#Browser Compatibility

Mankunku relies on modern Web APIs. This page documents compatibility requirements and known limitations.

#Required APIs

#Minimum Browser Versions

Not supported: Internet Explorer, Opera Mini, older mobile browsers without AudioWorklet.

#Mobile Considerations

#iOS Safari

• AudioContext resume — iOS Safari suspends the AudioContext until a user gesture. Mankunku handles this via Tone.start() called from the first tap.
• getUserMedia — Requires HTTPS. Works on iOS 14.3+ for Safari.
• Screen lock — Audio may stop when the screen locks. Users should keep the screen active during practice.
• Low-latency mode — iOS Safari has higher audio latency than desktop browsers. The latency correction in the scorer absorbs most of this.

#Android Chrome

• AutoPlay policy — Same as desktop Chrome: user gesture required to start AudioContext.
• getUserMedia — Works reliably on Chrome for Android.
• Performance — Lower-end devices may struggle with 60fps pitch detection. The requestAnimationFrame loop naturally adapts to device capability.

#PWA Support

Mankunku is configured as a Progressive Web App via @vite-pwa/sveltekit:

• Installable — Can be added to home screen on mobile and desktop
• Offline-capable — Static assets are cached via service worker
• Icons — App icons in static/icons/
• Manifest — Auto-generated by the PWA plugin

Note: Microphone access requires HTTPS in all browsers. Development via localhost is exempt.

#Known Limitations

#Permissions API

The navigator.permissions.query({ name: 'microphone' }) API:

• Not supported in all browsers (gracefully falls back to 'prompt')
• On macOS, may report 'denied' even when the user hasn't been prompted (browser-level permission not yet granted). Mankunku treats 'denied' as 'prompt' conservatively.

#AudioWorklet Module Loading

The onset detector loads its worklet via:

const workletUrl = new URL('./onset-worklet.ts', import.meta.url);
await context.audioWorklet.addModule(workletUrl);

This relies on Vite transforming the TypeScript worklet file into a loadable module. In production builds, Vite handles this correctly. In some development configurations with HMR, the worklet URL may need special handling.

#SoundFont Loading

First instrument load requires downloading a ~2MB SoundFont file. This is cached via CacheStorage (not localStorage) for subsequent visits. A loading indicator shows during this process.

#Pitch Detection Accuracy

• Polyphonic signals — The McLeod Pitch Method is designed for monophonic instruments. Background noise, multiple instruments, or harmonics from certain embouchures may cause detection errors.
• Low notes — Notes below ~80Hz require longer analysis windows and may have lower clarity scores.
• Very high notes — Above ~1200Hz, harmonics and overtones can cause octave errors.

#Audio Latency

Total system latency (mic → detection → display) is typically 50–150ms, depending on:

• AudioContext buffer size (4096 samples ≈ 85ms at 48kHz)
• requestAnimationFrame interval (~16ms)
• Browser audio pipeline latency

The scoring system's latency correction absorbs constant delays, so latency primarily affects real-time visual feedback rather than scoring accuracy.

#HTTPS Requirement

Microphone access (getUserMedia) requires a secure context:

• https:// in production
• http://localhost in development (exempt)
• http://127.0.0.1 in development (exempt)

Deploy behind HTTPS for production use.