ByteDance and Tsinghua just introduced DreamID-Omni, an AI model that turns reference portraits, text, and voice clips into talking videos. It eliminates lip-sync confusion in multi-person scenes and supports full video face-and-voice editing, with open-source code expected in March.
Burger King just rolled out an OpenAI-powered voice assistant called Patty directly into workers' headsets. Beyond answering recipe questions and alerting managers to stockouts, Patty actively tracks whether employees say "please" and "thank you" to grade store friendliness. It's currently being piloted across 500 locations.
OpenAI just added gpt-realtime-1.5 to its Realtime API to level up speech-to-speech interactions. The new model brings more reliable instruction following, improved tool calling, and better multilingual accuracy. Why it matters: developers building real-time voice agents can now have AI run tools and fetch data mid-conversation with far fewer glitches.
Wispr Flow just launched its Android app, bringing cross-app AI voice typing to mobile via a floating bubble after earlier debuts on desktop and iOS. The new release cleans up filler words, contextually formats text, translates over 100 languages, and runs 30% faster thanks to an infrastructure rewrite, alongside a new transcription model for Hinglish.
Apple just opened CarPlay to third-party AI chatbots in iOS 26.4, allowing drivers to chat hands-free with ChatGPT, Claude, and Gemini from their car dashboard for the first time. The update introduces a dedicated voice control interface with visual feedback, though bots won't get wake-word support or permission to control vehicle settings. Why it matters: Siri's exclusive grip on the steering wheel is finally loosening.
ElevenLabs just launched ElevenAgents for Support, conversational AI agents that handle customer service across voice, phone, email, and WhatsApp in over 90 languages. The platform boasts sub-second response times, over 10,000 expressive voices, and tone adaptation to de-escalate heated issues without awkward pauses. Teams can deploy them directly from existing SOPs with zero coding and hand off complex tickets to human reps with full context.
The Vatican just teamed up with language provider Translated to bring AI live translation to Holy Mass in 60 languages. Visitors at Saint Peter's Basilica can simply scan a QR code to stream live audio or read text directly in their mobile browser with no app required, powered by Translated's Lara AI.
Moonshine AI just released Moonshine Voice, an open-weight speech-to-text model engineered for real-time on-device use on iOS, Android, and Raspberry Pi. Its 200M parameter streaming model outperforms Whisper Large v3 in word error rate while featuring built-in voice activity detection and intent recognition.
T-Mobile just announced Live Translation, an AI feature that translates phone calls across more than 50 languages in real time. Because the AI runs directly on the carrier's network via VoLTE, you don't need a dedicated app or even a modern smartphone—it even works on older flip phones. Calls aren't recorded or saved, and beta testers can trigger it by dialing 87.
ElevenLabs just rolled out the first insurance policy for AI voice agents, backed by the AIUC-1 security certification. After passing over 5,000 adversarial tests, companies can now insure their bots against hallucinations and prompt injection attacks. Why it matters: over 95% of enterprise AI pilots stall over legal and security fears, and financial underwriting might be the exact safety net businesses need.
ElevenLabs just introduced Audiobooks in ElevenCreative, turning raw manuscripts into fully voiced audiobooks with a single click. Creators can fine-tune emotional delivery scene by scene, clone their own voice, and publish straight to major listening platforms from one unified dashboard. Producing a professional audiobook just went from weeks in a studio to an afternoon project.
OpenMOSS just open-sourced the MOSS-TTS family, a versatile speech suite that clones voices from just a few seconds of reference audio. The lineup ranges from multi-speaker dialogue and voice design to ultra-low-latency streaming, with 8B and 1.7B checkpoints readily deployable on local GPUs.
Soul-AILab just open-sourced SoulX-Singer, a zero-shot singing voice synthesis model. Feed it a few seconds of raw speech and a reference melody, and it makes that voice sing whatever you want without fine-tuning. The best part: at under 3GB, it runs smoothly on basic GPUs and even CPUs.
Open-source tool Just Dub It just released a lightweight LoRA adapter for the LTX-2 foundation model, letting you dub videos into languages like French and German while jointly generating matching lip movements. It preserves the speaker's facial identity and runs on consumer GPUs, with code and weights available on GitHub and Hugging Face.
UniAudio 2.0 just open-sourced a unified audio language model that handles text-to-speech, sound effects, music generation, and vocal style editing in a single architecture. Trained on 160B tokens, it proves multimodal audio no longer needs a stack of separate specialist models.
Mistral AI just dropped Voxtral Transcribe 2, introducing two speech-to-text models for batch and live audio. The standout is Voxtral Realtime, a 4B-parameter model released under Apache 2.0 that runs on-device with latency down to sub-200ms. It brings near-offline accuracy to real-time voice agents across 13 languages without sending private audio to the cloud.
OpenBMB just rolled out MiniCPM-o 4.5, a 9B omnimodal model that sees, listens, and speaks simultaneously in real time. Thanks to an end-to-end full-duplex design, it doesn't just wait its turn—it monitors inputs at 1Hz and can proactively interrupt with reminders. It matches Gemini 2.5 Flash on key multimodal tasks while running locally on a Mac or single GPU.
ElevenLabs just made its Eleven v3 text-to-speech model generally available across all platforms. The update heavily refines tricky contextual notation like phone numbers, currencies, and sports scores, cutting reading error rates from 15.3% to 4.9% with a 72% user preference over the alpha.
MOSS-Audio just open-sourced a unified audio-language model available in 4B and 8B sizes, each offering Instruct and Thinking variants. It packs speech recognition, speaker analysis, emotion detection, environmental sound, and music understanding into one model, significantly outpacing Gemini-3.1-Pro on timestamped ASR accuracy.
Apple just acquired secretive Israeli AI startup Q.ai for nearly $2 billion, marking the tech giant's second-largest acquisition ever behind Beats. The company specializes in machine learning for facial micro-expressions and "silent" voice input. Why it matters: This could fundamentally reshape hands-free interactions for upcoming wearables like AirPods and Vision Pro without requiring users to speak out loud.
