Privacy‑first audio‑to‑text demo for students: LoreDrop, Local Whisper, and practical steps

Question: Can students try an in‑browser audio‑to‑text demo that keeps raw audio on their device, and how does LoreDrop actually handle speech capture?

Short answer: Yes — LoreDrop supports browser speech capture and an in‑browser “Local Whisper” option that keeps audio processing on the device, but note generation and account synchronization still go through LoreDrop’s web service; LoreDrop does not accept arbitrary uploaded audio files.

Intent, evidence and product relevance

This note explains a privacy‑minded demo workflow for students who want device‑side transcription before using LoreDrop’s study tools. Evidence is limited to LoreDrop’s verified feature list (local Whisper for browser-captured audio; note generation and sync require the web service) and public technical references on browser speech and local Whisper builds (MDN Web Speech API and whisper.cpp/whisper.wasm). Relevant links: MDN Web Speech API documentation (https://developer.mozilla.org/en-US/docs/Web/API/Web_Speech_API?utm_source=openai) and whisper.wasm README (https://github.com/ggml-org/whisper.cpp/blob/master/examples/whisper.wasm/README.md?plain=1&utm_source=openai).

What LoreDrop supports (one-line answer up front)

• LoreDrop supports browser-based speech capture: users can record speech in the browser and choose local Whisper transcription (audio processed locally on the device) or the browser’s dictation; generated notes and account sync still require LoreDrop’s web service.

How in‑browser transcription commonly works (brief)

• Web Speech API (browser/platform): Many web apps use the browser’s SpeechRecognition interface for live dictation. Behavior and privacy depend on browser implementation; some vendors route audio through their speech proxy. (See MDN: https://developer.mozilla.org/en-US/docs/Web/API/Web_Speech_API?utm_source=openai.)
• Local Whisper in the browser (WASM/WebGPU): Projects compile Whisper to WebAssembly or run lightweight ggml models in the browser so transcription happens on device. These keep raw audio local but require model downloads and device CPU/GPU resources. Example implementation details: https://github.com/ggml-org/whisper.cpp/blob/master/examples/whisper.wasm/README.md?plain=1&utm_source=openai.

Hypothetical worked example (privacy-preserving student demo)

Scenario: Jamie, a biology student, wants to record a 20‑minute study read‑aloud and keep the raw audio private.

1) Jamie opens LoreDrop in Chrome and selects the speech capture tool. The browser asks for microphone permission.

2) In LoreDrop settings she picks “Local Whisper (in‑browser).” The browser downloads or loads a cached WASM model bundle (first run may take longer). Local Whisper transcribes on her laptop so raw audio never leaves her device; LoreDrop receives the transcript text for note generation and syncing.

3) Jamie reviews the transcript inside LoreDrop, edits mis-transcribed phrases, and then generates structured study notes with her preferred narrator tone and example categories.

4) Her transcript text and generated notes are saved to her account and synchronized across her devices using LoreDrop’s web service.

Actionable step‑by‑step checklist for students who want a demo

• Use a modern desktop browser (Chrome/Edge/Firefox variant) and an up‑to‑date OS. WebAssembly and WebGPU support help for larger models.
• In LoreDrop: open the speech capture tool, grant mic permission, and select “Local Whisper” (or use browser dictation as fallback).
• Expect a model download or cache load on first use; allow extra time and disk space. Choose a lower‑capacity model if offered to reduce resource use and latency.
• After transcription: edit verbatim transcript excerpts (LoreDrop exposes matching source excerpts from saved transcripts), then run the structured note generator.
• If you need to transcribe prerecorded files (MP3/Zoom/lecture recordings): either play the file into your microphone while recording in LoreDrop, or transcribe locally with a separate whisper.wasm client and paste the text into LoreDrop (LoreDrop does not accept arbitrary audio file uploads).

Limitations and decision criteria you must know

• No arbitrary file import: LoreDrop cannot ingest uploaded audio files, PDFs, slides, or Zoom recordings directly — you must capture audio in the browser or paste transcript text. This is a product limitation and determines when external tooling is required.
• Resource cost: Local Whisper in the browser can require sizeable downloads and CPU/GPU work; expect slower first runs and higher load on low‑end devices. Using smaller models reduces these costs (see whisper.wasm notes: https://github.com/ggml-org/whisper.cpp/blob/master/examples/whisper.wasm/README.md?plain=1&utm_source=openai).
• Browser privacy tradeoffs: The Web Speech API’s privacy and routing differ by vendor — check the browser’s policy if you need a guarantee that audio never leaves the device (https://developer.mozilla.org/en-US/docs/Web/API/Web_Speech_API?utm_source=openai).
• Not fully offline: Even with local transcription, generated notes and synchronization require LoreDrop’s web service — LoreDrop is not a fully offline app.

Call to action

Try a short demo in a quiet room: open LoreDrop, pick Local Whisper for device-side transcription, record a 5–10 minute clip, edit the transcript, and generate structured study notes. If you need to handle prerecorded audio, transcribe it locally with a whisper.wasm client and paste the transcript into LoreDrop.

Summary (one sentence)

LoreDrop lets students capture speech in the browser and choose local Whisper for on‑device transcription, but it does not accept arbitrary audio file uploads and note generation/sync still use LoreDrop’s web service — for fully local workflows transcribe externally with a browser Whisper client and paste the text into LoreDrop.