Hearspeech-to-text14
xAIModel providermodel release2026-09-18
xAI releases Grok Voice Transcribe 2.0 speech-to-text
Puts a hyperscaler STT refresh into the same price band as Transcribe 1.0, with a named Loom deployment and a timed 1.0 sunset — peers will pressure-test the vendor benches.
Details2 sources
xAI shipped Grok Voice Transcribe 2.0, a batch and streaming STT model it says is twice as accurate as Transcribe 1.0 at the same $0.10/hr batch and $0.20/hr streaming price.
Grok Voice Transcribe 2.0
Wispr Flow introduces Canto speech model for real-time dictation
Canto is a model-level addition to Wispr Flow's dictation product, with the published comparison focused on noisy real-world dictation rather than a general speech benchmark.
Also Use
Details1 source · 1 X post
Wispr Flow introduced Canto, a speech model for real-time dictation trained and evaluated on real-world dictations.
Canto
TwinMind for Desktop adds Windows to its dictation and meeting-capture app
Windows support gives TwinMind a desktop dictation reach that Mac-only tools such as SpeechFast (2026-10-18-speechfast-local-dictation-mac) lack, while Scrieb (2026-10-16-scrieb-1-0-offline-transcription) matches it on both platforms offline. The English-only Fastest mode against a 140-language Accurate mode is a speed-versus-coverage trade-off users have to pick by hand.
Also Use
Details3 sources
Mac and Windows users get one desktop app that pastes cleaned dictation at the cursor in any program and captures calls into notes, prompting them when a meeting starts.
TwinMind for Desktop
Treble raises $18M Series A-2 for voice and physical AI simulation
Amazon's Alexa team and Logitech are named users of Treble's simulated rooms, a sign that acoustic test data for far-field voice is becoming a funded category of its own at €36M raised.
Details2 sources
Iceland-based Treble Technologies raised an $18 million Series A-2 led by Paladin Capital Group, bringing total funding to €36 million.
Acoustic simulation platform
Speechmatics launches Agent STT powered by Linden 1
A dedicated agent STT surface from a long-running ASR vendor, with both Pipecat and LiveKit plugins available on day one and a public latency–accuracy claim that peers will pressure-test.
Details2 sources · 1 X post
Speechmatics released Agent STT, a speech-to-text API for production voice agents, powered by its new Linden 1 model with sub-350 ms finalisation, 55+ languages, live diarisation, and conversational events over `/v2/agent`.
Agent STT (Linden 1)
Pipecat 1.11.0 expands OpenAI, Whisper, Sarvam, AWS, and AssemblyAI STT knobs
STT latency/accuracy and vocabulary controls catch up across the major providers.
Details3 sources
pipecat-ai 1.11.0 adds OpenAIRealtimeSTT delay, OpenAISTT keywords, Whisper initial_prompt, Sarvam saaras:v4, AWS Transcribe partial_results_stability, and more AssemblyAI language_codes.
pipecat-ai 1.11.0
KrispPlatformpartnership2026-09-17
Krisp selects Soniox as its speech-to-text provider
The filing records a production-provider choice by an established voice-processing platform; it does not claim that Soniox replaced every Krisp speech component or publish a migration date.
Details1 source · 1 X post
Krisp selected Soniox to provide speech recognition behind Krisp’s real-time voice products.
Soniox speech-to-text provider · vendor Soniox
DeepgramModel providermodel release2026-09-17
Deepgram launches Nova-3 Pharma speech-to-text
Splits pharma entity accuracy from general medical STT — a measurable KRR claim competitors in clinical voice will be asked to match.
Details1 source · 1 X post
Deepgram released Nova-3 Pharma, a speech-to-text model trained for drug names and pharmaceutical vocabulary, available for batch and streaming on hosted and self-hosted deployments via model=nova-3-pharma.
Nova-3 Pharma
PindropPlatformlaunch2026-09-16net-new
Pindrop launches BotStopper to detect AI voice agents on enterprise calls
BotStopper adds an adversarial detection layer for calls involving synthetic or agentic voices; the announcement is about identifying bots, not building a voice agent.
Also Use
Details1 source
Pindrop launched BotStopper as a standalone product that detects AI voice agents during phone interactions.
BotStopper
AssemblyAI launches Dictation API for finished-text STT
Bundles transcript-plus-rewrite into one billable call for push-to-talk UIs — a different request shape from streaming agent STT already in the corpus.
Details3 sources · 2 X posts
AssemblyAI released the Dictation API, a sync endpoint that returns a verbatim transcript and an LLM-cleaned finished text from one short clip on Universal-3.5 Pro, priced at $0.62 per audio hour all-in.
Dictation API
Superwhisper launches its Android dictation app
This is the Android availability launch; the later floating-bubble feature filing is a separate product update, not a duplicate of the app launch.
Also Use
Details2 sources · 1 X post
Superwhisper made its voice-to-text dictation app available on Android, with a floating bubble for inserting polished speech into other apps.
Superwhisper for Android
Salesforce and AWS bring low-latency voice dictation to Slack on Trainium
The partnership puts Salesforce's speech stack into an employee-facing surface and links voice input to the same cross-cloud infrastructure push as the already filed contact-center handoff.
Also Use
Details1 source
Salesforce and AWS optimised Salesforce speech models on AWS Trainium to provide near-instant voice dictation inside Slack threads.
Slack voice dictation on AWS Trainium · vendor AWS Trainium
LiveKit Agents 1.8.2 adds ElevenLabs secondary languages for realtime STT
ElevenLabs multilingual realtime STT.
Details1 source
livekit-agents 1.8.2 adds secondary languages and language detection for ElevenLabs realtime STT, and migrates Speechmatics to Agent STT.
livekit-agents 1.8.2
Deepgram makes India voice endpoint generally available
Adds a managed onshore option after EU and Australia — relevant for Indian BFSI voice workloads that previously faced self-host or cross-border trade-offs.
Also Speak
Details1 source · 1 X post
Deepgram opened a generally available India regional endpoint at api.in.deepgram.com in AWS ap-south-2 (Hyderabad), with in-country storage and inference for STT, TTS, Voice Agent, and text intelligence at Global/EU/Australia pricing.
India endpoint (api.in.deepgram.com)
ThinkLLMs, speech-to-speech8
Alibaba ships Qwen3.8-Omni-Flash and a Realtime variant for live voice-and-camera agents
Alibaba shipped this beside Qwen-Audio 3.1 Realtime Plus (2026-09-20-alibaba-qwen-audio-3-1-realtime-plus) and Qwen3.8-LiveTranslate (2026-09-20-alibaba-qwen3-8-livetranslate) on the same day, so Model Studio offered two realtime voice-agent model lines at once; the claimed audio lead over Gemini 3.8 Flash rests on Alibaba's own evaluation.
Also HearSee
Details1 source
Qwen3.8-Omni-Flash reads text, images, audio, and video with a 1M-token context and calls tools on long audio-visual tasks; Qwen3.8-Omni-Flash-Realtime perceives and answers while live audio and video stream in.
Qwen3.8-Omni-Flash and Qwen3.8-Omni-Flash-Realtime
Alibaba ships Qwen3.8-LiveTranslate realtime interpretation
Vendor-claimed 2.3 s LAAL and speaker-attributed cloned speech on a 60/29-language realtime API — a direct peer for Gemini Live Translate-style interpretation workloads.
Details4 sources
Alibaba Cloud and the Qwen team released Qwen3.8-LiveTranslate, a WebSocket realtime interpretation model that understands 60 languages and speaks 29, with average lagging cut from 2.8 s to 2.3 s versus the prior generation.
Qwen3.8-LiveTranslate (qwen3.8-livetranslate-flash-realtime)
Alibaba ships Qwen-Audio 3.1 Realtime Plus for duplex voice agents
Makes 3.1 the documented default for Bailian realtime voice agents while leaving Flash as the cost-sensitive option; migration is a model ID and voice swap on the existing 3.0 protocol.
Details8 sources · 2 X posts
Alibaba Cloud Model Studio documents qwen-audio-3.1-realtime-plus as the recommended end-to-end speech-to-speech model for voice assistants and customer-service conversations, available in Singapore and China (Beijing).
Qwen-Audio 3.1 Realtime Plus
Pipecat 1.11.0 wires Gemini 3.8 Live and extended thinking
Gemini 3.8 Live becomes selectable the same week Pipecat UI lands for builders.
Details3 sources
pipecat-ai 1.11.0 adds gemini-3.8-live and gemini-3.8-live-extended-thinking on GeminiLiveLLMService with NON_BLOCKING tools and multi-turn_complete hold.
pipecat-ai 1.11.0
StepFunModel providermodel release2026-09-15
StepFun ships StepAudio 3 Realtime topping Full-Duplex Bench at 98.9
Shanghai lab posts a duplex S2S API family that leads AA conversational dynamics while still trailing peers on first-audio latency — the quality/latency trade is the story.
Also HearSpeakOrchestrate
Details5 sources
StepFun released StepAudio 3 Realtime, a full-duplex audio-language model with Think-While-Speaking and a Voice Agent, reporting 98.9 overall on Artificial Analysis Full-Duplex Bench alongside ASR Max, TTS, Gen, and Music siblings.
StepAudio 3 Realtime
GoogleModel providermodel release2026-09-15
Google ships Gemini 3.8 Live and Extended Thinking on the Live API
A hyperscaler S2S drop with async tools + visual grounding, same day both Pipecat and LiveKit wire it — the competitive reference for GPT-Live-1-class agent stacks, with Artificial Analysis and τ-Voice numbers that need independent checking.
Details3 sources · 5 X posts
On 15 Sep Google released Gemini 3.8 Live and 3.8 Live Extended Thinking for native speech-to-speech on the Gemini Live API and AI Studio, with async tool calls, visual grounding, and mid-conversation switching across 97 languages.
Gemini 3.8 Live and 3.8 Live Extended Thinking
DeepLModel providermodel release2026-09-15
DeepL Voice adds real-time voice preservation across languages
Moves meeting translation from shared synthetic voices to per-speaker identity — a different product axis from agent STT/TTS already tracked here.
Details3 sources · 1 X post
DeepL shipped new Voice models that preserve each speaker's voice, tone, rhythm, and expression during live multilingual translation, initially across 14 languages, plus a desktop app for Zoom, Teams, and Meet.
DeepL Voice (voice preservation)
Nuance Labs raises $50M Series A for full-duplex face-to-face AI
Funds a single-model face-to-face duplex bet against cascaded avatar stacks — research preview still ahead, but NVIDIA joined the round.
Also ShowSpeak
Details3 sources
Seattle’s Nuance Labs raised a $50 million Series A led by Lightspeed, with Accel, South Park Commons, NVIDIA, and Define Ventures, to build one full-duplex audiovisual model for face-to-face conversation.
Nuance Labs foundation model
Orchestrateframeworks, evals8
Pipecat walkthrough: Jev beats GPT-5.6 Luna on voice UI operator accuracy and latency
A dated head-to-head on the operator slot justifies why later multi-character and Unreal demos keep Jev on the control path instead of a general LLM tool-caller.
Also Think
Details1 source · 1 X post
Kwindla (@kwindla) amplifies Jon Taylor's walkthrough of Typesafe Jev versus GPT-5.6 Luna as the operator in a Pipecat speech-interface pipeline: Jev at 92.6% command accuracy / 296 ms median latency vs Luna at 81.3% / 1,008 ms (Luna without reasoning).
Jev vs GPT-5.6 Luna operator walkthrough · vendor Typesafe
.@jonptaylor recorded a detailed walkthough of Jev vs GPT-5.6 Luna as the "operator" element of a Pipecat speech interface pipeline.
GPT-5.6 Luna:
- 81.3% command accuracy
- 1,008 ms median latency
Jev
- 92.6% command accuracy
- 296 ms median latency
A few notes here ...
1) We expected to see a big latency benefit. But the higher accuracy is maybe more interesting. Jev (with a bunch of code wrapped around it) is much better at turning messy transcriptions from a noisy speech environment into correct command structs.
The very hand-wavy explanation here is something like: LLM tool calling is an amazingly flexible, general capability but is only one thing that LLMs are trained to do. Jev is trained to do a narrower thing: to produce probabilities over a fixed set of options.
If we can express our tool calling requirements as fixed set of choices, it's perhaps not surprising that the model trained specifically and only to evaluate choices does a better job at that task.
Some keywords to talk to your friendly neighborhood LLM about this topic, if you're interested in a deeper dive, are: discriminative vs generative models, calibrated probabilities, constrained decoding, and closed-set vs open-set prediction.
Note that this is GPT-5.6 Luna without reasoning enabled. With reasoning, today's LLMs are going to do much better on evals like this. Reasoning lets models build a scratch pad of options, which is massively helpful in classification tasks. But we generally can't use models in reasoning/thinking mode for applications like voice agents or voice user interfaces, because the latency is too high.
2) An application-specific eval lets us make informed choices about whether we're getting measurable benefits from different models and different architectures. It's gotten a lot easier to build pretty good, very specific evals as part of agent development. Fable and Astra know how to do this and mostly get things right without a lot of guidance.
3) Using Jev feels more like programming than using an LLM does. Mostly, with LLMs, we're doing "context engineering." Which is just a fancy way of saying we're trying to feed the LLM a big pile of natural language and hoping we're aiming for the right part of the latent space.
In most of our sophisticated agents, there's a lot of code to do this context engineering: constant summarization, switching prompts for different parts of the conversation, etc. But the end result is always just really long prompt strings!
With Jev, we're sending the model natural language, but in addition we're doing very explicit mapping between application state and Jev's input data structures.
https://x.com/kwindla/status/2101394993957274021
TavusPlatformlaunch2026-09-18
Tavus adds MCP Connectors so PALs can take actions during conversations
Tavus moves PALs from conversational guidance into authenticated multi-step work across external applications, giving the avatar agent a dated action layer rather than only a response surface.
Also ConnectUse
Details2 sources
Tavus added MCP Connectors so a PAL can access apps such as Gmail, Google Calendar, and Linear, or a third-party MCP server, and take real actions during a conversation.
MCP Connectors for PALs
Pipecat demo hands voice UI control to Jev and generations to PhoneLLM
Splitting command/UI control (Jev) from conversational generation (PhoneLLM) is the operator pattern later scaled to multi-character and Unreal NPC demos — filed for the control handover, not Jamcat music generation.
Also Think
Details1 source · 1 X post
Jon Taylor (@JonPTaylor) demos Pipecat orchestrating a realtime handover between a Typesafe Jev command agent driving the UI and a PhoneLLM voice agent for generations, shown inside his Jamcat session (UI/control path — not a music-model ship).
Jev voice-driven UI handover demo · vendor Typesafe
Wow, @typesafeai Jev is a game changer for voice-driven UI. Responsiveness really matters for an experience to 'feel' right, and this certainly does the job!
Latency here is not optimal given I'm in London and calling out to us-west. But still, super snappy.
In this video, Jamcat is using Pipecat to orchestrate a real-time handover between a command agent (Jev) that drives the UI, and voice agent (PhoneLLM) for handling generations and (not in the video) discussing ideas and session setup with the user.
https://x.com/JonPTaylor/status/2100736122502390211
Pipecat demo runs the same agent code fully on-device on iOS 26
Same-code on-device iOS is a stronger portability proof than cloud-only examples — hear/think/speak stay local while orchestration abstractions still reach cloud models when needed.
Also HearThinkSpeak
Details3 sources · 2 X posts
Kwindla (@kwindla) posts Jon Taylor walking through an on-device Pipecat iOS agent using Apple Speech Analyzer STT, Apple Foundation Models on-device LLM, and Kyutai PocketTTS, with the same Pipecat code path as cloud agents via pipecat-build-ios.
Pipecat on-device iOS agent demo · vendor Apple
You can run @pipecat_ai voice agents entirely locally on iOS.
Here's @JonPTaylor talking through how this works. Jon demonstrates an on-device agent that uses the Apple Speech Analyzer transcription model, the Apple Foundation Models on-device LLM, and @kyutai_labs PocketTTS.
This is running on iOS 26. The transcription model is quite good. The LLM is ... small. But if you're using the iOS 27 beta, you can experiment with the next generation of the Apple Foundation Models, too. There's definitely progress happening in these ~3B parameter models.
The key thing to me, though, is that the code on iOS is the same Pipecat code you run in the cloud, or on a bigger device.
That means you can use Pipecat distributed agents to build applications that leverage multiple models, you can route tasks to a bigger LLM in the cloud, you can use Pipecat Flows for conversation design. All of the 180+ services and architecture tools in Pipecat are available for local realtime AI development on a phone.
https://x.com/kwindla/status/2100719640481804752
https://x.com/kwindla/status/2100719642256048400
Pipecat 1.11.0 expands eval suites with multi-scenario files and llm_marker
Release eval packs become compact multi-case files with marker-aware assertions.
Details3 sources
pipecat-ai 1.11.0 lets scenario files hold many cases under scenarios:, adds llm_marker expectations, function_call eval judges, and manifest name/concurrency/data fields.
pipecat-ai 1.11.0
Hume AIModel providerother2026-09-17
Hume publishes Voice Controllability Leaderboard for TTS direction-following
Separates identity cloning from controllability. Same model can win motivational casting and lose meditation; Fish's voice-design swings are the clearest example in the board.
Details2 sources · 1 X post
Hume published a Voice Controllability Leaderboard scoring 17 TTS models on voice design, instruction-following, inline tags, and role fit with blind human raters across 13 languages.
Voice Controllability Leaderboard
A result from the Voice Controllability leaderboard we published today:
Fish's voice design model won nearly every head to head for motivational speaker voices and lost nearly every one for meditation voices. Same model, same prompt format, and the job you're casting for flipped
https://x.com/hume_ai/status/2100584836864024785
SoundHound ships Human Assisted Resolution inside OASYS
Separates brief human judgment from escalation — a concrete HITL pattern for branded voice/chat agents that otherwise double-pay for transfer.
Details1 source · 1 X post
SoundHound launched Human Assisted Resolution (HAR), an opt-in OASYS feature that lets an AI agent ask a colleague a specific question in real time and finish the conversation without a full handoff.
Human Assisted Resolution (HAR)
ElevenLabs adds call queueing and hold audio to ElevenAgents
Turns concurrency limits from hard rejects into a held queue — material for telephony agents that spike above seat count during campaigns.
Also Connect
Details2 sources
ElevenLabs shipped concurrency call queueing for ElevenAgents: callers over the agent limit hear hold audio and receive queue_status events until admitted or timed out, with APIs to upload custom hold clips.
ElevenAgents call queueing
Useagents in market16
Vocaleo Bridge lets a calling agent hold the other party and patch in its user
Escalation runs from the consumer's own agent back to the consumer, the reverse of a contact-centre transfer to a human rep; per-call opt-in and a fixed verified number are the guardrails, and Pro Mode calls on GPT-Live-1 (2026-09-13-vocaleo-pro-gpt-live-1, W37) cannot use it.
Also Connect
Details2 sources · 2 X posts
When the user has allowed it for that call, the agent rings the owner's verified number, briefs them privately, and connects the two calls only after they agree.
Vocaleo Bridge
Pipecat 1.11.0 switches init clients to Pipecat UI
The default builder-facing client is now the Pipecat UI console product.
Details3 sources
pipecat-ai 1.11.0 makes pipecat init React clients use Pipecat UI (ui.pipecat.ai) instead of voice-ui-kit, with console connect, transcript, metrics, and live events.
pipecat-ai 1.11.0
MetaModel providerlaunch2026-09-17
Meta Muse beta adds outbound calls to US businesses
Consumer agent stacks now dial merchants, not just chat — product parity with Instinct Concierge (2026-09-16-instinct-concierge-calling), filed from TechCrunch plus the Muse product lead's X post rather than a Meta blog.
Also ThinkConnect
Details2 sources · 1 X post
Meta expanded the Muse beta so the agent can place outbound phone calls to US businesses, rolling first to users who had already asked Muse for calling.
Muse outbound business calling
we just expanded the @Muse beta for outbound calls to US businesses, prioritizing folks who'd asked their Muse to let us know they wanted it first :)
if you're in, give it a shot and tell us what you think. phone calling was one of our top requests and your feedback helped make it happen! 📞
https://x.com/wailord/status/2100342273854894533
Betterbot ships multifamily Voice AI with a native Leasing Dashboard CRM
Multifamily operators get phone in the same agentic context as chat and SMS on 17 September, with Entrata/Yardi/RealPage sync — a vertical peer to EliseAI-class housing voice rather than a greenfield CCaaS.
Also Connect
Details2 sources
Betterbot opened a new Voice AI lane and a Leasing Dashboard CRM so one agentic stack answers, qualifies, and routes multifamily prospect and resident calls alongside chat, SMS, and email.
Voice AI and Leasing Dashboard
Assort expands direct NextGen Enterprise EHR actions for specialty practices
Moves vertical voice agents past intake into write-back scheduling and referrals on a major ambulatory EHR — the bottleneck after the phone is answered.
Details2 sources
Assort Health announced expanded Platinum API Tier access so its AI agents can read appointment slots and create, cancel, or reschedule appointments directly in NextGen Enterprise EHR, plus referrals, notes, and payment workflows.
NextGen Enterprise EHR direct integration (expanded) · vendor NextGen Healthcare
MarchexPlatformlaunch2026-09-16net-new
Marchex turns a paid AI Voice Agent pilot into a commercial deployment with an advertising and media customer
Call-analytics incumbents are selling voice agents into their installed base, as Insight7 later did (2026-10-15-insight7-voice-ai-agents, W42); the $400,000 figure covers the whole customer relationship, so it does not price the agent itself.
Details2 sources
The unnamed customer's agent qualifies prospective buyers on calls when sales representatives are unavailable, and Marchex says it has signed further customers and is in pilot talks with several large enterprises.
Marchex AI Voice Agent
Instinct Concierge lets the personal agent place outbound phone calls
Consumer text agents that already run email and trusted-network assistant-to-assistant contact now dial merchants for tasks with no web form — same week Meta Muse opened US business calling (2026-09-17-meta-muse-outbound-calls).
Also Connect
Details3 sources · 1 X post
Early-access Concierge has Instinct dial businesses for high-touch errands such as offline restaurant bookings, dentist cancellation lists, and cable-bill disputes.
Instinct Concierge
ElevenLabs launches Reception, an AI receptionist for small businesses
Packages ElevenAgents as an SMB receptionist SKU (calendar + after-hours) against dedicated receptionist vendors, not only API buyers; HIPAA stays on the broader platform.
Details2 sources · 1 X post
Reception by ElevenAgents answers inbound calls, books appointments from a website scan, and goes live with a phone number in minutes; plans start at $29/month with 70+ languages.
Reception (Reception.ai by ElevenAgents)
Aristotle launches nationwide voice-first AI tutoring with a $5M True Ventures seed
A consumer voice tutor sold to parents at human-tutor prices — the bet is productive struggle via live voice + whiteboard, not another homework chatbot; outcomes evidence still pending beyond engagement metrics.
Details4 sources
Aristotle opened nationwide access to its voice-first AI tutor for students roughly ages 13–18 after a closed beta of 1,000+ students and 1,500+ tutoring hours, alongside a $5 million seed led by True Ventures.
Aristotle voice tutor
AmazonPlatformlaunch2026-09-16
Amazon launches Alexa+ in India with English, Hindi, and Hinglish code-switching
This is the first Alexa+ market where Amazon leads with code-mixed speech rather than a localized voice, putting a mass-market consumer assistant on the same Hindi/Hinglish ground as Sarvam's Indic voice stack; the rest of the 2026 country rollouts (UK 19 March to Australia 6 August) were Early Access expansions of the same product and are deliberately not filed.
Also ThinkHear
Details2 sources
Alexa+, Amazon's generative-AI assistant, began rolling out across India on 16 September through a free Early Access program for eligible Echo and Fire TV owners; it speaks English, Hindi, and Hinglish and handles switching languages within a single sentence.
Alexa+ (India Early Access)
State Bar of Nevada puts Abby Connect AI Receptionist on its lawyer-referral line
This is a dated public-sector deployment where voice intake and browser automation work around a client system with no public API, while human escalation remains the safety net.
Also OrchestrateConnect
Details2 sources
The State Bar of Nevada combined Abby Connect’s AI Receptionist with its human team to automate lawyer-referral intake around the clock without changing its existing software.
Lawyer Referral and Information Service AI Receptionist · vendor Abby Connect
Smith.ai makes its full AI Receptionist free for 25 answered calls a month
A real forever-free production tier lowers the trial barrier for small service businesses while preserving a paid human-escalation path; the 25-call cap and per-call overage still make it a starter plan rather than unlimited coverage.
Also ConnectOrchestrate
Details2 sources · 1 X post
Smith.ai launched a free-forever AI Receptionist plan that answers 25 real calls a month with the same core intake, routing, bilingual, integration, and testing capabilities as paid plans.
AI Receptionist Free plan
Hello Patient acquires Converse Health for referral and back-office AI agents
Moves a voice-first patient-access vendor into EHR document workflows, so outbound referral calls sit on the same agent stack as inbound phones rather than a bolted-on fax tool.
Details3 sources
Hello Patient bought Converse Health so its patient-conversation agents can also do referral and fax intake, chart-driven follow-up, authorisation paperwork, and records work inside outpatient EHRs.
Converse Health acquisition (back-office agents)
AmbiPlatformlaunch2026-09-15net-new
Ambi launches an ambient-intelligence app for context and memory
The app is voice- and context-led, but the public launch material does not publish adoption or accuracy figures; early access and hardware availability are separate availability constraints.
Also Hear
Details2 sources · 1 X post
Ambi introduced an ambient-intelligence system that listens to permitted context, builds memory, and helps turn a user’s intentions into ongoing work.
Ambi
Superhuman acquires Fathom and brings its AI meeting notetaker into the Superhuman suite
Fathom's transcripts become context for Superhuman Go's agents inside a suite that already includes Grammarly and Mail, a different route for a meeting notetaker from Otter.ai's $320 million non-equity growth financing a month later (2026-10-20-otter-ai-320m-cvf-financing, W43); deal terms and any change to Fathom's standalone plans were not disclosed.
Also Hear
Details2 sources · 1 X post
Superhuman bought Fathom, whose meeting capture, transcripts, live summaries, and CRM integrations will feed Superhuman Go and its agents across email, calendar, documents, and databases.
Fathom acquisition
ApplePlatformlaunch2026-09-14
Apple releases Siri AI in beta with expressive voices and on-device dictation upgrades
Apple's assistant now runs on Apple Foundation Models built with Google's Gemini, the same week Gemini 3.8 Live reached the Live API (2026-09-15-google-gemini-3-8-live). Apple publishes no latency or accuracy figures for the new voices or dictation, and the beta is English only.
Also SpeakHear
Details1 source
Apple's 2027 software releases bring a rebuilt, conversational Siri with personal context, onscreen awareness, app actions, and a dedicated Siri app, rolling out in an English beta on Apple Intelligence devices.
Siri AI