingual on-device speech recognition

Puts a ~0.1B-LM multilingual ASR stack on phone/ONNX paths months before Edge0's streaming Infinite card (`2026-09-23-edge0-audio8-asr-infinite`) — vendor Open ASR mean WER 7.03 and ~200 MB iOS peak claim, CC-BY-NC-4.0 weights.

Details7 sources · 1 X post

Edge0 released Audio8-ASR-0.1B, a compact autoregressive multilingual ASR model with about 0.1B language-model parameters for English, Chinese, Cantonese, French, German, Japanese, and Korean, plus ONNX Runtime and iOS Apple Neural Engine deployment builds.

Audio8-ASR-0.1B

Meet Audio8-ASR-0.1B.Small model. Full-scale capability.

https://x.com/SamuelZengML/status/2085321310067134527

Speaktext-to-speech3

LiveKitPlatformlaunch2026-08-07

LiveKit Agents 1.6.9 adds Deepgram Flux TTS, Bland TTS, and Baseten Qwen3-TTS

Large TTS provider wave in one cluster.

Details1 source

livekit-agents 1.6.9 adds Deepgram Flux TTS (v2/speak), a Bland TTS plugin, Baseten Qwen3-TTS, and Fish Audio quality-guard on all TTS requests.

livekit-agents 1.6.9

ElevenLabsModel providerAPI2026-08-06

Dubbing v2 opens on the ElevenAPI

Studio dubbing as an API call across 90+ languages, with delivery conditioned on the original performance.

Details1 source · 1 X post

ElevenLabs opened Dubbing v2 on the API so developers can embed performance-preserving dubbing across 90+ languages.

Dubbing v2

Dubbing v2 is now available on ElevenAPI, enabling developers to embed AI dubbing directly into their products and workflows.

https://x.com/ElevenLabs/status/2085383576959197680
Bland AIModel providerlaunch2026-08-04net-new

Bland Speech v3: TTS trained on 100M+ real phone conversations

Phone-conversation training set plus a named Audio Realism bench. After starter credits, $0.015 / 1,000 characters.

Details1 source

Bland launched Speech v3, a TTS trained on 100M+ real phone conversations, claiming second place to real humans on Design Arena Audio Realism.

Bland Speech v3

ThinkLLMs, speech-to-speech2

Boson AIModel providermodel release2026-08-04

Boson AI ships Higgs Realtime speech-to-speech API at production price

Full-duplex S2S with OpenAI Realtime drop-in at sub-$0.02/min audio-out — Boson's first filed realtime agent model after the TTS/avatar line.

Also Speak

Details1 source

Boson AI launched Higgs Realtime, an audio-native speech-to-speech model and OpenAI-Realtime-compatible API with ~700 ms speech-in to speech-out, 100+ languages, and metered rates of $0.0023/min in and $0.014/min out.

Higgs Realtime

TelnyxPlatformAPI2026-08-03

Telnyx AI Assistants add WebSocket voice conversations without telephony

Decouples Telnyx voice agents from PSTN while keeping Mission Control assistant definitions; complements same-day Edge Compute filing.

Also OrchestrateConnect

Details1 source

Telnyx opened a direct WebSocket conversation endpoint for AI Assistants so clients can stream PCM16 audio and receive transcripts, assistant audio, and tool-call events without a phone number or PSTN hop.

Voice AI WebSocket conversations

Orchestrateframeworks, evals2

SierraPlatformlaunch2026-08-07

Sierra Voice Personas: brand voice, pacing, and mid-call language switch

Brand voice is a control on the agent platform here, rather than a TTS dropdown in another vendor’s console.

Also Speak

Details1 source · 1 X post

Sierra launched Voice Personas so one agent can carry brand-specific personality, pacing, interruption behaviour, and mid-call voice or language switching, available on Sierra that day.

Voice Personas

TelnyxPlatformlaunch2026-08-03

Telnyx launches Edge Compute so voice agents run next to speech and calls

CPaaS vendor closes the gap between telephony and agent runtime; comparable connect+orchestrate move to Twilio Agent Connect-class ships.

Also Connect

Details2 sources

Telnyx made Edge Compute generally available so agent code, state, inference, speech, and telephony can run on one Telnyx network with under-200ms round-trip targets.

Edge Compute

Useagents in market2

CarMaxEnterpriselaunch2026-08-06net-new

CarMax puts Sierra voice agents on inbound sales calls

A national used-car retailer on inbound sales voice is a US enterprise proof point next to SoftBank’s Japan channel deal, and it still arrived via a trade-press write-up.

Details1 source

CarMax said Sierra voice agents have been live on inbound sales calls since May, answering hours and availability questions and handing off to associates; appointment scheduling is next.

Inbound sales voice agents (Sierra) · vendor Sierra

HyroPlatformpartnership2026-08-05

Hyro adds voice automation on ServiceNow Healthcare Operations

Second major 2026 Hyro platform wire after Five9 — voice agents into enterprise IT/ops workflows, not only patient scheduling.

Also ConnectOrchestrate

Details2 sources

Hyro partnered with ServiceNow so health systems can reach ServiceNow Healthcare Operations workflows over Hyro voice and digital agents, starting with employee FAQs, conversational ticket management, and IT desk voice ticket intake via a ServiceNow Store app.

AI Agents × ServiceNow Healthcare Operations · vendor ServiceNow