hotwirednews Install bot

Week 2

5–11 Jan 20262026-W02 · 5 filings · 4 companies · 4 net-new

The Orchestrate layer led the week with 2 of 5 filings, from Daily / Pipecat.

Hearspeech-to-text1

ElevenLabsModel providermodel release2026-01-09net-new

ElevenLabs ships Scribe v2 batch STT with keyterm prompting and entity detection

Batch STT sibling to realtime Scribe — keyterm prompting and 56-category entity timestamps target subtitle and compliance pipelines rather than agent turn latency.

Details2 sources

ElevenLabs released Scribe v2 for batch transcription, subtitling, and captioning, with keyterm prompting (up to 100 terms), native entity detection across up to 56 categories, automatic multi-language transcription, and speaker diarization.

Scribe v2

Announcementelevenlabs.io

Speaktext-to-speech1

Boson AIModel providermodel release2026-01-09net-new

Boson AI ships Higgs TTS 2.5 as a 1B-parameter production voice model

1B TTS that claims to beat the prior 3B on speed and SEED-TTS WER — Foundry catalog path matters for Azure-bound voice agents.

Details1 source

Boson AI released Higgs TTS 2.5, shrinking the prior 3B Higgs TTS 2 stack to 1B parameters while claiming faster first-token audio (~150ms vs ~200ms), stronger voice cloning, and expressiveness control tags.

Higgs TTS 2.5

Announcement

Orchestrateframeworks, evals2

Daily / PipecatPlatformlaunch2026-01-08net-new

Pipecat Cloud reaches general availability for managed voice agents

Managed Pipecat scale with Daily’s realtime network underneath, while the same agent code still self-hosts — the nine-month beta already claimed more than 1,000 teams before GA on 8 January 2026.

Also Connect

Details2 sources

Daily made Pipecat Cloud generally available after a nine-month beta with more than 1,000 teams, as a vendor-neutral managed host for open-source Pipecat agents.

Pipecat Cloud

Announcementdocs.pipecat.ai
Daily / PipecatPlatformmodel release2026-01-07net-new

Pipecat Smart Turn v3.2 cuts short-utterance errors and adds noise robustness

Short yes/okay turns and noisy rooms are where agents interrupt or stall — a 40% short-utterance error cut on the public bench is the practical gain over v3.1.

Also Hear

Details2 sources

Pipecat released Smart Turn v3.2, an open turn-detection model with about 40% fewer short-utterance misclassifications and cafe/office noise added to training.

Smart Turn v3.2

Announcementhuggingface.co

Useagents in market1

PrestoPlatformfunding2026-01-08net-new

Presto adds $10M led by Metropolitan Partners Group for QSR Voice AI

Puts a January 2026 primary on Presto’s production-scale drive-thru claim — hundreds of live sites and millions of 2025 interactions — before the later Remus-affiliate top-up.

Details1 source

Presto Phoenix added $10 million led by Metropolitan Partners Group, with Remus Capital, Link Ventures, and angels including the CEO of ElevenLabs, to scale live Presto Voice drive-thru deployments.

Presto Voice

Announcement