Archive, page 2
Drama 3 is a 35B-parameter mixture-of-experts text-to-speech model with about 3B active parameters. Fish Audio released the weights, and users on Reddit and MarkTechPost call it a new leading open TTS option.
A British lawmaker became the first person to use an AI-generated voice to deliver a speech in the UK parliament, with software recreating her former voice during a health debate to read a speech she wrote herself.
xAI updates its Grok iOS app with improvements to Chat, Voice and Imagine.
Baroness Campbell addressed the House of Lords using her own AI-recreated voice built with ElevenLabs, the first AI-voiced speech in UK Parliament history.
A project repurposes Google's EmbeddingGemma 2 embedding model into a speech-to-text system, with the write-up and demo shared on Hacker News.
Voice AI company Gladia is set to join OVH Groupe, aiming to accelerate Europe's sovereign voice AI capabilities with European-hosted speech technology.
Deepgram is expanding its Flux voice model to 10 languages and targeting air-gapped laptops, letting voice AI run locally without cloud connectivity for sensitive environments.
Google is testing a redesigned Gemini Live mobile interface with a new waveform animation and a dedicated voice output control.
Tesla is bringing the Grok Bot to Cybercab and robotaxi fleet vehicles, enabling voice-controlled tasks during rides, and expanding Grok Bot voice tasks to Model Y.
TII releases Falcon-ASR, a 1.6B-parameter Arabic speech recognition model with an Emirati focus, reporting 20.92% average WER on six test sets. Weights and API are not yet available.
Rivian is rolling out a new AI voice assistant in its vehicles, with InsideEVs detailing what it can do.
Atticus is a Mac voice assistant: hold Control-Option-Space, speak or type a task, and it runs it, sending simple tasks to cheaper handling. Built by a 9th-grade developer and shown as Show HN.
Open WebUI is a self-hosted, user-friendly chat interface for Ollama and OpenAI-compatible APIs. v0.12.0 adds realtime voice calls using an OpenAI Realtime voice model, switchable by administrators in audio settings.
EMA Lightning is a tiny Turkish TTS system: a 5.6M-parameter DiT acoustic model plus a 3M-parameter vocoder, about 34 MB total, Apache 2.0. It claims 0.92% WER on Freya-TR-Eval and 3.86 ms first audio.
Benzinga reports SoundHound AI shares are in freefall, examining why the Nvidia-backed voice AI company's stock has dropped sharply.
TestingCatalog reports that Anthropic might be developing its own voice models. No launch or official confirmation has been reported.
Google's Gmail Live feature may soon add seven new voices powered by Gemini, according to Android Authority. The change is not yet rolled out.
PocketWebTools for Mac bundles local AI chat, transcription and video tools in one app that runs on your Mac. Launched on Product Hunt.
Automation Anywhere agreed to acquire conversational voice AI company Boost.ai from Nordic Capital to expand customer-facing voice agents and its 'autonomous enterprise' push.
Willow Care, maker of AI dictation app Willow Voice, chose CoreWeave for model training, citing forward-deployed support rather than raw compute. The app combines speech recognition with dictation features.
OpenAI's Dots persistent agent is now fully rolled out on mobile, letting users issue voice commands to Codex for task handling. The Register reports open-source imitators have followed.