Speakr is a personal, self-hosted web application designed for transcribing audio recordings
Python
Add a 'Privacy notes' collapsible section to the system-audio help modal — covers the trade-offs of each capture mode in plain terms so users encountering the modal also see the security implications of the choices it documents; collapsed by default (uses a <details> element so it doesn't dominate the modal's primary setup focus); content sections: (1) Microphone / System Audio / Mic+System — standard browser permission model, no new privacy surface; (2) Disable echo cancellation / noise suppression / auto-gain — not a privacy setting at all, just captures more of the ambient room; (3) Linux virtual source (speakr_loopback) flagged with a warning-coloured header because the pactl module-virtual-source wrap explicitly undoes Chrome's filtering of PulseAudio monitor sources which is a deliberate privacy protection (without it any site with mic permission could enumerate speakr_loopback and capture system audio silently), with three concrete mitigations: load-on-demand instead of persistent default.pa, audit mic permissions in chrome://settings/content/microphone, use a less obvious source name; (4) where audio ends up — self-hosted Speakr → configured ASR endpoint, so local Whisper keeps everything local and cloud ASR sends audio to that provider, a deployment-config concern not a capture-mode one
Murtaza Nasir1998f74e7bd6
diff --git a/templates/modals/system-audio-help-modal.html b/templates/modals/system-audio-help-modal.htmlindex 7f3cb8e3..fc89b643 100644--- a/templates/modals/system-audio-help-modal.html+++ b/templates/modals/system-audio-help-modal.html@@ -188,6 +188,64 @@</div></div>+ <!-- Privacy notes — collapsed by default so the modal+ stays focused on setup; expandable for users who+ care about the trade-offs each capture mode makes,+ especially the Linux virtual-source workaround+ which removes a Chrome-level privacy protection+ on the user's machine. -->+ <details style="margin-top: var(--sp-4);">+ <summary class="cursor-pointer t-body-strong" style="color: var(--text-secondary);">+ <i class="fas fa-shield-halved"></i>+ Privacy notes+ </summary>+ <div class="space-y-3" style="margin-top: var(--sp-3);">+ <p class="t-body" style="color: var(--text-secondary);">+ Most of these capture modes use standard browser APIs and have the same privacy properties as any other web app that records audio. A couple of them have specific trade-offs worth knowing about.+ </p>++ <div>+ <div class="t-body-strong" style="margin-bottom: var(--sp-1);">Microphone, System Audio, Mic + System</div>+ <p class="t-body" style="color: var(--text-secondary);">+ All gated by the browser's permission system. The browser shows an explicit prompt before capture starts, displays a persistent recording indicator while it's running, and lets you revoke access at any time from the site settings. No new privacy surface beyond what's already in any voice-recording web app.+ </p>+ </div>++ <div>+ <div class="t-body-strong" style="margin-bottom: var(--sp-1);">Disable echo cancellation, noise suppression & auto-gain</div>+ <p class="t-body" style="color: var(--text-secondary);">+ Not a privacy setting — these are audio-quality processors. With them off the recording captures more of the ambient room (background voices, typing, notifications). The data path doesn't change; everything still goes to the same place.+ </p>+ </div>++ <div>+ <div class="t-body-strong" style="margin-bottom: var(--sp-1); color: var(--text-warn, var(--text-accent));">+ <i class="fas fa-triangle-exclamation"></i>+ Linux virtual source (<code style="background: var(--bg-tertiary); padding: 0 4px; border-radius: 3px;">speakr_loopback</code>)+ </div>+ <p class="t-body" style="color: var(--text-secondary);">+ This is the one to think about. Chrome's filtering of PulseAudio monitor sources is a deliberate privacy protection — without it, a malicious site with mic permission could silently record everything you're listening to (calls, sensitive content, private music) with no "screen sharing" indicator. The <code style="background: var(--bg-tertiary); padding: 0 4px; border-radius: 3px;">module-virtual-source</code> wrap explicitly undoes that protection on your machine.+ </p>+ <p class="t-body" style="color: var(--text-secondary); margin-top: var(--sp-2);">+ After the module is loaded, any site you've granted mic access to can enumerate <code style="background: var(--bg-tertiary); padding: 0 4px; border-radius: 3px;">speakr_loopback</code> and capture system audio through it.+ </p>+ <div class="t-body-strong" style="margin-top: var(--sp-2); margin-bottom: var(--sp-1);">Mitigations:</div>+ <ul class="list-disc list-inside space-y-1 t-body" style="color: var(--text-secondary);">+ <li><strong>Load on-demand</strong>: don't add the <code style="background: var(--bg-tertiary); padding: 0 4px; border-radius: 3px;">load-module</code> line to <code style="background: var(--bg-tertiary); padding: 0 4px; border-radius: 3px;">~/.config/pulse/default.pa</code>. Run <code style="background: var(--bg-tertiary); padding: 0 4px; border-radius: 3px;">pactl load-module …</code> before recording and <code style="background: var(--bg-tertiary); padding: 0 4px; border-radius: 3px;">pactl unload-module module-virtual-source</code> after.</li>+ <li><strong>Audit your mic permissions</strong> at <code style="background: var(--bg-tertiary); padding: 0 4px; border-radius: 3px;">chrome://settings/content/microphone</code> and remove any site that doesn't strictly need it.</li>+ <li><strong>Use a less obvious source name</strong> (e.g. <code style="background: var(--bg-tertiary); padding: 0 4px; border-radius: 3px;">aux_in</code> instead of <code style="background: var(--bg-tertiary); padding: 0 4px; border-radius: 3px;">speakr_loopback</code>). Doesn't stop a determined enumerator but raises the bar.</li>+ </ul>+ </div>++ <div>+ <div class="t-body-strong" style="margin-bottom: var(--sp-1);">Where the audio ends up</div>+ <p class="t-body" style="color: var(--text-secondary);">+ Captured audio goes to your Speakr server (self-hosted by you) and then to whichever ASR endpoint you've configured. If you're running a local Whisper / self-hosted ASR, audio never leaves your network. If your ASR is OpenAI or another cloud provider, audio gets sent there — same as any other recording you upload to Speakr. That's a deployment-config decision, not specific to these capture modes.+ </p>+ </div>+ </div>+ </details>+</div><div class="modal-footer"><button @click="showSystemAudioHelp = false" class="btn btn--primary">Got it</button>