Phonon TTS puts studio-grade voice fully on-device, free for English

A 65MB Apache-2.0 model runs offline in a browser and tops the TTS leaderboard — plus Copilot in Xcode, a new Claude Code build, and open world models.

Nowline Nov 7 3:00 AM banner

Top AI stories from the last hour

Top AI stories from the last hour

Copy markdown

  • Offline voice, in a browser tab

    Phonon's English model (phonon-31m, Apache-2.0) is about 65MB and runs on phones, laptops, or a browser tab with no network — the text never leaves the device. It hits 28ms time-to-first-audio on CPU and ranks #1 for German and Spanish intelligibility on Hugging Face's Open TTS Leaderboard, so you can ship real-time voice with zero API cost or latency.

  • Copilot lands inside Xcode

    GitHub Copilot now plugs into Xcode's built-in coding assistant, and reached JetBrains Air the same week. iOS and macOS devs get Copilot completions and chat without switching to a third-party editor — it's a Nov 4 release on the Copilot changelog.

  • Claude Code 2.1.318, and a deprecation clock

    The current build fixes a `--continue` regression that could silently drop earlier conversation after a failed tool call, and stops macOS sessions stalling as memory grew. Bigger heads-up: Opus 4.8 'fast mode' is deprecated and removed Nov 30 — migrate those workflows before then.

  • NVIDIA opens four Cosmos 3 world models

    NVIDIA released four open-weight world-simulation models (16B params each) for interactive and driving scenes. Open weights mean you can fine-tune and self-host them for robotics policies, simulation, and synthetic training data.