{"slug": "claude-meetings-joins-the-call-your-overlay-still-lives-on-dual-track-wasapi", "title": "Claude Meetings Joins the Call — Your Overlay Still Lives on Dual-Track WASAPI Capture", "summary": "Anthropic launched Claude Meetings on 10 November 2026, adding Claude as a visible participant on Google Meet (beta for Team and Enterprise) that can draft plans, sketch timelines, answer chat questions, and post notes and action items to claude.ai after the call. A developer building the Windows app Talkaside argues the feature is orthogonal to live translation needs because it consumes the platform's mixed audio stream rather than capturing separate local mic and loopback tracks, and cannot render an overlay invisible to screen share. The developer recommends dual-track WASAPI capture with a lightweight audio callback, track alignment, and a local or BYOK STT/translation path, noting end-to-end partials typically land in the 80–150 ms range.", "body_md": "Anthropic launched Claude Meetings on 10 November 2026. The feature adds Claude as a visible participant on Google Meet (beta for Team and Enterprise). While people are still talking it can draft the plan under discussion, sketch a timeline on a whiteboard, answer questions in the meeting chat, and pull data from connected apps. Notes and action items land in claude.ai after the call. It posts a notice when it joins and refuses external guests unless an admin has authorized them.\n\nThat is a clean product move for knowledge work. It is also orthogonal to the problem that actually bites when you need live language help on a Windows call or interview.\n\nClaude Meetings joins the call. It receives the mixed audio stream the platform already has, plus any shared screen or chat. It does not sit on your local audio endpoints. It does not give you a second, private track of your own microphone that never enters the call. It does not let you paint a translation overlay that is invisible to screen share. And it does not run when you are on a platform it has not integrated with yet, or when you prefer not to invite a third-party participant at all.\n\nFor a developer who needs real-time speech-to-text and translation during a technical interview, a sales call, or a support session in another language, those constraints matter. The useful path is still local capture of two tracks, local or BYOK processing, and an overlay that never appears in the shared desktop.\n\n*Classic amplitude/frequency sketch. Real Windows capture graphs also include shared-mode mixing, communications enhancements, and device-specific jitter that no remote meeting participant can remove.*\n\nA single mixed stream loses speaker identity. For live translation you need:\n\nThat separation lets the STT engine produce clean transcripts for each side and the UI show them side-by-side without injecting a bot into the meeting.\n\nOn modern Windows the practical stack is:\n\nA “sub-50 ms time-to-final” cloud model is real once it has clean PCM. It is almost irrelevant if the samples have already been delayed or dropped inside the audio engine. End-to-end partials on a good day land in the 80–150 ms range; Bluetooth or a busy shared-mode session pushes that higher.\n\n*Multiple waveforms. Dual-track capture keeps the mic and the remote side separate so each can be biased, timed, and translated independently.*\n\n```\n// Pseudocode shape used by many Windows audio tools\n// cpal or windows-rs WASAPI exclusive/shared, loopback on render device\nlet mic_stream = device.build_input_stream(&config, |data| {\n    // copy to ring buffer, never do heavy work here\n    ring_mic.push(data);\n}, err_fn)?;\n\nlet loopback_stream = render_device.build_input_stream(&loopback_config, |data| {\n    ring_loopback.push(data);\n}, err_fn)?;\n\n// separate thread\nloop {\n    let (mic_chunk, loop_chunk) = align_and_read(&ring_mic, &ring_loopback);\n    if vad_active(&mic_chunk) || vad_active(&loop_chunk) {\n        stt_send(mic_chunk, loop_chunk); // local model or BYOK\n    }\n}\n```\n\nKeep the audio callback under a few hundred microseconds. Do the format conversion, VAD, and network/model work on another thread. Pin the capture thread and the audio interrupt affinity if DPC spikes are eating your buffers (Process Lasso’s LatencyTest or raw ETW still surface those).\n\nClaude Meetings is a useful new participant for teams that want the model inside the call. For the developer who needs the language barrier gone on their own machine, without joining a bot and without leaking the overlay, the work is still the same: open two WASAPI clients, keep the callback empty, align the tracks, and feed a tight STT/translation path. The model improvements keep arriving. The capture budget does not move until you measure and tighten it yourself.\n\nI'm building [Talkaside](https://talkaside.com/), a Windows app for job interviews and calls in another language.\n\n*Drafted with AI assistance from this project's code and logs.*", "url": "https://wpnews.pro/news/claude-meetings-joins-the-call-your-overlay-still-lives-on-dual-track-wasapi", "canonical_source": "https://dev.to/madnessbrains_i/claude-meetings-joins-the-call-your-overlay-still-lives-on-dual-track-wasapi-capture-ifa", "published_at": "2026-11-11 07:19:35+00:00", "updated_at": "2026-11-11 07:21:58.974123+00:00", "lang": "en", "topics": ["ai-products", "ai-tools", "natural-language-processing", "developer-tools"], "entities": ["Anthropic", "Claude Meetings", "Google Meet", "Talkaside", "WASAPI", "claude.ai"], "also_reported_by": [], "alternates": {"html": "https://wpnews.pro/news/claude-meetings-joins-the-call-your-overlay-still-lives-on-dual-track-wasapi", "markdown": "https://wpnews.pro/news/claude-meetings-joins-the-call-your-overlay-still-lives-on-dual-track-wasapi.md", "text": "https://wpnews.pro/news/claude-meetings-joins-the-call-your-overlay-still-lives-on-dual-track-wasapi.txt", "jsonld": "https://wpnews.pro/news/claude-meetings-joins-the-call-your-overlay-still-lives-on-dual-track-wasapi.jsonld"}}