tomai
Log in
☁️ Cloud · credits

Audio Denoise

Remove background noise, hiss and hum from recordings with adaptive spectral filtering

🪙 Server-side noise filtering: 2 credits per job

Adaptive spectral noise suppression — Strong removes more hiss but may soften quiet vocals.

Server-side FFmpeg handles any format — including HEVC/H.265 files your browser can't play. Files are deleted from the server within 30 minutes.

How it works

  1. 1

    Upload the noisy recording — MP3, WAV, M4A, FLAC or any other audio format, and most videos.

  2. 2

    Pick the strength: Light for barely audible hiss, Medium for typical room noise, Strong for very noisy sources.

  3. 3

    Choose an output format and download the cleaned file. Jobs are processed on our server by FFmpeg's adaptive spectral denoiser.

Clean speech without the underwater sound

Three strength levels tuned for voice: light hiss removal, standard room-noise cleanup, and aggressive fan/AC suppression that still keeps consonants crisp.

🌫️

Smart noise removal

Spectral noise reduction removes hum, hiss and room tone while keeping the useful signal — no expensive hardware needed.

🎚️

Three strength levels

Light, medium or strong — pick how aggressively to clean depending on how noisy your recording is.

🔊

Speech stays clear

The voice or music you actually recorded stays front and center — only the noise floor is lowered.

How spectral noise reduction works here

The worker analyzes the first seconds to profile the noise floor, then applies spectral gating across the file:

Output keeps the original sample rate and container (MP3 in, MP3 out; WAV stays WAV). Music is deliberately not the target — for songs, use the vocal separator instead.

Frequently asked questions

How does the denoiser remove noise without damaging the audio?

Adaptive spectral noise reduction (FFmpeg afftdn): the filter continuously estimates the noise floor of each frequency band and suppresses it, leaving speech and music mostly untouched. It is not a simple high-pass filter — it removes noise inside the voice range too.

When should I use Strong instead of Medium?

Strong drives down heavier noise (traffic, crowds, loud fans) but can slightly soften quiet vocals. Start with Medium for typical room noise — Strong is for genuinely noisy recordings.

Will it remove constant hum (50/60 Hz)?

Yes — steady hum and hiss are exactly what spectral noise reduction is best at. Sudden one-off sounds (a door slam, a cough) are not filtered, because they are not background noise.

Related tools

What is audio denoising?

Audio denoising separates the sound you want from the steady background noise hiding underneath it — room tone, fan hum, traffic, keyboard clicks or tape hiss. This tool runs FFmpeg's built-in adaptive spectral filter on our server: instead of a blunt frequency cut, the filter continuously estimates the noise floor of every frequency band and suppresses exactly that level, so speech and music stay almost untouched while the hiss disappears. You upload any audio file (most video files work too), choose how aggressively the noise should be reduced, pick an output format, and download a clean recording. Podcasters, students, interviewers and anyone rescuing old recordings can use it without touching a single filter setting.

What the denoiser does

  • 🌫️ Removes steady background noise — room tone, fan hum, traffic, keyboard clicks and tape hiss
  • 📈 Estimates each frequency band's noise floor and suppresses only that level, leaving speech and music intact
  • 🎚️ Three strength presets: Light for faint hiss, Medium for typical room noise, Strong for very noisy sources
  • 🔌 Handles constant 50/60 Hz mains hum, the classic low-end buzz
  • 🎧 Accepts any audio format and most video files; exports MP3, WAV, OGG, AAC or FLAC
  • ⚡ Runs entirely on our server with FFmpeg's built-in filter — nothing to install

Good times to use it

  • 🎙️ Cleaning a podcast or voice memo recorded with a fan or air conditioner running
  • 🎮 Removing keyboard clicks and PC fan noise from a game recording or lecture capture
  • 📼 Rescuing an old interview or cassette rip that hisses and hums
  • 🤝 Polishing a call recording before sharing it with a team or client

Your recording is sent to our server, where FFmpeg's spectral denoiser processes it, and no copy is kept afterwards — uploads and results are wiped on a 30-minute schedule. The filter targets steady background noise: sudden one-off sounds like door slams or coughs are left alone, and Strong reduction can slightly soften quiet vocals, so start with Medium for typical room noise. Files up to 1 GB are supported, and each job costs a small number of credits.

Related tools