Events 🎤
Audio Layer hosts “Voice x Robotics,” an SF meetup with live robot demos (Sep 15, San Francisco, LUMA)
StepFun unveils StepAudio 3 at a live SF meetup, with real-time demos and a builders’ panel (Sep 16, San Francisco, LUMA)
Regal hosts Regal Rise, its flagship conference on rebuilding customer interactions around AI voice agents. (Sep 17, NYC + virtual, LUMA)
Top Updates 💪
Apple’s Watch normalizes always-on listening, with Siri Recap ambiently noting conversations through the day and drawing fresh privacy scrutiny. (CNET)
Meta Muse Voice Transcribe is built for always-on AI glasses, with an 80ms listen-or-write loop that raises bystander-consent questions as it diarizes whole rooms. (shattered.io)
OpenAI brings GPT-Live-1 to the API at $0.05 a minute, giving developers its full-duplex voice model plus 12 new real-time voices. (OpenAI)
Microsoft ships Azure AI Speech LLM 2607, cutting latency up to 3x with better mixed-language accuracy for enterprise transcription. (Microsoft)
Listen Labs walks away from a $1.5B round amid talks for Salesforce to acquire the voice-AI research startup for about $2B. (TechCrunch)
Rime partners with Dialpad as the premier text-to-speech provider behind all of Dialpad’s agentic voice AI products. (WebWire)
PolyAI ships Wren, a coding agent inside Dialog Studio that drafts a first version of a voice agent from a natural-language description. (AI Magazine)
Observe.AI launches Performance Agents, connecting conversation analysis, coaching, and performance measurement for contact centers. (CMSWire)
Amazon Prime Video adds AI lip-sync dubbing, altering on-screen mouth movements to match English dialog on a German-language series. (NYT)
Bodhan AI, IIT Madras, and NVIDIA release open Indic models covering speech, translation, text-to-speech, and OCR across 20+ languages. (Free Press Journal)
Krisp reframed as a contact center platform, extending from noise cancellation and notes into QA, compliance, translation, and voice security. (The Globe and Mail)
Blue Machines AI launches Aurora, a BFSI-native speech-to-text model built for India’s code-switching, noisy telephone conversations. (Economic Times)
AudioCodes bets big on voice AI as conversational AI grows over 50% YoY, even as its cash reserves decline. (Insider Monkey)
SAINS and 8nabler sign an MoU to expand voice AI in Sarawak public services, after handling 4,000+ ICT support calls autonomously. (Borneo Post)
Viaim launches the Rise AI agent earbuds at IFA 2026, a “Listen, Think, Act” hearable that turns conversations into project context. (KLGadgetGuy)
SCMP rebuilds its article narration in-house, replacing external providers with its own AI text-to-speech for natural voice. (SCMP)
A new breed of conversational AI is reviving voice in customer service, as agentic systems move past scripted bots. (Current Analysis)
Engineering Corner 😎
ARTPARK and IISc release the VAANI Noise Event Dataset, real-world spontaneous speech with timestamped noise across 100+ Indian languages. (Hugging Face)
A dev.to walkthrough drives macOS workflows with Gemini 3.5 Transcribe, combining spoken instructions with on-screen context. (dev.to)
Candor-LR is a new audio-visual speech recognition benchmark built from 1,656 natural dyadic video calls with overlapping, unscripted speech. (arXiv)
A paper tackles Korean ASR error correction for call centers, introducing a large dialogue-level benchmark for text-based post-editing. (arXiv)
A Frontiers study fine-tunes Wav2Vec2 and Whisper for Telugu, benchmarking speech-to-text on a low-resource Indian language. (Frontiers)
HackerNoon covers Dr. Sunday David Ubur’s affective architecture, arguing STT discards the acoustic cues that carry human intent. (HackerNoon)
A “self-listening” paper targets voice AI interruptions, giving agents a way to monitor their own output to handle barge-in. (AI Weekly)
Audio8 releases TTS Preview 0.6B, a compact multilingual model with zero-shot voice cloning across 11 languages. (HackerNoon)
Aqua Voice targets professionals with the Avalon model, a voice-to-text tool that beats Whisper and Scribe on technical and coding vocab. (Dynamic Business)