ClearMic

Studio-grade voice cleanup, in your browser.

Try it →

---

Noise out. Voice forward.

Background hum, room reverb, street bleed, fan whine — they ruin recordings. ClearMic can remove noise, enhance speech, transcribe, or chain those steps together. No desktop apps, no plugin chains, no learning curve.

Drop a file, choose at least one processing step, get clean audio and/or a transcript back in under a minute.

Built for

• Podcasters rescuing takes from bad room acoustics
• Journalists salvaging interviews recorded in noisy locations
• Voice-over artists working without a treated booth
• Field recordists fighting wind, traffic, and HVAC
• Anyone who's ever recorded a voice memo and wished it sounded cleaner

Why ClearMic

	ClearMic	Desktop pro tools	Most online tools
Cost	Free, 10/day	Hundreds of dollars	Often paywalled
Setup	None	Hours of plugins	Sign-up funnels
Speed	Under a minute	Manual editing	Long queues
Privacy	Files purge in 24h	Local-only	Stored indefinitely
Access	Any browser	macOS / Windows only	Varies

How it works

1. Drop in audio — WAV, MP3, M4A, FLAC, OGG. Up to 200 MB.
2. Clean it up — Pick noise removal, enhancement (Low, Medium, or High strength), transcription, or any combination. Steps apply in order, top to bottom.
3. Verify your email — a 6-digit code. No passwords.
4. Wait under a minute — close the tab if you want, your spot is saved.
5. Download, or come back later — Grab the cleaned audio and transcript right from the page. Your last 24 hours of jobs stay in the Recent library for one-click re-download. Opt in to email and we'll deliver 24-hour download links to your inbox the moment it's ready.

Privacy by default

• All audio is automatically purged 24 hours after upload
• No tracking, no analytics, no third-party ads
• Sessions expire after 30 days
• 10 processing jobs per day per email — keeps costs sustainable and abuse contained

The engine

ClearMic runs a selectable Replicate pipeline:

• Noise removal — playmore/speech-enhancer
• Enhancement — resemble-ai/resemble-enhance, with Low, Medium, and High presets
• Transcription — vaibhavs10/incredibly-fast-whisper

When multiple steps are selected, each step receives the output of the prior step. Transcription always uses the latest audio in the chain and does not change the downloadable audio.

---

Technical overview

Built on the Cloudflare edge — single Worker, no backend servers.

Layer	What
Frontend	Static assets served from public/
Runtime	Cloudflare Workers
Storage	R2 (audio), D1 (sessions + job metadata)
Queue	Cloudflare Queues for async job kickoff
Email	Cloudflare Email Service
Bot check	Cloudflare Turnstile
AI	Replicate (playmore/speech-enhancer, resemble-ai/resemble-enhance, vaibhavs10/incredibly-fast-whisper)

Browser uploads go directly to R2 via 15-minute presigned PUT URLs, so the Worker never buffers large upload bodies. The presigned URLs sign exact Content-Type, Content-Length, and custom metadata; the completion endpoint verifies the stored object before queueing. Sessions live in D1 (not KV) for stronger consistency and revocability. Replicate output URLs are validated against an allowlist of replicate.delivery hosts before fetching.

Project layout

Path	What
src/index.ts	API router + Queue handler + cron scheduler
src/auth.ts	Email OTP, D1 sessions
src/uploads.ts	Direct-to-R2 upload intents, quota
src/jobs.ts	Job status, input, webhook, download
src/replicate.ts	Prediction lifecycle, output persistence
src/pipeline.ts	Multi-step processing selection + sequencing
src/notifications.ts	Completion-email rendering and dispatch
src/email-template.ts	Shared email shell, brand mark, button helpers
src/cleanup.ts	Scheduled deletion of expired R2 objects + D1 rows
src/r2.ts	Presigned R2 PUT URL signing (aws4fetch)
src/turnstile.ts	Turnstile siteverify
src/db.ts	D1 helpers, public job shape
src/audio.ts, src/model.ts, src/http.ts, src/types.ts	Shared helpers + types
public/index.html, styles.css, app.js	UI
public/icon.svg	App icon (favicon + apple-touch-icon + email brand mark)
migrations/	D1 schema

API

Method	Path	What
POST	/api/auth/request-otp	Send email login code
POST	/api/auth/verify	Verify code, set session cookie
GET	/api/me	Current user + remaining quota
POST	/api/logout	Clear session
GET	/api/config	Public Turnstile site key
POST	/api/uploads	Create presigned R2 upload
POST	/api/uploads/:id/complete	Verify upload, queue job
GET	/api/jobs	List active jobs + quota
GET	/api/jobs/:id	Job status
GET	/api/jobs/:id/download?token=…	Download clean output with job token
GET	/api/jobs/:id/transcript?token=…	Download transcript text with job token

Upload body: { fileName, fileType, fileSize, noise_removal, enhance, transcribe, enhancement_preset, email_on_completion }

• Select at least one of noise_removal, enhance, or transcribe
• enhancement_preset — low (32 evaluations) · medium (64) · high (128), used when enhance is selected
• Enhancement always uses solver: "Midpoint" and prior_temperature: 0.5; only number_function_evaluations changes by preset
• email_on_completion — optional boolean; sends 24-hour download links after the final selected step completes

Local development

npm install
cp .env.example .env   # fill in secrets
npm run db:migrate:local
npm run dev:tunnel     # public URL via Cloudflare Tunnel

Replicate fetches uploaded audio from a public HTTPS URL, so end-to-end processing against localhost won't work. Use the dev:tunnel URL in your browser, not localhost.

During local tunnel development, browser uploads go through the Worker into Wrangler's local R2 simulation instead of using presigned R2 URLs. This avoids the local-R2 versus remote-R2 mismatch while still giving Replicate a public URL to fetch /api/jobs/:id/input. Deployed production traffic still uses direct-to-R2 uploads.

Turnstile automatically uses Cloudflare's official always-pass test keys on localhost, 127.0.0.1, and *.trycloudflare.com. Production hosts still use TURNSTILE_SITE_KEY and TURNSTILE_SECRET_KEY from the environment.

Don't create .dev.vars — it overrides .env in Wrangler and we expect .env.

Deploy from scratch

Provision Cloudflare resources:

npx wrangler r2 bucket create clearmic-audio
npx wrangler queues create clearmic-audio-jobs
npx wrangler d1 create clearmic-db
npx wrangler r2 bucket lifecycle add clearmic-audio delete-after-1-day --expire-days 1
npm run r2:cors

Paste the database_id from wrangler d1 create into wrangler.jsonc.

Set secrets:

npx wrangler secret put REPLICATE_API_TOKEN
npx wrangler secret put EMAIL_FROM
npx wrangler secret put TURNSTILE_SITE_KEY
npx wrangler secret put TURNSTILE_SECRET_KEY
npx wrangler secret put R2_ACCOUNT_ID
npx wrangler secret put R2_ACCESS_KEY_ID
npx wrangler secret put R2_SECRET_ACCESS_KEY

• EMAIL_FROM must be a verified sender on a Cloudflare Email Service domain
• R2_* should be scoped to the clearmic-audio bucket with object read/write
• R2_CORS_ORIGINS in .env must include every deployed frontend origin that uploads directly to R2

Migrate D1 and deploy:

npm run db:migrate:remote
npm run deploy

Abuse protection

• Turnstile validates every OTP request before email is sent
• 10 processing jobs per email per rolling 24h window (counts queued jobs + pending uploads)
• Direct R2 uploads via 15-minute presigned URLs — Worker never buffers upload bodies
• Upload URLs sign exact Content-Type, Content-Length, and ClearMic metadata; /complete re-verifies the stored object
• Per-job random tokens guard Replicate input, webhook, and download routes
• Completion emails only include tokenized links that expire with the job
• R2 lifecycle hard-deletes after 1 day; D1 expires_at blocks access at exactly 24h
• R2 CORS pins browser upload origins
• OTPs: 10 min TTL, single-use, locked after 5 wrong attempts

---

_{Built on Cloudflare Workers · Powered by Replicate}