Spaces:
Running
title: Hearlark — Private Audio Transcription
emoji: 🤫
colorFrom: indigo
colorTo: purple
sdk: static
app_file: index.html
pinned: true
license: apache-2.0
short_description: Private speech-to-text in your browser (Whisper)
tags:
- speech-to-text
- whisper
- transcription
- subtitles
- srt
- transformers.js
- webgpu
- privacy
- automatic-speech-recognition
- free-tool
- cybermax
models:
- onnx-community/whisper-tiny
- onnx-community/whisper-base
- onnx-community/whisper-small
thumbnail: >-
https://huggingface.co/spaces/CyberMax-tools/hushscribe/resolve/main/thumbnail.png
Hearlark: private audio & video transcription
Free, no sign-up. How to use:
- Drop an audio or video file, or record from your mic.
- Pick a model size and language (or auto-detect).
- Export TXT, SRT, WebVTT or JSON. The audio never leaves your device.
Need more? Turn the recording into captioned shorts: Clipwhit Pro, $12/month →
Part of CyberMax free tools · all CyberMax tools
Drop an audio or video file (or record from your mic) and get a timestamped transcript, without uploading anything. OpenAI's Whisper runs inside your browser with transformers.js: WebGPU when your browser has it, otherwise CPU (WASM).
- Private by design: the file is decoded and transcribed on your device. No server, no account, no upload.
- Exports: TXT, SRT subtitles, WebVTT and JSON (segments with start/end times).
- 99 languages with auto-detect, plus translate to English.
- Three model sizes: Tiny (fastest), Base (default), Small (most accurate). Downloaded once, then cached.
Good for interviews, lectures, podcasts, voice memos, meeting recordings and video subtitles when the audio shouldn't leave your computer.
FAQ
Why not Otter or TurboScribe? They are good server-based services, but your file is uploaded and free use is capped: Otter's free Basic plan has 300 minutes a month and 3 lifetime file imports (otter.ai/pricing, Sep 2026); TurboScribe's free plan allows 3 files a day of up to 30 minutes. Hearlark never uploads the audio and has no account or quota.
How fast is it? On a plain cloud CPU an 11-second clip took 10.8 s on Tiny and 14.9 s on Base (25 Sep 2026). WebGPU (recent Chrome or Edge) is usually much faster. Long files on CPU take a while, so keep the tab open.
Is it really free? Yes. The page and the model files are served by Hugging Face; nothing runs on our servers, so there is nothing to meter.
More from CyberMax
- All CyberMax free tools
- CyberMax MCP & API kit for AI agents
- Pay-per-result web data tools on the Apify Store
- CyberMax Store: every CyberMax product, API plans compared and bundles, with secure Stripe or Polar checkout.
Hearlark © 2026 CyberMax. Uses Whisper model weights (Apache-2.0) converted to ONNX by onnx-community, and transformers.js (Apache-2.0). Not affiliated with OpenAI or Hugging Face. Accuracy depends on audio quality; check names and numbers before publishing a transcript.