Meeting transcription
The meeting transcription feature records audio conversations and produces timestamped transcripts. Speedwave captures audio directly on the host running the Desktop application and transcribes it locally using whisper.cpp, so recordings never leave your machine. The container running Claude has no access to audio hardware.
Audio capture and processing
Section titled “Audio capture and processing”Each recording supports three audio sources: the full meeting (system audio and microphone in separate channels), system audio only, or microphone only. A Silero VAD gate suppresses silent intervals so Whisper does not invent text during pauses. A level meter shows the input while you record, one bar for the meeting and one for you. Transcripts carry timestamps but no speaker labels; Claude can infer speaker turns in chat.
Enabling and configuring transcription
Section titled “Enabling and configuring transcription”- Open Settings -> Beta features and enable the beta toggle to display the Meeting transcription tab.
- Under Settings -> Meeting transcription, download the two models Speedwave picks for your hardware (table below). Downloads are verified with SHA-256, capped at 12 GiB in total, and follow redirects only to Hugging Face, GitHub and their CDNs. After that, transcription runs offline.
- Open the Meeting transcription tab, select the audio language (Polish or English; no automatic detection), and click Start.
| Hardware | Live model | Final model |
|---|---|---|
| Mac (Metal) or Windows with a dedicated GPU (Vulkan) | large-v3-turbo, about 1.5 GiB | large-v3, about 2.9 GiB |
| Integrated GPU or CPU only | small, about 0.5 GiB | large-v3-turbo, about 1.5 GiB |
The control panel shows the acceleration in use: Metal (GPU), Vulkan (GPU), Vulkan (integrated GPU) or CPU.
Live preview and final pass
Section titled “Live preview and final pass”During recording, the live model writes an auto-scrolling preview. When you stop, the final model re-transcribes the whole recording and replaces the preview; without the final model, the live one does this pass and the panel warns you. You can resume finished recordings. If the final pass fails (for example on empty audio), the session shows a failed: <reason> badge and keeps the live draft.
Clearing the live transcript checkbox makes a recording audio-only until you stop, which keeps the CPU free during the meeting. It is on by default on a Mac or with a dedicated GPU, and Speedwave remembers your choice. While a recording runs, the tray icon and the sidebar tab show a recording marker, and the tab stays reachable with beta features off.
Sending transcripts to Claude
Section titled “Sending transcripts to Claude”Transcripts stay on your machine until you share them. Send to new chat opens a new conversation and Add to current chat uses the open one. The transcript is attached to the composer with a suggested prompt you can edit, and nothing is sent until you press send. Only text is transmitted, never audio.
Local storage of recordings
Section titled “Local storage of recordings”Each recording is saved under ~/.speedwave/transcripts/<id>/ as audio.wav and transcript.json, readable only by your account, unencrypted at rest, and never deleted automatically. Audio goes to disk as it is captured; the live pass holds only a short window in memory.
Meeting transcription requires macOS 14.4 or later. On Windows, system audio is captured via WASAPI. On macOS, the first recording triggers a microphone permission prompt, which can be reset using tccutil reset Microphone pl.speedwave.desktop.audio-capture.
For details on how transcription data is handled, see Data and privacy.