Hearspeech-to-text17
ReplayPlatformlaunch2026-11-14net-new
Replay launches a bot-free, on-device Mac meeting recorder
Local capture removes bot admission and recurring per-seat pricing while open files reduce lock-in, but the product requires Apple silicon and macOS 26 and publishes no independent transcription evaluation.
Also UseOrchestrate
Details2 sources
Replay records calls from any Mac app, transcribes them locally with speaker names, and makes the resulting meetings searchable and available to Claude for grounded questions.
Replay
onoatsPlatformlaunch2026-11-14net-new
onoats launches a local-first meeting recorder that writes plain transcripts
A no-bot recorder that leaves source-separated transcripts as ordinary files gives local-first meeting capture a different path from hosted note-taking apps; local speech-to-text is optional, and the product leaves post-call summarisation to the user's own tools.
Also Use
Details3 sources · 1 X post
onoats records microphone and system audio as separate streams, transcribes them, and writes plain JSONL and Markdown files without an account or database.
onoats
FreeFlow launches on-device Mac dictation with a Claude Code voice hook
FreeFlow combines local speech recognition with a one-time licence and an agent-specific voice control path, unlike subscription cloud dictation; it publishes no independent recognition benchmark.
Also UseOrchestrate
Details3 sources
FreeFlow Dictation turns spoken words into cleaned text at the cursor on Apple Silicon Macs and lets users answer Claude Code prompts by voice without sending audio to a server.
FreeFlow Dictation
SaymilaPlatformlaunch2026-11-13net-new
Saymila launches on-device voice dictation for every app
Saymila joins the crowded any-app dictation field with a single mobile-and-desktop product and on-device recognition, while its cleanup and style transformations go beyond literal transcription; no independent quality results are published.
Also Use
Details2 sources
Saymila provides voice dictation on iPhone, Android, Mac, and Windows, cleaning hesitations, repetitions, punctuation, and style as text lands in any app.
Saymila
LiveKit Agents 1.8.12 adds new realtime STT providers and turn detection
Four new agent-facing transcription paths and a turn-detector input expand LiveKit's speech-input surface after the 1.8.11 provider release, while the 13 November notes leave production limits and comparative performance open.
Details2 sources
LiveKit Agents 1.8.12 adds xAI, AssemblyAI, and Baseten transcription paths, Lokutor Conv 2.0, and a local turn detector with turn-length input.
livekit-agents 1.8.12
iFLYTEK launches SparkAudio 1.0 API for native audio understanding
A single API spanning speech and non-speech audio gives iFLYTEK a broader foundation-model route than the existing Spark ASR 2.0 row (2026-09-23-iflytek-spark-asr-2-0), while the 99-language and 202-dialect coverage is a vendor description and independent quality evidence is absent.
Also Think
Details3 sources
SparkAudio 1.0 is available on iFLYTEK's Open Platform for speech transcription, translation, dialect recognition, environmental-sound understanding, speaker separation, emotion analysis, and audio question answering.
SparkAudio 1.0
FluidVoice releases Fluid Speech Pro with open weights and five runtimes
The open-weight card turns Fluid Speech Pro from FluidVoice's October beta into a portable model for CPU, GPU, Apple Silicon, and Apple Neural Engine deployments; leaderboard, speed, and under-1B claims are vendor-reported.
Details6 sources · 1 X post
FluidVoice published a 0.63-billion-parameter multilingual ASR model under CC BY 4.0 with Transformers, NeMo, ONNX, MLX, GGUF, and Core ML paths.
Fluid Speech Pro open weights and runtimes
Salesforce releases SOV-Transcribe-1, an open-weight ASR model for real-world speech
The release gives enterprises a sizeable open-weight ASR path with natural-language transcript-style controls and entity-focused evaluation, while the non-streaming design leaves realtime deployment to a separate serving layer.
Also Orchestrate
Details5 sources · 3 X posts
SOV-Transcribe-1 is a 4-billion-active-parameter Apache 2.0 speech model for non-streaming ASR with multilingual support, entity accuracy, accent and noise robustness, and instruction-controlled formatting.
SOV-Transcribe-1
MicrosoftModel providermodel release2026-11-12
Microsoft makes MAI-Transcribe-2 generally available in Foundry for production
The GA wording separates the batch model’s production status from the October MAI-Transcribe-2-Streaming public preview, giving Foundry buyers a clearer production path for the lower-cost non-streaming SKU.
Details2 sources · 1 X post
MAI-Transcribe-2 is now generally available in Microsoft Foundry for production transcription at $0.10 per audio hour.
MAI-Transcribe-2 production availability
KoljaBModel providermodel release2026-11-12net-new
KoljaB releases RealtimeVAD 0.0.5 for earlier speech-off signals
A median 100 ms speech-off lead on 3,279 real turns gives voice-agent builders an early, cancellable signal before Silero's turn decision, but the alpha evaluation uses one speaker and capture chain with simulated quiet and phone input.
Also Orchestrate
Details2 sources
RealtimeVAD 0.0.5 is an alpha Python VAD package that adds quiet-input handling, short-backchannel detection, and a LiveKit Agents plugin for earlier end-of-speech signals.
RealtimeVAD 0.0.5
Pipecat 1.16.0 adds diarisation controls across speech-to-text services
The 11 November 2026 release makes speaker identity and endpointing controls part of more Pipecat transcription paths, but provider diarisation support and language coverage still vary by service.
Details2 sources
Pipecat 1.16.0 adds Sarvam HTTP transcription, wider diarisation support, and speaker metadata and controls across its STT services.
pipecat-ai 1.16.0
Reson8Model providermodel release2026-11-10
Reson8 releases Rumble-2 for multilingual speech-to-text
A 54 ms p95 Coval latency claim and coverage of more than 50 languages put Reson8's production model against AssemblyAI, Deepgram, and ElevenLabs on third-party boards, but several cited figures are vendor-copied and the Coval WER is stated two ways in the same post.
Details2 sources · 1 X post
Reson8 launched Rumble-2, a production speech-to-text model for more than 50 languages, now live on its Prerecorded, Realtime, and Turns APIs.
Rumble-2
Pipecat releases Voice Reader rc5 for noisy speech cues
The 10 November rc5 commit makes Pipecat's audio-side perception path less wasteful on short turns and more robust for tiny deployments, but the card remains experimental, English-only, and its evaluations use audio-LLM judges.
Also Orchestrate
Details2 sources
Pipecat Voice Reader rc5 updates the experimental English ONNX model with a shorter adaptive read window and a noise-trained tiny variant for voice-agent perception.
Pipecat Voice Reader rc5
ElevenLabs ships Scribe v3 Live for realtime speech-to-text
Context-aware endpointing and realtime speaker labels move two agent-facing controls into the STT model, while the 43% semantic-WER comparison is measured against ElevenLabs' prior realtime model and not an independent test.
Also Orchestrate
Details3 sources · 2 X posts
Scribe v3 Live brings multilingual realtime transcription, speaker diarisation, and context-aware turn detection to ElevenAgents and ElevenAPI.
Scribe v3 Live
ElevenLabs ships Scribe v3 for multilingual recorded-audio transcription
The batch model extends ElevenLabs' existing Scribe line from general transcription into larger multilingual and multi-speaker workloads, with the 27% WER reduction reported against Scribe v2 rather than an independent benchmark.
Details2 sources · 1 X post
Scribe v3 provides recorded-audio speech-to-text across 102 languages with speaker labelling, keyterm prompting, and formatting controls through ElevenAPI.
Scribe v3
AssemblyAI adds Veterinary Mode and updates Medical Mode for transcription
Domain-specific transcription now covers veterinary conversations as well as the existing clinical path, but AssemblyAI's roadmap labels the veterinary release a preview and gives no independent medical or veterinary accuracy result.
Details2 sources · 1 X post
AssemblyAI adds a Veterinary Mode preview for domain-tuned transcription and updates Medical Mode for pre-recorded and realtime clinical audio.
Medical Mode update and Veterinary Mode
AirCaps AIModel providermodel release2026-11-09net-new
AirCaps AI ships A5S Async v1 for speaker-attributed transcription
Speaker-attributed batch transcription that reports 23.88% cpWER on VoiceArena's 139-conversation board gives AirCaps a differentiated multi-speaker position, while the benchmark and comparison figures are vendor-published and the primary research note is dated 3 November.
Details4 sources · 1 X post
A5S Async v1 provides batch speech recognition with speaker attribution through the AirCaps API for meetings, calls, and other multi-speaker audio.
A5S Async v1
Speaktext-to-speech9
LiveKit Agents 1.8.12 adds Lokutor Versa 2.0 TTS
Nine-language TTS coverage adds a new provider path to the runtime alongside the release's STT integrations, but the changelog supplies no independent comparison or commercial terms.
Details2 sources
LiveKit Agents 1.8.12 adds Lokutor Versa 2.0 text-to-speech across nine languages.
livekit-agents 1.8.12
HeyGenModel providerAPI2026-11-13
HeyGen adds audio tags to its Voice speech API
Inline delivery controls give HeyGen's existing API a text-level route for expressive speech without a separate prosody parameter, while professional-voice support and tag semantics remain constrained by the documentation.
Details3 sources · 1 X post
HeyGen Voice API users can put tags such as [whispers] and [laughs] in text to control delivery on public and instant voices across the three HeyGen Voice models.
HeyGen Voice audio tags
Speechify puts Simba 3.5 on OpenRouter at $15 per million characters
OpenRouter gives Speechify's multilingual streaming TTS a visible public price and an additional developer distribution path, while the underlying model remains the already-filed Simba 3.5 rather than a new version.
Details2 sources · 1 X post
Speechify says Simba 3.5 is live on OpenRouter as a streaming TTS model covering 48 languages, with OpenRouter's model page listing $15 per million characters.
Simba 3.5 on OpenRouter · vendor OpenRouter
ChariotModel providermodel release2026-11-12
Chariot ships Chariot-1-TTS-Multilingual for 11 Indic languages and English
Indic-language code-mixing and context-aware reading target the commerce and conversational cases where generic multilingual TTS often needs application-side normalisation; the comparative scores are Chariot's own evaluation and do not establish independent quality.
Details1 source
Chariot released a TTS model covering 11 Indic languages and English with more than 50 voices, code-mixing, instant voice cloning, and context-aware pronunciation.
Chariot-1-TTS-Multilingual
Pipecat 1.16.0 expands realtime TTS controls and pronunciation settings
The 11 November 2026 release gives builders more direct control over delivery and pronunciation across several TTS integrations, although each setting remains bounded by the provider's own API and model support.
Details2 sources
Pipecat 1.16.0 adds steering, speed, noise reduction, pronunciation, and clone-strength settings across its realtime and provider TTS paths.
pipecat-ai 1.16.0
UnderdogModel providermodel release2026-11-10
Underdog ships Woof JP, a 40.5 MB Japanese browser TTS model
A 40.5 MB Japanese model that runs in-browser lowers the hardware and network barrier for local TTS demos, while the non-commercial licence and vendor-reported 14.5x CPU speed limit its production use.
Details2 sources · 1 X post
Woof JP adds Japanese speech to Underdog's browser TTS line with Misa and Mikasa voices, open weights for non-commercial use, and offline-capable in-browser generation.
Woof JP
ZyphraModel providermodel release2026-11-09
Zyphra releases ZONOS3-Mini, an on-device open-weight TTS model
A 100-million-active-parameter model that Zyphra says starts speaking in under 150 milliseconds on an M3 MacBook Air brings Apache 2.0 voice cloning onto phones and in-browser WebGPU runtimes, though the first-place ZTTS2-Eval ranking and the tenfold word-error-rate claim versus DAC come from Zyphra's own measurements.
Details4 sources · 1 X post
ZONOS3-Mini is a sparse mixture-of-experts text-to-speech model with about 100 million active and 800 million total parameters, Apache 2.0 weights, zero-shot voice cloning, and runtimes for browsers, phones, and servers.
ZONOS3-Mini
OpenAIModel providerlaunch2026-11-09
OpenAI adds Faro and Tonada voices to ChatGPT Voice
Named Argentinian-Spanish voices extend ChatGPT's regional speech coverage after GPT-Live-1.1's 15 October multilingual voice rollout (2026-10-15-openai-gpt-live-1-1), but OpenAI has not published a separate model identifier, pricing, or API path for Faro or Tonada.
Details2 sources · 2 X posts
ChatGPT Voice users can choose Faro and Tonada, two new voices for Argentinian Spanish, with global availability according to an OpenAI Voice team post.
ChatGPT Voice Faro and Tonada
HeyGenPlatformlaunch2026-11-09
HeyGen opens its Platform and Playground for Voice and avatar APIs
HeyGen puts direct model testing and API production paths together on 9 November 2026, making its existing Voice model easier to evaluate and integrate without treating the platform launch as a new speech model.
Also Show
Details3 sources · 1 X post
HeyGen's developer platform exposes its Voice, video, and avatar models in a playground and provides API paths for production use.
HeyGen Platform and Playground
ThinkLLMs, speech-to-speech10
Retell AIModel providermodel release2026-11-13
Retell introduces Tell-1, an interleaved-reasoning LLM for contact-center calls
Interleaving a fast spoken lead with silent reasoning makes contact-center response latency part of the model design rather than a separate orchestration workaround; the score, ranking, latency, and cost figures are Retell-run or company-reported.
Also OrchestrateUse
Details2 sources · 1 X post
Tell-1 interleaves an immediate spoken lead with silent reasoning for Retell voice agents, targeting low latency, tool use, instruction following, naturalness, and contact-center cost.
Retell LLM (Tell-1)
LiveKit Agents 1.8.12 adds Gemini, Perplexity, and Decisions API models
Named reasoning and decision-provider paths broaden the model choices around voice-agent sessions after LiveKit Agents 1.8.11's Responses and decision support, but the release gives no independent model evidence or commercial limits.
Details2 sources
LiveKit Agents 1.8.12 adds Google Gemini 4 models, a Gemini Decisions API model, and Perplexity Decider models to its inference paths.
livekit-agents 1.8.12
Bland AIModel providermodel release2026-11-13
Bland AI previews Tofu-1, a full-duplex model for high-stakes calls
Speaker attribution, interruption handling, and concurrent task work target failure modes in high-stakes phone calls, while the limited enterprise preview publishes no API, price, latency, or independent evaluation.
Also HearSpeak
Details2 sources · 1 X post
Tofu-1 is Bland AI's research-preview speech-to-speech model for calls with background noise, overlapping speakers, interruptions, and live task work.
Tofu-1
KugelAudio releases Fluid 1, a speech-to-speech model for phone conversations
Fluid 1 puts turn-taking and speech-to-speech handling into KugelAudio’s European-hosted voice stack, complementing its earlier Kugel TTS and Clarity rows rather than replacing them.
Also HearSpeakConnect
Details2 sources · 1 X post
Fluid 1 handles speech-to-speech phone conversations with interruption-aware turn taking, OpenAI Live compatibility, and coverage of 24 languages and 52 dialects.
Fluid 1
Pipecat 1.16.0 adds Zoom AI Voice Agent and Gemini Interactions services
The 11 November 2026 release adds both a native speech-to-speech provider and a stateless multimodal Gemini route, but availability, billing, and behaviour remain tied to Zoom and Google's separate APIs.
Details2 sources
Pipecat 1.16.0 adds a speech-to-speech Zoom AI Voice Agent service and a stateless Gemini Interactions API service with streaming, tools, and multimodal context.
pipecat-ai 1.16.0
DeepLModel providermodel release2026-11-11
DeepL launches Aure, a real-time voice translation model that preserves expression
Aure follows DeepL's September voice-preservation release (2026-09-15-deepl-voice-preserve) by making expressive, identity-preserving translation the named model surface across meetings, in-person conversations, and the Voice API; the page publishes no independent quality evaluation or pricing.
Also HearSpeak
Details2 sources · 1 X post
DeepL made Aure available across DeepL Voice's meetings, in-person conversations, and API paths, translating a speaker's rhythm, intonation, emphasis, and voice identity in real time.
Aure
DecagonModel providermodel release2026-11-11
Decagon introduces Alto, a conversational model for duplex voice agents
Alto makes turn judgment—whether to answer, wait for confirmation, or keep the caller engaged while a tool runs—a named model surface in Decagon's Voice 3 duplex stack; the latency and resolution gains are Decagon's production claims, and the post does not document a public model API or weights.
Also HearSpeak
Details2 sources · 1 X post
Alto is Decagon Labs' conversational model for duplex customer-service agents, deciding when to speak, wait, or keep a call moving.
Alto
Modulate adds Twilio Ventures and Paul English investment alongside Velma updates
A 1.6-second median detection time and greater-than-99.1% reported accuracy give Modulate a faster caller-screening pitch, while the undisclosed strategic investment and vendor-run evaluation leave the commercial and independent evidence open.
Also HearOrchestrate
Details2 sources · 1 X post
Modulate announced new investments from Twilio Ventures and Paul English alongside Velma updates for synthetic-caller detection and audio-native voice-agent monitoring.
Velma audio intelligence updates and strategic investment
Fish Audio launches Live 3 full-duplex model for voice agents
Putting backchannels, pauses, and interruption handling inside a full-duplex model gives voice-agent builders an alternative to separately tuned turn-taking thresholds, although the 186 ms response figure is Fish Audio's claim and English is the only announced language.
Also HearSpeak
Details2 sources · 1 X post
Fish Audio Live 3 is a full-duplex English voice model that listens while speaking, handles interruptions and backchannels, and is available through the Fish Agent Platform and API.
Live 3
Nuance Labs introduces Emma, a full-duplex persona built on Nuplex
Emma's simultaneous-response behaviour and audiovisual model claim put Nuance Labs on the full-duplex persona track after its 14 September 2026 Series A, while access remains a research preview and the 18-month research history is company-reported.
Also HearSpeakShow
Details3 sources · 2 X posts
Nuance Labs presents Emma as a research-preview AI persona built on Nuplex, its full-duplex audiovisual foundation model.
Emma and Nuplex
Orchestrateframeworks, evals25
LiveKit Agents 1.8.12 adds GPT-Live delegation recovery
Recovery for missed GPT-Live delegations targets a concrete failure between spoken caller answers and backend actions, extending the task-routing path from LiveKit Agents 1.8.11 without publishing recovery rates or timing guarantees.
Details2 sources
LiveKit Agents 1.8.12 can recover GPT-Live caller answers that were transcribed but never delegated after a quiet grace period.
livekit-agents 1.8.12
DeepSeekModel providerAPI2026-11-13net-new
DeepSeek makes its full App capabilities available through the API
This is a broad model-provider API surface rather than a standalone speech-model launch: ASR and TTS are part of a larger application-tool suite, while the official page does not identify separate speech model names or independent voice-quality measurements.
Also HearSpeakThink
Details4 sources
DeepSeek exposes web search, file input, URL reading, safety checks, OCR, ASR, and TTS through its API for building applications with the capabilities of the DeepSeek App.
DeepSeek Full-Suite App API
Artificial Analysis launches a Diarization Index for speech-to-text APIs
Speaker attribution is a separate production failure mode from word transcription accuracy, so a WDER board gives voice-agent and meeting builders a more targeted comparison than aggregate WER alone; the benchmark remains a vendor-run evaluation whose dataset and weighting should be read with its methodology.
Also Hear
Details2 sources · 1 X post
The Artificial Analysis Diarization Index compares speech-to-text APIs by Word Diarization Error Rate, measuring whether words are assigned to the correct speaker.
Artificial Analysis Diarization Index
Talkdesk launches Voyager for long-horizon customer journeys
Voyager shifts Talkdesk’s automation framing from resolving an individual interaction to coordinating a governed journey across departments and systems, making persistent context a core product surface.
Also UseConnectThink
Details2 sources · 1 X post
Talkdesk Voyager coordinates AI agents, employees, workflows, and enterprise systems around customer goals that can run over days, months, or years.
Talkdesk Voyager
Retell AI reports $100M annualized run rate after raising $4.6M
The small disclosed raise alongside a large run-rate claim is a capital-efficiency and distribution signal for the voice-agent platform segment, while the figures are not independently audited in the announcement.
Also UseConnect
Details1 source
Retell AI says it reached a $100 million annualized run rate with $4.6 million raised and more than 4,000 customers running over 100 million calls per month.
$100M annualized run rate and $4.6M raised
LiveKit opens its Cloud API beta for coding-agent voice workflows
LiveKit makes project administration and call-level debugging available to coding agents under the user’s own permissions, extending the existing agent-building path into operations and evaluation.
Also ConnectUse
Details2 sources · 1 X post
The LiveKit Cloud API lets coding agents create and manage projects, members, webhooks, and voice-agent call data through user-authorised REST and CLI access.
LiveKit Cloud API
Fish Audio launches Fish Agent, a no-code platform for custom voice agents
Fish Audio moves from speech models and APIs into a self-serve agent-building surface, lowering the setup barrier while leaving production controls and commercial limits open.
Also ConnectSpeakUse
Details1 source · 1 X post
Fish Agent lets users build custom AI voice agents that make and receive phone calls without writing code.
Fish Agent
CrestaPlatformlaunch2026-11-12
Cresta launches Journey Intelligence across customer touchpoints
The product extends Cresta’s conversation-intelligence and agent-operations line from individual interactions toward cross-touchpoint journey analysis, while the public page leaves the data connectors and evaluation methodology open.
Also UseConnect
Details2 sources · 1 X post
Cresta Journey Intelligence connects calls, chats, and other customer touchpoints into a longitudinal view for diagnosing repeat contacts and supporting agents.
Journey Intelligence
CekuraPlatformlaunch2026-11-12
Cekura expands its Voice Quality benchmark to 16 platforms and 1,395 live calls
The expansion moves Cekura’s public evaluation from separate model boards toward a production-platform quality view, while the new weighting and changed test set mean buyers should not read the November leaderboard as a time-series improvement over September.
Also HearSpeakThink
Details2 sources · 1 X post
Cekura’s November Voice Quality run compares 16 production voice platforms across 31 phone lines, 15 scenarios, and 1,395 live calls using 12 quality metrics.
Voice Quality benchmark
VapiPlatformlaunch2026-11-11
Vapi opens Agent Loop early access for self-improving voice agents
Agent Loop packages the monitoring, simulation, experiment, and traffic-routing work that Vapi says customers were building by hand, but access begins with a waitlist and the launch's customer results are not independently measured.
Also ThinkConnect
Details2 sources · 1 X post
Vapi's Agent Loop builds, evaluates, tests, and incrementally deploys voice agents against a business goal, with early access available by waitlist.
Agent Loop
Pipecat 1.16.0 adds rubric judging and expressive perception controls
The 11 November 2026 release moves Pipecat evaluation beyond simulated calls by adding a saved-conversation rubric, while the classifier and perception paths remain provider- and model-dependent.
Details2 sources
Pipecat 1.16.0 adds EvalRubric for judging recorded conversations and Google Decisions plus tone and perception changes for agent evaluation.
pipecat-ai 1.16.0
GoogleModel providerAPI2026-11-11
Google launches Gemini Decisions API for fast multimodal choices, including audio
An audio-aware decision endpoint gives voice-agent builders a classifier and routing surface that can inspect a recording without asking a generative model to transcribe or produce a long answer, though the voice relevance is an audio-classification path rather than spoken interaction.
Also HearThink
Details2 sources
Gemini Decision 1 and the Decisions API score up to 1,000 bounded questions over shared text, image, audio, video, or PDF input through Google AI Studio.
Gemini Decision 1 and Decisions API
ElevenLabs launches a partner program and marketplace for enterprise agents
The two launches broaden ElevenLabs' enterprise distribution beyond direct agent configuration, but Marketplace purchasing is not available until next year and the release gives no partner-level pricing or deployment outcomes.
Also Use
Details3 sources · 1 X post
ElevenLabs launched a partner program for enterprise ElevenAgents deployments and a curated marketplace of industry-specific AI and voice applications.
ElevenLabs Partner Program and Marketplace
ElevenLabs and NVIDIA expand their partnership for enterprise voice agents
The expansion links a named NVIDIA multimodal model and compute capacity to ElevenLabs' agent platform while pointing to a government/healthcare deployment path; the project details remain an announced collaboration rather than a documented production rollout.
Also ConnectUse
Details1 source
ElevenLabs and NVIDIA expanded their partnership with Nemotron 3.5 Super VL in ElevenAgents, additional Vera Rubin compute, and an on-premises Polish e-health voice-agent project.
ElevenLabs and NVIDIA enterprise voice-agent expansion · vendor NVIDIA
ElevenLabs previews Interaction Engine for model-aware voice-agent conversations
Model routing and conversational timing are presented as one control surface rather than separate prompt and endpointing settings, but the summit post describes this as a preview rather than a generally available product.
Also Think
Details2 sources · 1 X post
ElevenLabs previewed Interaction Engine, a system that manages turn-taking and interruptions while routing simple and complex parts of a conversation to different models.
Interaction Engine
ElevenLabs adds Memory and Routines to its ElevenAgents roadmap
Together the features extend ElevenAgents from a conversation surface into a longer-lived customer context and workflow layer; the public announcement describes planned behavior rather than a documented API or production rollout.
Also UseConnect
Details3 sources · 2 X posts
ElevenAgents Memory retains useful details across returning customer conversations, while Routines automates multi-step work before, during, and after those conversations.
ElevenAgents Memory and Routines
Aristotle publishes Pivotal Turns evaluation for real-time AI tutoring
Pivotal Turns makes teaching decisions—when a voice tutor should hold back, hint, or intervene—a first-class evaluation target beside call-quality and turn-taking benchmarks, but the labels, judge calibration, and results are Aristotle's own and the public post does not release the full dataset.
Also ThinkSpeak
Details3 sources · 1 X post
Aristotle published Pivotal Turns, an evaluation of whether voice tutors provide the right amount of support at pivotal moments in live tutoring sessions.
Pivotal Turns
GooglePlatformlaunch2026-11-10
Google adds offline ASR and TTS pipelines to AI Edge Gallery
The release gives developers a modular offline voice path beside Google's earlier local Foresight meeting companion (2026-10-06-google-ai-edge-foresight), but it combines an app showcase, system APIs, and open-weight bundles rather than introducing one separately named speech model.
Also HearSpeakUse
Details1 source
Google AI Edge Gallery now offers on-device speech recognition and speech synthesis for voice-enabled AI Chat, XR language tutoring, and developer integrations across Android and other platforms.
Google AI Edge Gallery on-device ASR and TTS
NiCE Cognigy launches its next-generation agent platform with Agentic Experience Studio
NiCE Cognigy is moving from an agent-builder and voice-gateway line toward a governed workspace for designing, evaluating, and operating a broader AI workforce; its voice claims remain first-party product positioning rather than an independent latency or quality result.
Also SpeakConnectUse
Details2 sources · 1 X post
NiCE Cognigy launched a new agent platform with Agentic Experience Studio, AI-assisted agent building, tracing and evaluation, governance, and real-time voice-agent capabilities for enterprise customer experience teams.
Next Generation of NiCE Cognigy
CeipalPlatformlaunch2026-11-10net-new
Ceipal launches Ora with a phone and video Voice Agent for recruiting
Ora puts a phone-screening agent inside a staffing system's ATS, VMS, and workforce workflows, so voice is one governed step in a recruiting operating system rather than a separate call tool; the adoption and time-saving figures are Ceipal claims.
Also Use
Details2 sources
Ceipal Ora launches an agentic recruiting platform whose Voice Agent calls candidates by phone or video, conducts prescreening and technical screening, and returns a structured summary and score for recruiter review.
Ceipal Ora Voice Agent
Bandwidth unveils Call Intelligence preview and network-level SI innovations
Call Intelligence puts conversation observability and compliance inside a CPaaS orchestration layer rather than requiring separate point solutions, but the preview status leaves its rollout and production limits open.
Also HearConnect
Details1 source
Bandwidth previewed Call Intelligence for its Maestro platform, combining call recording, transcription, sensitive-data redaction, and compliance monitoring across AI and human call flows.
Bandwidth Call Intelligence
Trillet ships a redesigned workflow builder with testing and call controls
The release puts regression testing, call evidence, and transfer handling into the same builder used to ship agents; its broad feature bundle publishes no pricing, adoption, or independent accuracy evidence.
Also HearSpeakConnectUse
Details1 source
Trillet’s redesigned builder adds versioned workflows, pre-publish evaluation, call review, emotion detection, smarter warm transfers, and new voice and language options.
Redesigned workflow builder
ThunderPhone launches v3 architecture and Zeus for voice-agent tuning
ThunderPhone reports a 99.4% Big Bench Audio score and cost-plus call pricing from about 2¢ per minute, but both the benchmark and the savings case are company claims and Zeus remains an early product.
Also ThinkConnectUse
Details1 source
ThunderPhone v3 gives phone and web voice-agent builders a configurable multi-model stack, while Zeus builds, tests, and tunes each agent against simulated and real calls.
ThunderPhone v3 and Zeus
SoundHound adds multimodal simulation to OASYS agent testing
Testing bad phone lines, background noise, and frustrated callers gives OASYS a pre-production voice-evaluation path beyond clean chat prompts; the launch's telemetry and scenario claims are from SoundHound, with no independent evaluation supplied.
Also HearConnect
Details3 sources · 2 X posts
SoundHound's OASYS adds voice simulation, realistic customer personas, and cross-channel A/B testing for agents before they reach customers.
OASYS multimodal simulation
ElevenLabs launches Campaign Manager for outcome-based outbound calls
Phone calls and WhatsApp are live on 9 November 2026, while email and SMS are still coming soon, so ElevenAgents gains a managed outbound-campaign surface without requiring teams to rebuild their existing agents.
Also ConnectUse
Details2 sources · 1 X post
ElevenAgents Campaign Manager lets outbound teams enrol contacts against an outcome, then run scheduled follow-ups across phone and WhatsApp until the outcome is reached or the contact opts out.
ElevenAgents Campaign Manager
Connecttelephony, channels6
Twilio makes Smart Turn Detection GA for Conversation Relay
Semantics-aware end-of-turn metadata gives phone-agent builders a provider-neutral control for barge-in and orchestration, while the Deepgram Flux exception and Twilio-native scope define the current boundary.
Also OrchestrateHear
Details2 sources
Twilio Conversation Relay now uses an in-house semantics-aware model to detect caller end-of-turn across supported STT providers.
Smart Turn Detection for Conversation Relay
Telnyx adds pre-storage redaction to Voice API and TeXML recordings
Redacting upstream before storage removes the interval in which sensitive audio would otherwise sit on disk, but English-only coverage and Deepgram-rate billing limit the current documented scope.
Also Orchestrate
Details2 sources · 1 X post
Telnyx call recordings can now silence card, personal, and health details before storage across Voice API and TeXML.
Call Recording Redaction for Voice API and TeXML
QuiqPlatformlaunch2026-11-12
Quiq launches Voice AI integration for Zoom Contact Center
The App Marketplace route ties an enterprise voice-agent product to a named contact-centre distribution and handoff surface, reducing a deployment barrier without introducing a new speech model.
Also OrchestrateUse
Details1 source
Zoom Contact Center customers can add Quiq’s Voice AI agent from the Zoom App Marketplace, with human escalation carrying call details and context into the Zoom desktop app.
Quiq Voice AI integration with Zoom Contact Center · vendor Zoom Contact Center
Pipecat 1.16.0 adds Genesys AudioHook pause and resume controls
The 11 November 2026 release gives Genesys AudioHook pipelines an explicit hold-control path, while the surrounding telephony changes still require callers to align provider and pipeline sample-rate settings.
Details2 sources
Pipecat 1.16.0 adds explicit pause and resume requests to Genesys AudioHook and updates telephony serializers for current sample-rate behaviour.
pipecat-ai 1.16.0
3CXPlatformlaunch2026-11-11
3CX ships V20 Update 10 with MCP, stronger AI agents, and local Mac transcription
The final Update 10 release turns the earlier AI Server and alpha work into a broader phone-system update, pairing local speech processing with MCP and agent capabilities for self-hosted deployments.
Also OrchestrateThinkHear
Details2 sources · 1 X post
3CX V20 Update 10 adds MCP support, improved AI agents, local transcription and call summaries on Mac mini, and broader phone-system controls.
V20 Update 10
Upfirst adds automatic replies and human handoff for incoming texts
The update turns Upfirst’s SMS handoff from a call follow-up surface into a two-way agent channel for small-business reception; the changelog does not publish message limits, pricing, or rollout geography.
Also Use
Details1 source
Upfirst’s agents can now reply to incoming text messages, hand conversations to a team member, and resume the exchange when the operator returns control.
Automatic text replies
Useagents in market22
TaciePlatformlaunch2026-11-14net-new
Tacie launches a bot-free Mac meeting assistant that answers questions in the room
Retrieval during the conversation is a different job from post-call summarisation, and local capture avoids bot admission, but Tacie publishes no speech-engine, accuracy, or retention benchmark.
Also HearOrchestrate
Details2 sources
Tacie records meetings on a Mac without joining as a bot, retrieves answers from past meetings and connected tools during the call, and writes decisions and tasks afterwards.
Tacie
WilloPlatformlaunch2026-11-13net-new
Willo launches a voice companion for World of Warcraft players
A game-specific assistant turns voice interaction and live game context into an in-play companion rather than a general chat surface, but Willo publishes no speech-model, latency, or quality details.
Also HearSpeak
Details2 sources
Willo is a free-to-start Windows, Mac, and iPhone companion that listens to spoken questions about World of Warcraft and answers aloud, with optional game context from the Lore Forever add-on.
Willo
Survive Dinner launches a two-minute AI voice game for nosy family questions
The launch is a compact example of conversational speech interaction used for entertainment: the player's spoken answers drive an AI-relative scenario instead of a conventional text-only game loop.
Also SpeakThink
Details2 sources
Survive Dinner is a browser voice game in which a player chooses a difficult family situation and three AI relatives, then answers questions aloud until dessert.
Survive Dinner
SpeakEQPlatformlaunch2026-11-13net-new
SpeakEQ launches an AI communication coach for real-time voice feedback
SpeakEQ adds a consumer-facing speech-analysis layer that turns vocal delivery and emotional signals into immediate coaching rather than only post-call transcription or meeting summaries.
Also HearThink
Details2 sources
SpeakEQ is an AI communication coach that analyzes tone, pace, energy, and vocal patterns and gives real-time feedback on delivery and expression in its early-access mobile app.
SpeakEQ AI communication coach
OntadaEnterpriselaunch2026-11-13net-new
Ontada makes Commure-powered AI capabilities available in iKnowMed
Embedding ambient documentation and cursor dictation in an oncology-specific EHR gives more than 3,300 iKnowMed providers a documented voice-and-workflow path, while the release publishes no adoption, accuracy, or rollout-timing metrics beyond availability.
Also HearOrchestrate
Details3 sources · 1 X post
Ontada made ambient documentation, chart and visit summaries, smart coding, a Practice Assistant, and cursor-based dictation available to iKnowMed customers.
iKnowMed integrated AI capabilities · vendor Commure
NACHC makes Abridge Assistant free to 108,000 community health center clinicians
The arrangement extends Abridge's clinical decision-support access beyond its existing health-system customer base into community health centers at national-member scale, but the release does not publish adoption, implementation, or clinical-outcome data.
Also Hear
Details2 sources · 1 X post
Abridge Assistant will be available at no cost to 108,000 clinicians at all NACHC member community health centers, whether or not they use Abridge's clinical documentation product.
Abridge Assistant · vendor Abridge
ServiceTitan rebuilds its trades voice agent on Vapi in two days
A named trades-software operator gives Vapi a concrete customer-deployment signal for configuration-driven voice-agent rebuilding, while the story publishes no call volume, containment rate, or independent quality result.
Also OrchestrateConnect
Details2 sources · 1 X post
ServiceTitan rebuilt a voice agent for trades-business callers on Vapi in two days after a six-month initial build, and says its customers can run customised agents for calls, lead capture, and bookings.
Trades voice agent on Vapi · vendor Vapi
HoyPlatformlaunch2026-11-12net-new
Hoy launches a private Mac channel app with voice notes and read-along transcription
A Mac communications app folding read-along transcription into no-server iCloud channels gives speech a first-class place beside files and screen recordings; the November 2026 changelog does not identify the speech engine or publish transcription quality.
Also Hear
Details3 sources
Hoy 1.0 puts iCloud-backed people and channels in the Mac menu bar and turns voice notes, video notes, and screen recordings into read-along text.
Hoy 1.0 for Mac
DNBEnterpriselaunch2026-11-12net-new
DNB deploys Genesys Cloud AI in its contact centre and plans Agentic Virtual Agent
The announcement separates a measurable contact-centre AI deployment from a future autonomous-agent phase: DNB has current advisor-assist and summarisation use, while its planned Agentic Virtual Agent rollout gives Genesys a path toward more self-service in a regulated financial-services environment.
Also OrchestrateConnect
Details3 sources · 1 X post
DNB is using Genesys Cloud AI for advisor assistance and interaction summaries in its contact centre, while planning a later deployment of Genesys Cloud Agentic Virtual Agent for more customer self-service.
Genesys Cloud AI adoption · vendor Genesys
Anthropic launches Claude Meetings beta for live video calls
This gives Anthropic a live enterprise meeting surface built around speech capture and shared work rather than a spoken-response assistant, with transcription consent and organisation-level access controls as part of the rollout.
Also HearOrchestrateShow
Details1 source · 1 X post
Claude Meetings joins Google Meet calls on Team and Enterprise plans to take notes, answer questions, and build shared documents and whiteboards while people talk.
Claude Meetings
American Heritage Credit Union’s SoundHound AI agent surpasses half a million member conversations
A named credit-union deployment crossing 500,000 conversations gives SoundHound a concrete production scale signal in financial services, while the release does not establish resolution quality or the division between voice and other channels.
Also ConnectOrchestrate
Details2 sources · 1 X post
American Heritage Credit Union reported that its SoundHound AI agent has handled more than 500,000 member conversations.
SoundHound AI Agent · vendor SoundHound AI
Typeless partners with Anker soundcore on the Work voice-dictation wearable
The partnership connects a consumer wearable's capture and meeting-note workflow to Typeless' cross-app dictation surface; public materials do not disclose the underlying speech model, accuracy method, or commercial terms between the partners.
Also Hear
Details3 sources · 2 X posts
Typeless and Anker soundcore announced the soundcore Work clip-on wearable, which captures speech and turns it into polished writing through Typeless.
Typeless × Anker soundcore Work · vendor Anker soundcore
InDebted launches Voice for outbound collections calls and payment-plan follow-up
The launch adds a collections-specific outbound voice product with payment-plan and hardship guardrails to InDebted's collections infrastructure platform, while the twice-as-likely retention result comes from the company's own accounts and no independent production evaluation is supplied.
Also OrchestrateConnect
Details2 sources
InDebted launched Voice, an AI agent that makes outbound collections calls, helps customers set payment plans, and hands sensitive conversations to people on the collections team.
Voice
HelixPlatformfunding2026-11-11net-new
Helix raises $6.2M and launches on-device voice human verification
Voice biometrics becomes a human-verification and age-assurance product rather than only an authentication feature, adding a privacy-first, on-device trust layer to the emerging voice-agent ecosystem.
Also HearThink
Details2 sources
Helix raised $6.2 million in pre-seed funding and launched voice-based biometric identity and age verification with its first partners.
Voice-based human verification
GooglePlatformlaunch2026-11-11
Google brings Gemini Live to Apple CarPlay
CarPlay adds a mainstream in-car voice-assistant surface for Gemini Live across supported languages, but it is a consumer app rollout rather than a new model or developer API.
Also ThinkSpeak
Details2 sources · 1 X post
Gemini users on iOS can now start hands-free Gemini Live conversations from an Apple CarPlay display, with calendar, inbox, planning, and brainstorming tasks available while driving.
Gemini on Apple CarPlay
Fireflies.ai launches Fred as an AI teammate with voice agents
Fireflies is combining voice-agent calls with meeting and connected-app context in one subscription, while its 140% usage increase and nearly 100 million task figures are company claims and Fly remains prospective.
Also HearOrchestrateConnect
Details1 source
Fred combines meeting notes with voice typing, command mode, connected-app actions, and voice agents for candidate screenings, sales discovery, customer support calls, and user interviews.
Fred
ElevenLabs opens free early access to ElevenAI for workplace conversations
ElevenAI gives ElevenLabs a workplace-facing surface alongside customer-facing ElevenAgents, with consent and context handoff as part of the product design; access remains early and the post does not establish production performance.
Also HearOrchestrate
Details2 sources · 1 X post
ElevenAI is an enterprise work assistant that turns consented meetings and spoken dictation into notes, next steps, and live context shared between employees and agents.
ElevenAI
SangomaPlatformlaunch2026-11-10net-new
Sangoma launches Sonella as a production agentic voice assistant for business calls
Sonella gives an established UC and communications vendor a first-party business voice-agent path that sits inside its existing phone system, with vertical guardrails and human escalation rather than a standalone chatbot add-on.
Also OrchestrateConnect
Details2 sources
Sangoma Sonella is production-available for businesses that want natural-language voice agents to answer calls, schedule appointments, route requests, and create support tickets.
Sonella
PromptPath raises $12M Series A for automotive dealership Voice AI
PromptPath adds a dealership-specific conversation-assist platform to the automotive Voice AI field, with the funding aimed at expanding beyond phones into F&I and service workflows rather than introducing a separately named speech model.
Also Orchestrate
Details1 source
PromptPath raised a $12 million Series A to expand its end-to-end Voice AI conversation-assist platform for automotive dealerships, from phones through F&I and service.
PromptPath Voice AI conversation-assist platform
JampaiEnterprisepartnership2026-11-10net-new
Jampai uses Gradium's Jette voice across advertising and customer phone calls
The named customer is using a consistent synthetic voice across brand marketing and a live phone touchpoint, a deployment signal for voice identity rather than a new Gradium model release; performance and commercial details are unpublished.
Also Speak
Details3 sources · 1 X post
German insurance broker Jampai uses Gradium's Jette voice in its advertising and as the voice customers hear when they call the company.
Gradium voice Jette · vendor Gradium
MitraPlatformlaunch2026-11-09net-new
Mitra adds an MCP connector so agents can call from a user's number
Personal-number calling gives MCP clients a telephony path that avoids unfamiliar agent numbers; the 9 November product page leaves call pricing, supported countries, and model details unstated.
Also Connect
Details3 sources
Mitra's MCP server lets ChatGPT, Claude, and Grok ask its phone assistant to make and answer calls from the user's number, returning results to the AI app.
Mitra MCP
Granola brings free cross-app Voice Typing to iPhone
iOS availability extends Granola's cross-app dictation from its Mac and Windows beta into a mobile typing surface on 9 November 2026, while the public materials provide no independent accuracy or latency result.
Also Hear
Details2 sources · 1 X post
Granola Voice Typing is available on iPhone for free users, turning speech into formatted text in the app where they are writing.
Granola Voice Typing for iOS