RNNoise evaluates short speech frames at its native 48 kHz processing rate.
DENOISE / AVAILABLE NOW
Noise Remover
Bring the voice forward. Reduce room noise, hiss, and low-level background texture in recordings.
DENOISE / dedicated toolSpeech-focused local neural filter
Before and after previews appear here.
GUIDE / remove background noise online
How does RNNoise remove background noise online?
To remove background noise online, SoundTools resamples speech to 48 kHz and runs RNNoise on consecutive 10 millisecond frames. A reduction-mix control blends the neural-filtered and original signals, then restores the source sample rate for local WAV preview and export. The selected recording is never sent to a denoising API.
RELATED USES
- noise remover free
- RNNoise browser
- clean voice recording
- reduce audio hiss
FIELD NOTES / 01
Match the reduction mix to the kind of noise
RNNoise is trained for speech cleanup, so it works best on a voice with steady fan noise, low room texture, or light hiss. Start near 50–80% wet mix and compare consonants, breaths, and word endings. Move toward 100% only when the filtered result remains clear; a lighter blend often sounds more natural than the fully processed signal.
The worker resamples each channel to 48 kHz, filters 480-sample frames, blends the wet and original signals, and returns to the source sample rate before WAV export. It is less suitable for music, changing street noise, keyboard impacts, another person speaking, clipping, or strong echo because those sounds can overlap useful speech in time and frequency.
REFERENCE / MODE CHOICE
RNNoise vs EQ vs a noise gate
These tools reduce different kinds of unwanted sound. The source determines which approach is useful.
| Method | Best for | What it changes | Important limit |
|---|---|---|---|
| RNNoise speech filter | Steady background texture under speech | A neural model estimates speech and noise in consecutive 10 ms frames | Can damage music, ambience, reverb, or unusual voices |
| EQ or high-pass filter | Rumble, hum, or a narrow frequency problem | Selected frequency regions are reduced across the complete recording | Voice energy in the same band is reduced too |
| Noise gate | Silence between phrases when the voice is already clear | Level falls when the signal drops below a threshold | Noise remains while the person is speaking and abrupt settings can cut word endings |
PROCESS / THREE STEPS
How to remove background noise from a voice recording
- 01
Choose speech with a consistent background
Open a voice memo, interview, or call excerpt. The browser decodes it locally; use a separate trim first if long silent sections are not needed.
- 02
Set a moderate reduction mix
Begin around 50–80%. Higher values use more of the RNNoise result, while lower values retain more of the original voice and room character.
- 03
Compare words, breaths, and noise
Render the complete recording, switch between original and noise-reduced previews, and download WAV only when intelligibility improves without distracting artifacts.
SIGNAL NOTES / VERIFIED BEHAVIOR
What this tool actually does
Clear limits are part of a useful tool. These values describe the processor currently running in this page.
Blend the original and neural-filtered signals instead of accepting only an all-or-nothing result.
The worker returns processed channels to the decoded source sample rate before local WAV creation.
USE CASES / 02
Background noise this speech filter can address
- Steady room and fan noise
Reduce low, consistent background texture beneath one clear speaker without sending the recording away.
- Voice memo and interview cleanup
Prepare a more intelligible local WAV before trimming, transcription, or a level adjustment.
- Light headset or call noise
Soften a stable electronic hiss or room bed while checking that consonants and word endings remain intact.
QUESTIONS / PRACTICAL ANSWERS
Questions about this tool
Answers based on the current browser processor—not promises about a future version.
01What background noise can RNNoise reduce?
It is most useful for steady room texture, fan noise, light hiss, and similar noise beneath speech. Results depend on how strongly that noise overlaps the voice.
02What reduction mix should I start with?
Start around 50–80%, then compare consonants, breaths, and word endings. Use 100% only when the fully filtered signal stays natural and intelligible.
03Can this noise remover clean music?
It is not designed for music. RNNoise focuses on speech and may remove instruments, ambience, or vocal detail from a musical recording.
04Will it remove echo, clipping, or another person talking?
Not reliably. Echo and other speech overlap the wanted voice, while clipping has already discarded waveform detail. Those problems need different recording or restoration methods.
05Why does the result sound metallic or watery?
The filter may be uncertain about whether part of the signal is speech or noise. Lower the reduction mix and use a cleaner, less reverberant source when possible.
06Does the output keep the source sample rate?
Yes. Processing uses RNNoise at 48 kHz internally, then the result is resampled back to the decoded source rate before WAV export.
07Is my recording uploaded for noise removal?
No. The recording is decoded, resampled, filtered, blended, previewed, and exported inside the browser and its dedicated worker.