# Audio Denoiser

> Remove background noise, hiss & hum from any audio file — in your browser

Free audio denoiser in your browser. Remove hiss, hum, rumble and ambient noise from MP3, WAV, M4A, FLAC. 4 presets, custom sliders, 100% private.

URL: https://tools.scoreroute.com/tools/audio-denoiser/

Markdown: https://tools.scoreroute.com/tools/audio-denoiser/.md

### Why this audio denoiser is different

AI chatbots like ChatGPT, Claude and Gemini cannot process your audio file with a real-time biquad-filter chain. They have no AudioContext, no OfflineAudioContext, no way to render a denoised AudioBuffer or encode a 16-bit PCM WAV for download. This tool runs the entire pipeline natively in your browser using the Web Audio API - the same engine that powers online DAWs and live broadcast software. Unlike paid services like Krisp.ai, Adobe Podcast (cloud) or Audo.ai (subscription), this is unlimited, free, and 100% private: your audio never leaves the device.

Click to choose, or drag an audio file here **MP3, WAV, M4A, FLAC, OGG, AAC, WEBM** · stays on your device 2. Choose Preset or Customise Voice Music Outdoor Custom High-pass cutoff (removes low rumble & hum) 80 Hz Powerline notch (mains hum removal) 60 Hz Set to off if your recording has no electrical hum. 60Hz is North America/Japan; 50Hz is Europe/Asia/Africa.


## How It Works

This denoiser uses an adaptive filter chain from the Web Audio API. Each stage targets a different kind of background noise:

Because every stage runs locally via OfflineAudioContext, your audio never touches a server. There are no uploads, no AI models, no daily limits and no watermarks.

- **High-pass filter** — Removes low-frequency rumble, microphone handling noise, plosives, traffic rumble and the bottom of mains hum. The default 80Hz cutoff is a sweet spot for speech: it kills air-conditioning drone and computer fan noise without thinning male voices.
- **Powerline notch** — A narrow notch filter at 50Hz (Europe/Asia/Africa) or 60Hz (North America/Japan) cancels the steady electrical hum produced by cheap power supplies, lighting ballasts and unbalanced cable runs. Disable it if your recording is clean of mains interference.
- **Low-pass filter** — Removes high-frequency hiss from cheap microphones, preamp self-noise and tape hiss. The Voice preset uses 8kHz to keep speech intelligible; Music uses 16kHz to preserve cymbals and air.
- **Noise gate** — A DynamicsCompressorNode configured as a downward expander: any signal quieter than the threshold (default -45dB) is squashed, while signal louder than the threshold passes unchanged. This silences the room tone between spoken sentences without cutting speech itself.
- **Make-up gain** — After everything else, a final gain stage restores perceived loudness so the cleaned audio is comfortable to listen to.


## Why AI Chatbots Cannot Do This

Even though modern LLMs can write essays, transcribe audio and translate between languages, they cannot denoise a user-uploaded audio file in their standard chat interface:

That is why this tool exists: it solves a real audio problem that AI assistants cannot address through their standard chat interface.

- **No audio graph runtime** — ChatGPT, Claude and Gemini accept text (and sometimes short audio clips). They have no AudioContext to build a BiquadFilterNode chain.
- **No biquad filtering** — Computing a difference equation `y[n] = b0*x[n] + b1*x[n-1] + b2*x[n-2] - a1*y[n-1] - a2*y[n-2]` on millions of samples per second is a DSP operation that text-based chat assistants cannot perform.
- **No WAV output** — A chatbot cannot render an OfflineAudioContext, take the resulting AudioBuffer and write a 16-bit PCM WAV header with the correct sample rate, bit depth and channel layout, or stream a downloadable binary blob.
- **No auditioning** — You need to *hear* whether denoising worked. AI can describe what clean audio *should* sound like, but cannot let you click Play on the result.


## Use Cases


### Podcast Cleanup

Strip room tone, computer fan noise and electrical hum before publishing an episode. Voice preset is tuned for spoken-word intelligibility.


### Voice Memos & Lecture Capture

Clean up phone-recorded lectures and voice memos recorded in noisy rooms. Make muffled phone calls sound like a professional recording.


### Outdoor Recording

Reduce wind noise and rumble from street interviews, field recordings and live event captures. Outdoor preset pushes the high-pass to 200Hz.


### Music Stem Polishing

Remove amplifier hum and tape hiss from older recordings before remastering. Music preset preserves the full frequency range of the original.


### Call Center & VoIP

Improve the listening quality of recorded Zoom calls, dictation and conference captures. Customise the gate to match the caller's mic environment.


### Accessibility & Transcription Prep

Reduce noise before feeding a file into automatic transcription: cleaner signal means higher word-recognition accuracy and fewer "inaudible" tags.


## Frequently Asked Questions

**How does this audio denoiser work?**

The tool applies a chain of Web Audio API filters in series: a high-pass biquad filter removes low-frequency rumble and mains hum below the cutoff, a powerline notch filter cancels 50Hz or 60Hz electrical interference, a low-pass biquad filter removes high-frequency hiss above the cutoff, a dynamics compressor set as a noise gate silences signal below the threshold, and a make-up gain stage restores lost loudness. All processing happens locally in your browser via OfflineAudioContext — no audio is uploaded to any server.

**Will it work on every recording?**

It works well on speech and voice recordings, podcasts, calls, lecture captures and interviews where background noise is mostly low-frequency hum, hiss, fan noise or steady ambient sound. For non-stationary noise such as crowd chatter, sudden bangs or music bleed-through, results vary — you may need to try multiple presets or fall back to a tool with AI spectral subtraction like Audacity.

**Is my audio file uploaded to a server?**

No. The entire pipeline runs locally in your browser using the Web Audio API and OfflineAudioContext. The audio file is decoded into PCM samples, processed in memory, encoded back to WAV, and downloaded. It never leaves your device.

**What audio formats are supported?**

The tool accepts any audio format your browser can decode with AudioContext.decodeAudioData: MP3, WAV, M4A, FLAC, OGG, AAC, WEBM and OPUS. The output is always a 16-bit PCM WAV file at the original sample rate and channel layout (mono or stereo).

**What is the difference between presets?**

Voice preset uses 80Hz high-pass, 60Hz notch, 8kHz low-pass and -45dB gate — optimized for speech clarity. Music preset uses 40Hz high-pass, no notch, 16kHz low-pass and -55dB gate — lighter touch that preserves bass and treble. Outdoor preset uses 200Hz high-pass, 50Hz notch, 5kHz low-pass and -35dB gate — aggressive wind and rumble reduction. Custom exposes every slider so you can fine-tune each stage.

**Can I preview before downloading?**

Yes. The waveform canvas supports side-by-side A/B comparison: the blue waveform is the original and the green waveform is the denoised result. Use the play buttons under each canvas to audition before downloading. You can also switch presets and click Reduce Noise again to compare settings.

**Why does the file size change after denoising?**

The output is encoded as 16-bit uncompressed PCM WAV at the original sample rate and channel layout, so the byte size is duration * sampleRate * channels * 2 bytes + 44 byte header. The same rules apply whether the audio was originally MP3, FLAC or WAV — the container changes to .wav but the underlying samples keep their native rate.

**How is this different from ChatGPT or Claude?**

AI chatbots cannot process a user-supplied audio file with a biquad-filter chain in their chat interface. They have no AudioContext, no OfflineAudioContext, no way to apply biquad filter coefficients, no way to render a new AudioBuffer, and no way to encode and download a 16-bit WAV. This tool runs the entire pipeline natively in your browser using the Web Audio API — the same technology that powers online DAWs, podcast editors and live broadcast software.

## References

- [MDN — Canvas API](https://developer.mozilla.org/en-US/docs/Web/API/Canvas_API)
- [MDN — HTMLMediaElement](https://developer.mozilla.org/en-US/docs/Web/API/HTMLMediaElement)