The Use layer led the week with 36 of 83 filings, from RuleInside, Laxis, Estaluna, Superwhisper and 29 more.
A new STT provider and the Voxtral 3 model family widen LiveKit's hearing path beyond the 1.8.9 Azure pronunciation-assessment work; both additions are release-level integration surfaces rather than maintenance-only changes.
LiveKit Agents 1.8.10 adds Reson8 batch and realtime STT with server-side turn detection and Mistral Voxtral 3 transcription models.
livekit-agents 1.8.10
Keeping audio on the PC addresses confidentiality for journalists, clinicians, lawyers, and researchers, but the product publishes no independent accuracy or diarisation result.
Also Use
SoundWiz transcribes recordings, meetings, and interviews on a Windows PC with speaker labels, timestamps, and no cloud upload.
SoundWiz
Offline recognition and command insertion make Sorimac a direct voice interface for writing and coding, while the site publishes no independent accuracy comparison.
Also Use
Sorimac turns speech into text at the cursor in Mac apps, with offline Qwen3-ASR and Whisper recognition, command controls, and 21 languages.
Sorimac
The 28 October 2026 release gives a Pipecat voice agent both a five-second audio-quality monitor and an explicit user-tone signal, while CPU model downloads, perception confidence, and alpha expressive-conversation APIs leave deployment and interpretation caveats.
Pipecat 1.15.0 adds AI-coustics Tyto audio analysis, Pipecat Tone reading of user emotion and sarcasm, and a perception field that carries those signals into the agent context.
pipecat-ai 1.15.0
One promptable checkpoint covers transcription, diarisation, translation, and audio questions, while NVIDIA reports 5.33% average English WER and about 4 billion active parameters; those benchmark figures are vendor-published and the available checkpoint is for recorded rather than live audio.
Also Think
Nemotron 3.5 Transcribe turns recordings into prompted transcripts, speaker labels, translations, or audio answers across 40 languages and is available on Hugging Face and ModelScope.
Nemotron 3.5 Transcribe
Meet NVIDIA Nemotron 3.5 Transcribe. An open-weight, promptable 30B MoE model with ~4B active parameters that brings transcription in 40 languages, speaker attribution, translation, contextual prompting and audio understanding into one checkpoint.
https://x.com/NVIDIAAI/status/2115462300241141882
Prisma v3.0 puts code-mix scripts, language detection, and customer vocabulary in the ASR request, while Gnani reports a 16.93 macro WER on Vistaar against 19.98 for ElevenLabs Scribe v2; the comparison is vendor-run.
Gnani AI's Prisma v3.0 recognises ten languages, including nine Indian languages and English, with code-mix, keyterm, and output-format controls for calls, recordings, and voice notes.
Prisma v3.0
Introducing Gnani Prisma v3.0, built for Indian calls that switch between Hindi, English and Kannada in a single sentence, run over busy streets and weak connections, and use brand names a model was never told about.
https://x.com/GnaniAi/status/2115337828699943189
Character-level timing gives transcription, dubbing, and lyric products one alignment layer instead of separate short-clip tools; ElevenLabs reports 179 ms average timing errors on full recordings versus 283 ms to more than 6,000 ms for the open-source tools it tested, so the comparison remains vendor-run.
Developers can use Cue for character-level timing in transcripts, subtitles, dubbing, and lyrics across the 29 languages in ElevenLabs' benchmark.
Cue forced alignment model and Forced Alignment API
New from ElevenLabs research: Today we’re sharing more about Cue, our new forced alignment model. It places every character on the audio timeline with more precision than any other aligner we tested. That means the right words show up at the right time. Forced alignment is the step that turns a transcript into timestamps. It’s how subtitles land on time, how dubbed lines match the speaker, and how clicking a word in a transcript jumps to the moment it was said. Most open-source aligners are built for short, clean clips in a limited set of languages. Cue is one model that handles recordings of any length in a single pass. On real-world audio, it was more accurate than every open-source aligner we tested in all 29 languages of our benchmark. It aligns an hour of audio in about half a second on one GPU. Cue already powers every Speech to Text transcription on ElevenLabs, and it’s available on its own through the Forced Alignment API. Read the full post: https://t.co/N1ORSOJVWK
https://x.com/elevenlabs/status/2115527286065914248
Separate local processing for both sides of a call targets speech intelligibility without uploading audio; the site warns that speakerphone use has not been validated.
Also Connect
OpenKrisp provides open-source local noise reduction for microphone and incoming-call audio on Apple Silicon Macs and x86-64 Linux.
OpenKrisp
Fish Audio extends its existing TTS and expressive-model line into the hear side of a live agent stack, with server-side turn semantics rather than transcript-only streaming; the preview has no published price or independent benchmark in its docs.
Also Orchestrate
Fish Audio's public preview streams live PCM audio over WebSocket and returns turn-by-turn drafts and final transcripts for voice agents, captions, commands, and phone calls.
Realtime STT (transcribe-stream-1 preview)
A bot that keeps transcripts and summaries inside Discord reduces meeting-tool switching, while retention and language support remain vendor documentation rather than independent tests.
Also Use
Vyach joins a Discord voice channel, records with participant notice, transcribes the conversation, and posts searchable summaries and answers.
Vyach
Vapi expands its provider catalog with a promptable OpenAI transcription path and a documented AssemblyAI option for HIPAA-enabled assistants, building on its existing version-traffic controls (2026-10-12-vapi-version-traffic-split) without requiring a separate dated platform launch.
Also Orchestrate
Vapi assistants can use OpenAI's gpt-transcribe with keyword and prompt controls, while HIPAA-enabled assistants can select AssemblyAI as their transcriber.
Vapi transcription and assistant controls · vendor OpenAI; AssemblyAI
Arrowhead's in-house TTS turns its earlier BFSI voice-agent deployments into a named regional speech-model layer; the million-call footprint, latency, and pronunciation gain are all company-reported, while no model card or independent evaluation is cited.
Also Use
Arrowhead's Fletch text-to-speech model is deployed for multilingual customer conversations in India and Southeast Asia, with support for 14 languages and code-mixed speech.
Fletch
The 1.8.10 speak cluster connects three current TTS paths—including the newly launched AssemblyAI-TTS-1 and Inworld's updated TTS-2 family—to the same agent runtime; it follows LiveKit's 1.8.9 provider cluster without re-filing that earlier release.
LiveKit Agents 1.8.10 adds Mistral Voxtral 3 TTS, Inworld TTS-2 Flash, and typed AssemblyAI TTS support with 36 Mistral voice presets.
livekit-agents 1.8.10
Markup-aware local read-aloud targets the gap between generic screen readers and AI coding output, while the site gives no independent voice-quality comparison.
Also Use
Hearby reads selected text and AI or coding output aloud with a local natural voice, skipping markup and code symbols.
Hearby
The 28 October 2026 release moves voice output beyond plain text into provider-adapted expression tags and adds a live AssemblyAI TTS path, but expressive conversation is explicitly alpha and each adapter performs only the subset of sounds or deliveries it supports.
Pipecat 1.15.0 adds alpha expressive conversation with delivery, sound, pause, and emphasis tags, plus an AssemblyAI Streaming TTS WebSocket service with word timings.
pipecat-ai 1.15.0
A context field moves expressive control into the TTS request instead of a separate tagging layer; the no-migration path preserves the `inworld-tts-2` integration, while duplex reasoning is still listed as coming soon.
Inworld's Realtime TTS-2 now supports opt-in delivery reasoning and a faster, steadier model behind the same public model ID at the same price.
Realtime TTS-2 reasoning update
Today we’re introducing Reasoning for Realtime TTS-2, making it the first TTS model in the world that can steer itself based on context. In consumer apps, your LLM generates every reply. Directing how each one sounds takes extra prompts, tags, and rules. Reasoning now handles that inside your TTS request. It reads the text and the context you pass, writes a delivery direction, which Realtime TTS-2 performs. For developers this means: - Emotion, tone, and pacing shaped by the context of each line. - No emotion-tag rules to maintain in your chat prompt. - Fewer LLM output tokens spent writing voice directions. The generated direction comes back in the API response, so you can inspect it. Your own instructions and inline tags always take precedence. Available today with Realtime TTS-2. Who are you building for?
https://x.com/inworld/status/2115866740705935621
A callable fal endpoint turns VEED's cloned-voice dubbing into usage-priced developer infrastructure at $0.11 per input-video second, although VEED's first-party API remains presented as a waitlist product and the ship is video-localisation rather than a voice-agent platform.
Also Show
VEED Dubbing Standard is live on fal for translating videos into 86 languages with cloned voices and synchronised lip movements while preserving speaker tone and emotion.
VEED Dubbing Standard · vendor VEED
VEED Dubbing is now live on fal. Translates videos with cloned voices and lip-sync. Pro supports 94 languages and 14 regional accents, preserves background audio, and handles multiple speakers. Provides target accent and voice cloning strength controls.
https://x.com/fal/status/2115448482735522026
A 90-millisecond production target and $15-per-million-character price put a live-call TTS option below the cost AssemblyAI says teams commonly pay; its 22 October waitlist and first listen preceded today's public availability, while the quality figures remain vendor claims.
AssemblyAI-TTS-1 is a generally available text-to-speech model for live voice agents with streaming audio, more than 4,000 voices, instant voice cloning, and no concurrency tier.
AssemblyAI-TTS-1
Introducing AssemblyAI-TTS-1, our first text-to-speech model. Natural speech for voice agents, streaming in 90 ms, at $15 per million characters. On accuracy, naturalness, and cost together, it's the best TTS model you can run at scale.
https://x.com/AssemblyAI/status/2115470615583056042
A public browser surface lets prospective Echo buyers test the same AI before physical delivery; the 100-plus language and vision claims are company claims, and advanced memory, custom voice, and subscription features remain planned for 2027.
Also Think
Ask Aria is available on Realbotix.com as a 3D avatar with speech recognition, text-to-speech across more than 100 languages, and camera-based Vinci visual awareness.
Ask Aria digital AI avatar
The 28 October 2026 release connects the new expressive-speech path to an avatar's face, but only Tavus replicas with expressive conversation enabled receive the mapping and unsupported emotion vocabularies fall back to neutral.
Pipecat 1.15.0 lets Tavus replicas change facial emotion as expressive bot speech is delivered when expressive conversation is enabled.
pipecat-ai 1.15.0
The shared speech, text, image, and tool-use context is a different deployment shape from a cascaded ASR–LLM–TTS agent, although the preview still relies on a separate Sonic-3.6 speech-synthesis layer in the demo.
Also HearSpeakOrchestrate
ServiceNow AI Research's Apriel Voice 27B Preview combines speech, text, image, reasoning, and tool use in one open-weight model for interactive voice agents.
Apriel Voice 27B Preview · vendor Cartesia Sonic-3.6
🚀 Meet Apriel Voice 27B Preview: an open-weight voice AI built to get work done.
https://x.com/ServiceNowRSRCH/status/2115851159282090235
Speech powered by Sonic-3.6.
https://x.com/cartesia/status/2117709023852240974
The financing is attached to a production speech-to-speech model rather than a research-only launch; the announced $31.5 million total and six-continent customer footprint remain Phonic's claims.
Phonic raised a $27.5 million Series A led by First Round, with participation from Lux Capital, Felicis, Definition, and Essence VC.
$27.5M Series A
Phonic-1 from @phonic_co is available in LiveKit Agents on Day 0. It's #1 on the Artificial Analysis Speech to Speech Index and built for long calls with dozens of tool calls.
https://x.com/livekit/status/2115846220610719838
A speech-to-speech model that prepares tool results during an unfinished turn targets the pause and hand-off costs of cascaded phone stacks; Phonic reports 83.2 on Sierra's multilingual full-duplex benchmark, while human listeners still preferred people about 60% of the time.
Phonic-1 is available for agentic calls with 50-plus languages, tool calls while the caller is speaking, and $250 in free credits for sign-ups by 30 November.
Phonic-1
Phonic-1 from @phonic_co is available in LiveKit Agents on Day 0. It's #1 on the Artificial Analysis Speech to Speech Index and built for long calls with dozens of tool calls. Try it out for your agents in one line of code: https://t.co/TxnkAsmVpz
https://x.com/livekit/status/2115846220610719838
Separating turn control from task execution gives voice builders a documented alternative to asking one general model to do both; Think-Live is generally available, while Think-Fast remains early access on 29 October 2026.
Deepgram makes Think-Live generally available in its Voice Agent API and offers Think-Fast in early access on the Deepgram LLM API for tool-driven spoken replies.
Think-Live and Think-Fast LLMs
Meet Think-Live LLM and Think-Fast LLM, two new LLMs built for voice agents. Think-Live runs the conversation: when to listen, wait, or respond. Think-Fast does the work: rules, tools, spoken replies. Our CPO @edanuff breaks down this, and more from #DeepgramSpeak, in his latest blog: https://t.co/dzufs8w2OB
https://x.com/DeepgramAI/status/2115874448066658754
A text checkpoint before speech gives regulated voice teams a place to apply policies and required disclosures, while on-premises and VPC deployment addresses the cloud-only constraint Rime identifies for frontier realtime APIs; the latency and compliance claims are vendor-published.
Also HearSpeak
Prism reasons over callers' audio, emits text for guardrail checks before speech, and runs in the cloud, on-premises, or a virtual private cloud.
Prism
introducing prism, the world’s first speech-to-speech model you can’t tell from a human in conversation prism has: > 300ms response time, 4x faster than OpenAI GPT-Live > outperforms every other s2s model on intelligence (with NO reasoning) > the first s2s model customizable through reinforcement learning we’re a team of 35 with $35M in funding, and we’re unlocking s2s model adoption faster than openai, google, and x. the first 100 people to sign up here get 1 month free: https://t.co/KC3KUq18bl
https://x.com/lilyjclifford/status/2115521147890520503
The 28 October 2026 release separates Bedrock reasoning from spoken output through explicit thought frames and preserves signed context across tools, but the feature depends on provider-specific Claude and Nova reasoning modes rather than making every LLM's chain of thought available.
Pipecat 1.15.0 makes AWS Bedrock reasoning content observable as LLM thought frames and carries signed Claude reasoning across later tool turns.
pipecat-ai 1.15.0
Liquid AI puts an open-weight speech-input and speech-output model with tool calling beside its small edge language models, giving local-agent builders a single audio path rather than a separately assembled ASR, LLM, and TTS cascade; no independent quality or latency result is supplied here.
Also HearSpeakOrchestrate
Liquid AI's open-weight LFM3-Audio-2B takes text or audio and returns text, speech, or both for local voice chat, transcription, translation, audio understanding, and tool calls.
LFM3-Audio-2B
An open Chinese-English path with voice cloning gives developers a local deployment option beside enterprise translation APIs such as Eleven Realtime Translation (2026-10-21-elevenlabs-eleven-realtime-translation); the roughly three-second StreamLAAL and benchmark comparisons come from Tencent's own evaluation.
Also HearSpeak
The open Hy-SimulMT model translates Chinese and English speech in both directions while the speaker is talking, returning translated speech with optional real-time voice cloning.
Hy-SimulMT
A foundation speech-model company is opening an enterprise-built harness to developers, but the 31 October 2026 announcement gives no customer names, pricing, language list, or independent performance figures.
Also HearSpeakConnectUse
Developers and product teams can configure and deploy voice agents through APIs or a voice harness across phone calls, websites, apps, and WhatsApp.
Vaani self-serve developer platform
The 31 October 2026 release turns Pipecat's W44 expressive-conversation APIs into separately downloadable perception and expression components that can be used with any RCS-capable LLM, including smaller models; the public card does not establish end-to-end voice-agent latency or listener-validated quality.
Also HearSpeak
Pipecat releases two small CPU models that read user emotion and sarcasm and add expression tags to an LLM's reply for the Real-time Context Standard.
Pipecat Tone Reader and Tone Tagger
The 30 October 2026 update gives Cognigy builders a documented cross-vendor delegation path alongside new low-latency and expressive TTS choices, but experimental A2A support is not a production guarantee and the release publishes no voice-quality or adoption figures.
Also ConnectSpeak
Cognigy.AI releases 2026.19/20 let agents call external A2A agents and add ElevenLabs v3 and Flash v2/v2.5 options to Voice Gateway.
Cognigy.AI releases 2026.19/20
Reusable connectors and workflows address the repeat integration work that sits after a first agent launch; the October 2026 release gives no pricing or measured deployment-time reduction.
Also UseConnect
Parloa Marketplace gives enterprise agent teams pre-built connectors, vetted partner solutions, and shareable workflows for launching new CX use cases.
Parloa Marketplace
Your first AI agent is live. Your next AI agent shouldn't mean starting from scratch. Too often, launching and scaling CX AI agents means repeating integrations, IT reviews, and conversation design. That's why we launched Parloa Marketplace. Parloa Marketplace offers pre-built integrations for systems like Salesforce, ServiceNow, and Microsoft Dynamics 365. It also includes industry solutions from Parloa and our partners: packaged workflows that combine the conversation logic, system actions, and routing needed to handle specific CX use cases. All within the platform where you build and deploy AI agents. For our customers, that means less repeated implementation work and a shorter path to getting AI agents and new CX use cases live. Learn more about Parloa Marketplace: https://t.co/uqeBc3MWuj
https://x.com/parloa_ai/status/2115850090300805265
A patch release adds two Nemotron 3.5 Bolt sizes across cloud, NIM, server, and single-GPU paths, but it is a hosting and recipe expansion rather than a new voice model; the reasoning-budget rule also changes behaviour below 192 Max Tokens.
Also HearSpeakThink
Nemotron Voice Agent 2.3.1 adds Nemotron 3.5 Bolt E4B across NVIDIA Cloud, server, and single-GPU recipes, plus Bolt 74B A9B on single-GPU hosts.
Nemotron Voice Agent v2.3.1
This turns Pipecat from a Daily-maintained framework into the neutral orchestration layer of an acquirer's full stack, but the foundation's operating model and transaction terms are not yet detailed; the current release explicitly promises continuity for Daily products.
Also Connect
The speech infrastructure company will combine Daily's realtime network and Pipecat's open orchestration framework with its speech models in a new voice-agent platform.
Daily and Pipecat
Today at #DeepgramSpeak, @trydaily, the company behind the open source @pipecat_AI framework, is joining Deepgram. We're thrilled to welcome @kwindla, our new Head of Open Source, and his team to Deepgram.
https://x.com/DeepgramAI/status/2115794199102492889
Cresta's June Conductor launch (2026-06-11-cresta-conductor) moved agent building into a governed harness; the 29 October release changes who can operate that lifecycle by putting the same CLI and voice-testing controls in customer and partner hands.
Cresta gives customer and partner developers self-serve access to Conductor, its CLI, and production workflows for designing, testing, deploying, and optimising voice agents.
Cresta self-serve model
Introducing self-serve for developers on Cresta! Your team can now build customer-facing AI agents and take them to production in as little as a week. With Cresta Conductor and the Cresta CLI, developers build, test, deploy, and optimize agents themselves, using the same trusted tools and infrastructure Cresta’s engineers use on enterprise deployments. Agent code lives in your repo and ships through your existing review and CI/CD pipeline. See how your team can build production-ready AI agents themselves on Cresta: https://t.co/QMIGqZLWKQ
https://x.com/cresta/status/2115865339439513924
This is a materially broader public relaunch than the August docs introduction (2026-08-27-cartesia-managed-agents), adding an end-to-end speech-to-speech framing and an explicit $0.03-per-minute launch price; the earlier row remains the first hosted-agent filing.
Also SpeakHearConnect
Cartesia publicly launches an end-to-end Managed Agents platform for enterprise calls with telephony, knowledge bases, tool integrations, and launch pricing of $0.03 per minute at 50% off.
Cartesia Managed Agents
Introducing Cartesia Managed Agents: our end-to-end voice agent platform for enterprise. #1 on @ArtificialAnlys Speech Agent Arena and @ServiceNow EVA-Bench, built by co-designing our voice models and harness as one system. Check it out → https://t.co/7kJhCYqjQo
https://x.com/cartesia/status/2115848186950791665
We’re launching with 50% off, just $0.03/min. Everything you need to go from prompt to live call in minutes: → Telephony → Tool-calling → Knowledge bases → Outbound campaigns Start building at https://t.co/VXNW5wMEYj.
https://x.com/cartesia/status/2115848190474027265
Configurable disclosures, local-time contact windows, holiday exclusions, frequency limits, and opt-outs turn regulatory settings into product controls for months-long outreach; Sierra describes calls, texts, and email but gives no pricing, named customer, or compliance-outcome figures.
Also Use
Compliance Console gives Sierra's Horizon agents configurable disclosures, contact windows, holiday exclusions, frequency limits, opt-outs, and compliance dashboards across calls, texts, and email.
Compliance Console
This gives Pipecat's classifier-led turn-control path an audio-native component beside the 1.15.0 text/classifier strategies (2026-10-28-pipecat-1-15-0-orchestrate); its published results are model-card measurements, not independent production or phone-codec validation.
Also Hear
Pipecat's open streaming classifier distinguishes user backchannels from floor-taking interruptions so agents can keep speaking or yield.
Pipecat Backchannel Detector
Using purpose-trained classifiers that are small and fast enough to not impact the voice loop can help close the gap in naturalness between a speech-to-speech and cascade pipeline.
https://x.com/JonPTaylor/status/2116554152444907966
The 28 October 2026 release gives voice-agent builders explicit control over turn completion and an early-delegation latency trade-off, but classifier spend, text-only predictive delegation, and changed default stop-strategy semantics make upgrades a configuration change rather than a drop-in speed claim.
Pipecat 1.15.0 adds classifier-based user-turn completion, predictive delegation for LLMWithBackend, richer scripted text expectations, and new turn-strategy combination controls.
pipecat-ai 1.15.0
Putting static skills, account actions, and terminal control together narrows the gap between agent coding and telephony operations; the docs publish no adoption, pricing, or call-quality figures, and the toolkit pages were last modified 23 October even though Vobiz's 27 October post says it is live.
Also Connect
Vobiz's toolkit gives developers Agent Skills, an MCP server, and a CLI for building and operating telephony agents through Vobiz.
Vobiz AI Toolkit
Live today: the Vobiz AI Toolkit. Agent Skills, an MCP server and a CLI. We've taken the guesswork out of building telephony, so you can get it right the first time. https://t.co/L4cg6cNcty
https://x.com/Vobiz_Ai/status/2114959072869106140
Signals gives event-driven systems a mid-conversation path distinct from a webhook that starts a new run, so live voice agents can react to changing operational facts without restarting the workflow.
Also Use
HappyRobot Signals delivers subscribed external events into in-progress workflow sessions so voice agents can change course during a call instead of waiting for a new run.
Signals
ElevenAgents now exposes a direct cost-versus-throughput control for voice-agent LLM work, alongside model selection and clone-strength knobs; the changelog makes the premium configurable but leaves workspace eligibility and measured savings open.
Also Speak
ElevenAgents can run supported LLMs on a paid fast service tier, while the 26 October changelog adds Realtime voice-mode defaults and clone-strength controls for v4 voices.
ElevenAgents fast LLM tier and voice controls
The partnership packages a private deployment route alongside Deepgram's AWS SageMaker guidance (2026-10-01-deepgram-aws-pipecat-sagemaker-voice-agents) and its self-hosted Flux path, giving regulated teams a named Dell/NVIDIA configuration instead of assembling infrastructure from scratch.
Also HearSpeakUse
Deepgram's Voice AI Platform has a validated path to run speech-to-text, text-to-speech, and voice agents on Dell infrastructure inside enterprise data centres.
Deepgram × Dell on-premises Voice AI · vendor Dell Technologies
The marketplace listing creates a dated distribution path for the white-label VoIP connector beyond the generic MCP release (2026-10-14-viirtue-mcp); the new post does not establish new telephony capability or usage volume.
Also Use
Viirtue says its phone-system MCP connector is live in OpenAI's plugin marketplace for ChatGPT and Codex, letting users ask plain-language questions against live phone data.
Viirtue MCP in OpenAI plugin marketplace · vendor OpenAI
Viirtue is now live in OpenAI’s plugin marketplace for ChatGPT and Codex. Partners and their customers can install Viirtue straight from the ChatGPT plugin directory, sign in with their Viirtue credential, and start asking their phone system plain-language questions. No custom connector build required. Things like: → Which calls did we miss yesterday? → Who’s logged into the support queue right now? → Is the front-desk phone online? → What did our AI Voice Agents handle this week? → Did that fax go through? Answers are pulled live from the customer’s account. Access is scoped to what the user’s credential already allows, and admins can revoke it anytime. Partners who want the experience under their own name can still deploy a white label custom connector. Reach out to your channel manager to get started. Install the plugin: https://t.co/18aI2PfD5h Full announcement: https://t.co/auivzZy0Kr #AIVoice #UCaaS #MSP #ChatGPT #WhiteLabel #Viirtue
https://x.com/Viirtue1/status/2115813407534243964
Mid-call changes to ASR/TTS providers, voices, hints, and DTMF settings let builders adapt a call without dropping it; the reconnect path and echo filter target two common failure modes in production voice agents.
Also OrchestrateSpeak
Twilio's Conversation Relay now supports live session changes, application reconnect, echo filtering, and ElevenLabs v4 Turbo voices in General Availability.
Conversation Relay GA updates
A single telnyx-edge ship command puts the phone and agent pipeline on the same edge platform, removing the separate telephony, media-hosting, and model-service assembly; the release does not publish pricing or call-quality measurements.
Also Orchestrate
Telnyx Edge can build, host, and connect Pipecat voice agents to phone numbers with call audio, Flux speech recognition, LLM inference, and text-to-speech behind one API key.
Pipecat on Telnyx Edge · vendor Pipecat
Pipecat agents now run on Telnyx Edge. telnyx-edge ship, connect a number, call it. Phone number, call audio, Flux STT, LLM inference and TTS on one platform, with one API key. Runs anywhere Edge runs.
https://x.com/telnyx/status/2115495097043124402
One quality rubric and staffing forecast now cover AI and human agents together, while mobile and Microsoft Teams access widen the operating surface; the release gives no pricing or general-availability timetable for the new AI capabilities.
Also OrchestrateUse
RingCX adds live supervision, quality scoring, journey analytics, staffing forecasts, and AI onboarding for contact centres managing AI and human agents together.
RingCX AI workforce capabilities
AI agents now handle a bigger share of customer calls. They should meet the same standard as human agents. New AI capabilities for RingCX give contact centers one place to supervise, measure, and plan for AI and human agents as a single workforce. More: https://t.co/uAtU1VbBcB
https://x.com/RingCentral/status/2115416475741622680
Connecting voice agents to live commerce data makes the product more operational than a generic FAQ bot, while its managed-service and workflow claims are vendor-published.
Also ConnectOrchestrate
RuleInside gives WordPress and WooCommerce stores managed AI agents that answer customer questions, recover carts, and handle phone, web chat, and WhatsApp enquiries.
RuleInside
Grounding spoken actions in meetings and documents differentiates Laxis from plain dictation, while its speed and language-count figures are vendor claims.
Also Hear
Laxis lets users speak requests such as replying with meeting facts, then drafts the action in Gmail, Slack, or another app alongside cross-device dictation.
Laxis Agentic Voice Keyboard
The product turns missed calls into a managed front-desk workflow, while the published price and capability claims come from Estaluna.
Also Connect
Estaluna answers business calls around the clock, handles questions, books reservations, takes messages, and transfers callers to named staff.
Estaluna AI Receptionist
A dictation app's built-in iOS keyboard turns Superwhisper's polished voice-to-text surface into a native typing-and-swipe workflow, with iPhone Duo support for Apple's new form factor; the public release gives no quality, latency, adoption, or pricing figures.
Also Hear
Superwhisper 3.0 for iOS adds an all-new keyboard, swipe typing, Liquid Glass support, and iPhone Duo optimisation, with the app available now.
Superwhisper 3.0 for iOS
Superwhisper 3.0 for iOS is available now with an all-new keyboard, iPhone Duo support, liquid glass, swipe typing, and more 📱
https://x.com/superwhisper/status/2116281810669748380
Record moves Sponkly's W40 live chat, call, and meeting translation (2026-10-04-sponkly-live-voice-translation) into an individual recording workflow that preserves a searchable artefact after the conversation; the 30 October feature page and launch post give no model vendor, quality metric, or public launch pricing.
Also HearSpeak
Sponkly users can record a lecture, talk, or meeting, read a live translation, and keep a cleaned transcript, summary, and audio.
Record
New in Sponkly: Record. Sit in a lecture, a talk or a meeting in a language you don't speak. Sponkly writes it down and translates it live, then gives you a clean transcript and a summary in both languages. Share it as PDF, Word or audio. https://sponkly.com/features/record
https://x.com/Sponklycom/status/2116049715380863125
This is an operator deployment of AI call protection rather than a standalone Hiya product launch; free automatic activation puts a vendor's scam-detection layer in front of every O2 customer, while the repeat-attempt and blocked-volume figures are O2 trial or operational claims.
Also Connect
O2 is rolling out free AI-powered blocking that stops suspected scam calls on its network before customers' phones ring.
O2 AI-powered scam call blocking · vendor Hiya
A listed US car and home insurer joins Marriott (2026-09-30-marriott-cresta-ai-agent) as a named Cresta AI Agent customer on voice, though the scope rests on one vendor post with no Mercury statement, metrics, or go-live date.
Cresta said Mercury Insurance (NYSE: MCY) will use its AI agents to automate customer conversations on voice and chat, with agentic assist for the human teams that take the remaining calls.
Cresta AI Agent · vendor Cresta
Together, we're introducing AI agents for voice and chat to automate conversations that drive fast and delightful customer experiences.
https://x.com/cresta/status/2116216564223819823
Answering from the current call moves a notetaker into live assistance, but Cluinn publishes no independent answer accuracy or meeting-containment result.
Also Hear
Cluinn listens to a call on a Mac, answers questions from what was said, surfaces unanswered questions, and puts cleaned dictation into any app.
Cluinn
Dental voice AI moves from making a note to coaching the practice around treatment conversations; availability for all Voice Notes customers is a clearer product step than the May select-customer introduction.
Also HearOrchestrate
Videa makes Ambient Coaching available to all Voice Notes customers, analysing dental patient-provider conversations for treatment, patient-question, and communication insights.
Ambient Coaching
A named eCommerce operator is using a platform agent to connect discovery, purchase, and post-sale support across brands and retailers; the published results are useful operating evidence but do not isolate voice from other channels.
Also OrchestrateConnect
Trademark Global is using Genesys Cloud Agentic Virtual Agent to guide product discovery and handle post-purchase service across its brands and retail channels.
Genesys Cloud Agentic Virtual Agent · vendor Genesys
The dated case study provides evidence about a voice-agent collections workflow but not a new API or general-availability announcement; the unnamed lender and vendor-reported benchmark constrain conclusions.
Also Orchestrate
Synthflow's Vector handled an early-delinquency collections pilot for a US branch-based lender, reaching 4.9% of attempts as the right borrower and securing promises to pay from 49% of those contacts.
Vector
The filing treats in-car voice as a licensable software asset inside Rivian's Volkswagen partnership, while the absence of vehicle, timing, and commercial terms limits what can be inferred about deployment.
Also SpeakConnect
Rivian says Rivian Assistant arrived on R2 in October and that its Volkswagen joint venture now licenses elements of the company's AI Voice Assistant technology.
Rivian Assistant and Volkswagen AI Voice Assistant licensing
A phone-line assistant makes generative voice AI available on basic phones without an app or internet bundle, with a distribution model aimed at rural and low-connectivity users; the launch is a production step beyond the pilot, while accuracy, affordability, and information freshness remain explicit open concerns in the launch coverage.
Also Connect
Miss Baza lets MTN Rwanda customers ask questions and receive spoken answers in four languages by dialing 845 without a smartphone or data bundle.
Miss Baza · vendor Viamo
Today, we officially launched Miss Baza, your AI-powered companion, ready to answer your questions! A special thank you to the Honorable Minister of ICT and Innovation, @MusoniPaula, for her support in advancing digital innovation in Rwanda. We also extend our appreciation to our partners, Viamo, @edencaremedical for being part of this journey. And this is just the beginning! Dial 845 and ask Miss Baza anything. 💛 #NicyoGihe
https://x.com/MTNRwanda/status/2115850477359570952
A meeting-notes app is moving into the cross-app dictation market with a free Mac surface, putting voice input directly into existing writing and AI workflows; the product page and help center do not publish transcription quality, model, or measured latency details.
Also Hear
Granola Voice Typing lets Mac users dictate polished text into the app under the cursor for free.
Granola Voice Typing
Granola now does voice typing. Unlimited words, for free. Granola already writes your meeting notes. Now it can do the typing for you, in any app.
https://x.com/meetgranola/status/2115837013975081208
The feature extends Fireflies' voice capture from its existing mobile dictation and meeting surfaces to X calls, but the announcement gives no independent documentation beyond the dated product post.
Also Hear
Fireflies says users can paste an X call link so Fred joins the call and returns notes, action items, and a recap.
Fireflies on X calls
Fireflies now joins X calls 📞 Paste the call link, Fred hops in, and you hang up with the notes, action items, and recap already done. Your meetings, handled. Even the ones on X. https://t.co/mZsoKvk3Zu
https://x.com/firefliesai/status/2115852279198650388
Approval before sending keeps a voice-first business assistant inside an operator’s control, while the product publishes no independent task-completion result.
Also Orchestrate
Empacer uses a business owner’s setup context to turn spoken requests into drafts and tasks such as customer replies, social posts, and invoices.
Empacer
Centralized routing connects the AI CSR's call and booking events to team workflows without requiring separate alert setup for each recipient; the page gives no adoption or outcome figures.
Also Orchestrate
Avoca's dashboard now centralizes alerts for call summaries, jobs booked by Responder or Simple Scheduler, and new Speed to Lead leads.
Team Notifications
An overlay removes the need to replace the active keyboard before dictating, giving a voice-first writing app a broader cross-app surface on Android; the release supplies no quality or pricing figures.
Also Hear
Typeless 2.10.0 adds an Android bubble that lets people dictate in any app while keeping another keyboard active, then edit the result with that keyboard.
Typeless 2.10.0 for Android
Typeless 2.10.0 is now live on Android! Meet Typeless bubble on Android! Dictate in any app while using another keyboard. Speak naturally and get clear writing. Make quick edits with your usual keyboard. Switch seamlessly between speaking and typing. Update now.
https://x.com/typelessdotcom/status/2115443469053579663
Legal intake routing becomes a configurable taxonomy of caller intent rather than a generic receptionist flow; the free-plan availability lowers the barrier for small firms to test it.
Also OrchestrateConnect
Smith.ai’s AI Receptionist now recognises ten caller types for law firms and lets teams configure routing, questions, and outcomes for each one, including on its free plan.
AI Receptionist Caller Type Identification
A voice channel can reuse Pypestream’s existing integrations, guardrails, escalation, and analytics instead of a separate programme; the release names no QSR brands and its 2.5 billion interactions and roughly nine-in-ten resolution figures are vendor claims.
Also ConnectOrchestrate
Pypestream's Voice AI is generally available for quick-service restaurant brands and carries live customer conversations on the same agent platform as its chat deployments.
Pypestream Voice AI
The changes remove two operational blockers from phone-based benefits and authorisation work—payor identification and bad-input diagnosis—while the 12% CVS Caremark improvement is vendor-reported and no new customer is named.
Also Orchestrate
Infinitus' October product update adds BIN-and-PCN PBM resolution, address-aware authorisation, field-team FieldForce deployment, and error-field reporting for healthcare agents.
Infinitus October 2026 updates
New month, new platform updates! Check out what's new at the blog: https://t.co/gMviYFB8bG
https://x.com/InfinitusAI/status/2115867303418245126
National patient-association distribution makes voice preservation available before speech loss; the partnership cites an estimated 4,000–4,500 people living with ALS in Spain but gives no uptake or clinical outcome.
Also Speak
The ElevenLabs Impact Program partnership with adELA gives eligible people with ALS across Spain free access to voice restoration and preservation technology.
ElevenLabs voice restoration with adELA
Aníbal, an engineer and professor, is losing his voice to ALS. With ElevenLabs, he's recreated it so he can keep teaching. We're partnering with @AsociacionELA, Spain's leading ALS association, to give people with ALS across Spain free access to our voice restoration technology.
https://x.com/ElevenLabs/status/2115374286273950186
The previously unannounced Series A makes the $52 million total a fuller view of Broccoli’s capital behind home-services voice agents; the amounts and customer count are reported by Fortune and Broccoli.
Broccoli AI raised $52 million across its Series A and Series B, led by Khosla Ventures with Y Combinator and Operator Partners participating, to expand home-service AI agents.
Series A and Series B
Broccoli has raised our Series A and Series B, both led by @khoslaventures with @ycombinator
https://x.com/call_broccoli/status/2115521811853312500
A connector gives a general-purpose assistant a phone identity and outbound calling path, but developer-mode access, one concurrent call, and a paid Bland plan keep the feature narrower than a native ChatGPT telephony capability.
Also Connect
Bland's Agent Phone Plan lets ChatGPT make and receive calls and texts through a dedicated US number after a connector sign-in in developer mode.
Bland ChatGPT connector
ChatGPT can make phone calls now. Setup takes 2 minutes. Add Bland as a connector in the ChatGPT plugin marketplace, sign in with Bland, done. Every new Bland account gets free credits, so your first calls from ChatGPT are on us.
https://x.com/usebland/status/2115504837718909316
Native replies extend Avoca's lead-response workflow beyond phone and the existing Yelp and Google Local Services paths; the possible ranking benefit is an Avoca claim, not an independently measured result.
Also Connect
Avoca's Speed to Lead now connects to Thumbtack so it can reply to service-business leads inside Thumbtack, alongside Yelp and Google Local Services.
Native Thumbtack integration
Combining reception, booking, and transcription makes Solwees a concrete small-business phone agent rather than a general chat wrapper; the site publishes no independent containment or quality result.
Also Connect
Solwees answers business phone calls, books and changes appointments, qualifies leads, and records and transcribes each call.
Solwees
Payment-plan setup gives a healthcare voice agent a financial decisioning task beyond balances and payments; RevSpring reports 36% higher payment rates and 27% fewer live-agent transfers, but names no customer or independent evaluation.
Also Connect
Let's Talk lets patients establish healthcare payment plans by phone through the same decisioning engine that powers RevSpring's PersonaPay portal.
Let's Talk payment plans · vendor PersonaPay
Work Mode connects Otter's meeting corpus to before-and-after workflow output, while Live Assist moves beyond the July 2026 enterprise launch towards paid-user availability; Otter gives no pricing or independent productivity result.
Also Hear
Otter's Work Mode generates pre-meeting briefs and post-meeting work from conversation context and connected apps, while Live Assist provides private, in-call coaching.
Otter Work Mode and Live Assist
Introducing the next chapter of Otter. Today, we're launching Work Mode and Live Assist, two new capabilities designed to help you prepare for meetings, get real-time coaching, and turn conversations into action. Reach what's new: https://bit.ly/4zhkybK
https://x.com/otter_ai/status/2115070493502083541
Local-number and dialect support makes the product a regional phone-agent offer rather than a generic chatbot; language coverage and pricing are vendor claims.
Also Connect
Grow50X.ai answers calls on a business number, replies across messaging channels, qualifies leads, books appointments, and writes to a CRM.
Grow50X.ai
Abridge is moving chart review, orders, documentation queries, and diagnosis capture into the same encounter workflow as recorded conversation; the release names evaluation methods and limitations but gives no independent time-saved or clinical-outcome result.
Also HearOrchestrate
Abridge's October platform update adds Visit Topics, staged in-visit orders, chart-informed inpatient notes, outpatient Query Engine coverage, diagnosis write-back, and an evaluation dashboard.
Now in Practice: October 2026
🧵 Now in Practice: October. Nine new capabilities on the Abridge platform, most of them built around one idea: move the work that happens after the encounter back into it, while the context is still fresh. https://www.abridge.com/blog/now-in-practice-october-2026
https://x.com/AbridgeHQ/status/2115170581402292376
A floating dictation surface brings Superwhisper beyond its desktop and iOS meeting workflows; the dated 0.6 availability entry is the product launch, while the later 0.6.1 entry is maintenance.
Also Hear
Superwhisper 0.6 makes its voice-to-text app available on Android, where a floating bubble can dictate into any app in supported languages.
Superwhisper for Android
Dictation on Android just got a whole lot easier... Our latest update introduces the floating pill. Drag it, dock it, and tap to dictate 🗣️
https://x.com/superwhisper/status/2115814312149499906
Google Cloud commitment spend adds a second hyperscaler procurement path after Sierra's OpenAI Marketplace listing (2026-09-29-sierra-openai-marketplace), but the 26 October release describes billing and governance rather than a new voice capability.
Also Orchestrate
Companies with Google Cloud commitments can buy Sierra's AI-agent platform through Marketplace with unified billing and Private Marketplace governance.
Sierra on Google Cloud Marketplace · vendor Google Cloud
Sierra is on Google Cloud Marketplace! Now it's even easier for more businesses to build with Sierra. Read more: https://t.co/TZChtWVCyw
https://x.com/SierraPlatform/status/2114806618219336035
The new SAP Solution Extension sale path is distinct from Parloa's May SAP Service Cloud integration (2026-05-12-parloa-sap-partnership); the release names an SAP portfolio product and support route but gives no pricing or customer-volume figures.
Also Connect
Parloa's Agent Management Platform is available as SAP Conversational Agent Management by Parloa for SAP Customer Experience customers.
SAP Conversational Agent Management by Parloa · vendor SAP
EU-hosted SaaS and on-premise deployment gives regulated customer-care teams one voice stack instead of a separately assembled model layer; the post names more than 50 Yampa clients and 400,000 monthly interactions but gives no independent latency or quality test.
Also HearSpeakOrchestrate
Gradium Text-to-Speech and Speech-to-Text are available inside Yampa's Voice Engine for customer-care agents on EU SaaS infrastructure or on-premise.
Gradium on Yampa · vendor Yampa
Gradium moves Phonon from the W43 open-weight release (2026-10-20-gradium-phonon-open-weights) into a consumer browser surface, while transcription and translation remain API-backed rather than local.
Also SpeakHear
Gradium's free Chrome extension reads selected web text aloud with on-device Phonon and can transcribe or translate tab audio through a Gradium API key.
Gradium Read & Transcribe
We're releasing a set of free tools powered by Phonon. The first is Gradium Read and Transcribe, a Chrome extension that reads any webpage out loud. It runs on Phonon, our 40M-parameter multilingual model, and handles everyday text. For hard cases like acronyms, phone numbers, ...
https://x.com/GradiumAI/status/2114764276627837259
A dental practice-management vendor is turning its existing patient record and AI Receptionist into a single voice workflow rather than leaving phones with a separate provider; the release gives no customer count or independent call-quality result.
Also ConnectHear
DI Voice gives new and existing Dental Intelligence customers an integrated VoIP phone system with call records, transcripts, summaries, sentiment, and AI Receptionist routing.
DI Voice