auphonic.com - Entries for the category Audio

Automatically Cut Music Segments from Your Productions

lukasm@auphonic.com (Lukas) — Wed, 11 Feb 2026 12:10:52 +0000

We’ve expanded our Automatic Cutting feature to make it even more useful for broadcast and long-form productions. You can now remove entire music segments with Auphonic. For example, songs in radio shows, podcasts, or recorded broadcasts, which must be removed due to licensing restrictions.

In many broadcasting workflows, this is not just a convenience but a requirement: music may be included in live transmissions, but archived or on-demand versions are often not permitted to make that music permanently available. Automatic music removal ensures compliance while keeping spoken content intact.

Designed for Longer Music Segments

The new Cut Music option is optimized for longer music tracks, like full songs or extended musical interludes. Typical use cases include:

Radio broadcasts with songs between interviews
Live shows with music breaks
Recordings where only spoken content should remain

Once enabled, Auphonic detects these music sections and removes them for you.

Auphonic Editor with Cut Music Segments

Refine your Automatic Results in the Auphonic Editor

The new Music Cutting Feature automatically cuts music segments, and lets you refine these cuts in the Auphonic Editor. Auphonic does the heavy lifting, and you decide exactly what stays and what goes.

Start with our Automatic Cuts and then fine-tune them by ear in the same interface. You can:

Zoom in and click on a cut to select it
Activate or Deactivate Cuts
Create your own custom cuts
Adjust cut regions to your liking and refine by ear

This combined workflow is the same philosophy we use in our Denoising Region Editor, where you refine the automatic denoising Auphonic applies to get exactly the output you're looking for.

NOTE: When you're happy with your cuts, press Apply Changes to reprocess with your new edits. Reprocessing your production will not cost any extra credits.

Cut Music vs. Speech Isolation

It’s important to distinguish between removing entire music tracks and removing background music.

Cut Music removes complete music segments from your production.

Speech Isolation (found under Noise Reduction) is the right tool for removing background music underneath speech.

You’ll find the option to remove music segments in the Automatic Cutting section of your production settings. Simply enable Cut Music, process your audio, and Auphonic will take care of the rest.

If you want to remove background music from your recording, use Noise Reduction > Speech Isolation. Here is additional information on how Auphonic Denoising works, and how we tackle the complex problem of automatically improving audio recordings.

As always, the goal is to save time while keeping full control over your final output.

Feedback

We’re continuously improving Auphonic based on real production workflows. If you have feedback, ideas or want us to think of other features, please reach out via our Contact Page.

You can always send us an email to support@auphonic.com, or give feedback on our results page.

RMS Loudness Normalization for Audible / ACX

grh@auphonic.com (Georg) — Thu, 15 Jan 2026 06:59:41 +0000

Audiobook production has a uniquely stressful property: technical delivery specs are strict, and a single out-of-range file can delay an entire release. Audible / ACX is a common example with strict criteria that uploaded files need to fulfil before they're accepted.

That’s why we’ve added an additional loudness normalization method in Auphonic: RMS-based normalization. It’s designed specifically for workflows that need RMS targets, including Audible / ACX.

What ACX requires (loudness and peaks)

ACX asks for each file to meet a level measured in RMS and a maximum peak limit. The headline values as stated in their ACX audio submission requirements are:

Average level: between -23 dB RMS and -18 dB RMS (per file)
Peak level: peaks must be below -3 dB

In practice, this means you need two things: a consistent overall level, and enough headroom so transients never exceed the peak ceiling.

Recommended ACX settings in Auphonic

If you want one simple, reliable starting point, use the preset below.
It targets the middle of the ACX RMS window and applies the correct peak ceiling.

Web Interface

In the Production Form, open Loudness Normalization and set:

Loudness Target: -20 dBRMS (Audible / ACX)
Maximum Peak Level: -3 dBTP
Normalization Method: RMS Levels (Audible / ACX)

Recommended Auphonic Loudness Normalization settings for Audible / ACX: -20 dBRMS, -3 dBTP, RMS Levels

Learn more about the Auphonic Loudness Normalization Settings!

API

For API users, send the parameter:

"loudnessmethod": "rms", "loudnesstarget": -20

Learn more about How the Auphonic API works!

LUFS vs RMS: why loudness measurement matters

Different measurement methods answer slightly different questions. Some aim to reflect how loud audio is perceived by a listener, others describe the average signal energy over time. Both are valid, but they are not interchangeable - and this distinction is exactly where ACX-related issues originate.

Understanding the difference helps explain why the same audio file can appear “in range” in one tool and fail a check in another.

Interested in learning more?
Here are detailed articles on How Loudness Normalization works in Auphonic.

LUFS: standardized perceived loudness

LUFS (Loudness Units relative to Full Scale) is a standardized loudness measurement defined in international broadcast standards. Its goal is to reflect perceived loudness as consistently as possible across different types of content.

To achieve this, LUFS measurements:

apply frequency weighting to approximate human hearing sensitivity
measure loudness over time, not just instant levels
use gating to reduce the influence of very quiet passages and long silences

This makes LUFS particularly robust for spoken-word content, music, and mixed material. Long pauses between sentences or chapters don’t disproportionately lower the loudness value, which means LUFS tends to align well with how loud a listener experiences the program overall.

RMS: average signal energy

RMS (Root Mean Square) is a way of calculating the signal energy that is no longer commonly used for loudness targets today. It measures the average energy of the audio signal over time, without attempting to model human perception.

In contrast to LUFS:

RMS does not apply perceptual frequency weighting
RMS typically does not include standardized gating
silence and very quiet passages directly lower the average value

This means RMS values are strongly influenced by pauses, gaps between phrases, and overall dynamics. Two narration files with identical spoken content but different pause lengths can produce noticeably different RMS readings.

Why this matters for ACX

ACX specifies its loudness requirements in terms of RMS.

For audiobook production, this difference is crucial. ACX requires every individual file to fall within a strict RMS range, and even a small deviation can lead to rejection and delays in publication.

Auphonic’s LUFS normalization has been used by many producers over the years. But if your workflow is built around RMS targets, or you want a method that aligns more directly with ACX’s published wording, RMS normalization is now available.

Feedback

We’re continuously improving Auphonic based on real production workflows. If you have feedback on RMS normalization for audiobooks, ideas or wishes for other features or feeback for existing ones, please reach out via our Contact Page, email support@auphonic.com, or send us feedback on our results page.

Denoising Editor: Choose Where (and How) to Reduce Noise

lukasm@auphonic.com (Lukas) — Tue, 02 Dec 2025 09:41:33 +0000

With our new Denoising Segment Editor feature, you can now control exactly where and how to reduce noise inside a production: Refine our automatically generated denoise segment boundaries, choose different denoising methods per segment, or switch denoising off completely for things like jingles, embedded music, or sound effects.

The result: you only clean what really needs cleaning, gain more creative control, and work with much higher precision inside a single production.

Only Denoise the Parts That Need It

Many productions contain a mixture of different situations:

Music jingles or sound effects combined with noisy speech
A clean studio introduction followed by a noisy on-location interview
Speech combined with short news clips, where ambiance and noises should not be removed
A short section where a fan or laptop noise becomes audible

Segment-Based Denoising lets you target specific segments of your production, and leaves the rest of the file untouched. The segments are created automatically by our Dynamic Denoiser, which classifies the noise conditions of your audio and generates initial segment boundaries. You can refine these segments manually now - similar to how you adjust cuts in our Cut Editor.

It is also possible to exclude denoising entirely in segments where you don’t want it at all - for example Jingles, embedded music, sound design elements, or external audio inserts.

Auphonic Editor with Output and Input waveforms and editable denoise segments

How Segments Work

The new segment editor for denoising is an additional layer on top of our Auphonic Editor:

You can split existing segments wherever you like.
You can move segment boundaries to match words, pauses, or noisy events more precisely.
You can create a sequence of smaller segments inside one automatic segment if needed.

The automatically detected segments give you a fast starting point which you can adjust for perfect results.

Denoising Segments in multiple colors - change methods or turn denoising on/off per segment

Every segment on the Denoiser layer can have its own processing setting. For each segment you can: turn denoising on or off completely or select which denoising method you want to use - for example our Speech Isolation, Static & Music Denoiser, or Classic Denoiser.

This keeps the interface simple and consistent. You only decide where to reduce noise and which algorithm to use.

How to Use Segment-Based Denoising

Run your production with Noise Reduction
Open the production result page to see our Auphonic Editor
Split or move segment boundaries in the Denoiser layer
Select the desired denoising method for that segment, or turn denoising off
Press Apply Changes or Edit Production to reprocess (reprocessing these edits does not consume additional credits)

Try our new Segment-Based Denoising now!

Conclusion

Noise in your recordings changes over time, so it makes sense that your denoising settings can change as well.
Segment-Based Denoising lets you:

Denoise only the parts that actually need it
Combine multiple denoising methods in a single production
Fix misclassified or delicate sections without touching everything else
Work with more precision while keeping the workflow simple

Open a finished production in our Auphonic Editor, switch to the Denoiser view, and start experimenting with segments.
You will quickly hear how much more control you gain over the sound of your final output.

Segment-Based Denoising is a major step in our broader direction: blend powerful automation with precise human refinement. Expect more features following this philosophy to arrive next year.

Feedback

As with all new features, we are very interested in how you use Segment-Based Denoising in your workflows. If you have feedback, questions, or suggestions, please Contact Us or send us a message via the production feedback form.

Multitrack Clarity Redefined: Introducing our new Mic Bleed Remover

grh@auphonic.com (Georg) — Wed, 08 Oct 2025 05:30:35 +0000

You know that faint trace of your co-host’s voice leaking into your mic channel?

That’s mic bleed—and it’s why multitrack sessions can sound like one noisy room instead of one clean mix. Every mic hears everything: voices, ambience, reflections. The result is overlapping speech and unclear separation that’s hard to fix in post.

Our new Mic Bleed Remover tackles the problem at the session level: it analyzes all tracks together, learns what belongs to each microphone, and removes the unwanted cross-talk while preserving natural ambience and timing.

Try Mic Bleed Removal

What is mic bleed and why is it hard to remove?

Mic bleed happens whenever multiple microphones are used in the same space. Each one not only records its intended speaker, but also the others nearby. You’ll hear it in podcast interviews, roundtable discussions, or any setup where several people talk at once in the same room. Even with one mic per person, voices still spill into neighboring tracks.

Traditional tools like noise reduction or gates can’t really fix this: they’re designed to remove constant background noise, not speech. Gates simply mute parts of a track, often cutting off room tone and making conversations sound unnatural. That’s why mic bleed is one of the harder audio problems to solve cleanly.

From Crossgate to Continuous Improvement

We first tried to solve this years ago with our Crossgate – a smart gate that muted bleed when it detected it. It worked, but only in ideal conversations: one person talking at a time, clean room, no overlaps. Real-world recordings are messier. People interrupt. Tracks drift. Rooms echo.

We weren’t satisfied with our old model, so we decided to build something entirely new.

The new model: True Mic Bleed Removal

Our new Mic Bleed Removal model takes a completely different approach. It’s the first system that actually understands your session – not just mutes it.

Instead of processing each track in isolation, it listens to all tracks together and learns what belongs to each microphone. It then removes only the parts that came from other mics, keeping track noise, ambience, music and everything else intact.

This allows it to handle even difficult cases: overlapping speech, misaligned tracks, strong bleed, and background noise. The result is clean, separated voices that still sound natural, with all the original ambience and timing preserved.

The new model is fully deployed and available right now in all Auphonic Multitrack productions.

Hear the difference

The best way to understand Mic Bleed Remover is to listen.

Mic Bleed Remover Inactive
Mic Bleed Remover + Denoise:

How to use Mic Bleed Removal

Cleaning up multitrack recordings is now easier than ever. Just upload your multitrack session, enable Mic Bleed Removal, and let Auphonic do the rest.

Select Remove Mic Bleed Option in Master Audio Algorithms Section

The algorithm automatically analyzes all microphones together, identifies what belongs to each track, and removes unwanted cross-talk while keeping ambience and timing perfectly natural.

No complex setup. No manual gating. Just clear, separated voices straight out of the mix. To guarantee everything works for everyone, we’ve also put together a list of Multitrack Best Practices .

Conclusion

Mic bleed used to be one of those problems you just had to live with. Now, you don’t.
Upload your multitrack, switch on Mic Bleed Removal, and hear your session open up.

One room, one take - perfectly separated.

Try it now → auphonic.com/engine/multitrack

Feedback

We drive innovation through your feedback and would love to hear how the new Mic Bleed Remover works for you. Send us your thoughts through the production feedback form or reach out via our Contact page. Every bit of input helps us fine-tune things further.

New Auphonic Static and Music Denoiser

lukasm@auphonic.com (Lukas) — Thu, 31 Jul 2025 09:43:32 +0000

Technology is evolving fast and we’re the ones pushing it forward.

Our new Static Denoiser removes steady background noise like hiss, hum, or fan noise while keeping music, ambience, and sound design fully intact. Perfect for audio dramas, videos, music, meditations, podcasts, or anything where clarity matters and the atmosphere does too.

Auphonic’s Vision for Noise Reduction

We firmly believe that there is no “one button to fix it all.”

But how can we say that when it is our mission to build exactly that?

Our goal is precision: Giving users full control over what stays and what goes.

That’s why we offer different tools for different kinds of noise:

Speech Isolation removes everything but your voice. It keeps only the speech you care about.
Dynamic Denoiser adapts in real time to changing environments - ideal for unpredictable noise patterns, while leaving music intact.
Static Denoiser now updated to precisely target stationary noise - like constant hiss or hum - while preserving music, ambient effects, and subtle details.

Think of audio dramas where sound effects should stay untouched, or meditation recordings with soft tonal elements that must remain intact. Unlike non-stationary noises (e.g. coughs, chair squeaks or mouth clicks), stationary noise is consistent over time - making it ideal for static removal without harming your content.

While we believe in precise user control, there is a way to reduce your audio editing work to almost one button: Saving your favorite settings as Custom Presets and applying those to your productions. You can even share your Presets with friends!
But more on that later 😉

How to Use the New Static Denoiser

Cleaning up your audio is easier than ever: Just upload your file, choose the Static Denoiser in the Production Form, and let Auphonic do the rest. No deep tech knowledge required.

Fine-tune the sound by adjusting the Remove Noise and Remove Reverb sliders. Lower them slightly to retain some natural texture while still boosting clarity and speech intelligibility.

And for everyone still wanting to use the Original Static Denoiser: Don't worry, the Legacy Denoising Version is still available. We renamed it to "Classic" and you can use it as normal. Just select it through the drop-down menu.

Audio Examples

It’s hard to read audio, but easy to hear the difference. Here is a sound comparison of our denoising models - please use headphones to hear all details!

New Static Denoiser vs. Speech Isolation

We processed a snippet from the History of Jazz podcast using both Speech Isolation and the new Static Denoiser. Speech Isolation removes everything but the speech - including music, background vocals, and ambience - resulting in a clean, voice-only track. In contrast, the Static Denoiser keeps the musical texture intact while removing just the steady background noise. Hear the difference for yourself!

Original:
Static Denoised:
Speech Isolated:

New Static Denoiser vs. Dynamic Denoiser

In this excerpt from the German audio drama Der Graue, the “Static Denoiser” perfectly preserves all the sound effects while removing the reverb and static noise, whereas the “Dynamic Denoiser” (or “Speech Isolation”) removes everything from the audio that is not speech:

Original:
Static Denoised:
Dynamic Denoised:

New Static Denoiser vs. Legacy Classic Denoiser

To show how far we’ve come, we processed a short segment from the radio play Around the World in 80 Days using both the new Static Denoiser and the legacy Classic Denoiser.
The Classic Denoiser extracts noise prints in speech pauses, which is not possible in this example because of the background music, whereas the new model removes noise cleanly.

Original:
Old Classic Denoised:
New Static Denoised:

Try It Out

The new Static Denoiser is live! Perfect for cleaning up hiss and hum while keeping music and sound effects intact.

If you're used to the legacy model and want to keep using it: No worries, the Classic Denoiser is still available. Want more aggressive cleanup or even full music removal? Try our Dynamic Denoiser or Speech Isolation models .

Re-processing is free - so go wild:
As long as you don’t change your input file, you can tweak your production and test different settings without using extra credits. Try it. Break it. Compare models. Save your findings as your favorite preset. And please, tell us what you think!

Feedback

We drive innovation through your feedback and would love to hear how the new Static Denoiser works for you. Send us your thoughts through the production feedback form or reach out through our Contact Page.

Every bit of input helps us fine-tune things further.

Fillerword Cutting Optimized for Greek, Romanian and Hungarian

lukast@auphonic.com (Lukas) — Tue, 22 Apr 2025 06:04:05 +0000

Fillerword detection is highly language specific. While “uhm” and “aah” might be more common across the board, each language has its own unique fillerwords. We take this into account when improving our algorithms. Auphonic lets you automatically cut unwanted parts from your audio production, easily removing silence and fillerwords in multiple languages.

Introducing Greek, Romanian and Hungarian

We develop and train all of our algorithms in-house - and in our latest rollout, we've improved the detection of fillerwords in Greek, Romanian and Hungarian.

Fillerwords from these languages will now be detected more reliably and automatically removed from your productions.

To try it out for yourself, simply select “Cut Fillers” with our Automatic Cutting Feature either in the production page:

Feature Selection in the Production Page

Or you can select the Auphonic Preset "Cut Filler Words and Silence” in our Quick Production Screen on the landing page:

Auphonic Quick Production Presets

What do you think?

We drive innovation through your feedback and we would love to hear your thoughts! Do you have ideas for what this feature could evolve in the next iteration? Or suggestions for new features you'd like to see in Auphonic?

Please, send us your thoughts HERE, or through the feedback form in your production (as seen below)!

Feedback Form in your Production Result

We're looking forward to hearing your ideas and wish you happy editing!

---
"Ακούγεσαι εξυπνότερος με την Auphonic!"
- Message from a Greek friend

New Bandwidth Extension Feature with Enhanced Voice AutoEQ

mpagavino@auphonic.com (Manuel) — Fri, 04 Apr 2025 04:57:59 +0000

Ever struggled with muffled, low-quality audio in voice calls or old recordings? Our latest feature — Bandwidth Extension, now integrated with the Voice AutoEQ — is here to change that.
Audio bandwidth extension (BWE) intelligently restores missing high frequencies, making speech clearer and more natural, even in archival or low-bitrate recordings. Listen to our BWE audio examples below to hear the difference for yourself!

What is Audio Bandwidth Extension (BWE)?

Audio bandwidth extension is a technique used to artificially recreate or “extend” higher audio frequencies that may have been lost or limited during recording, transmission, or compression. By analyzing the existing frequency information in an audio signal, our algorithm can predict and synthesize additional higher-frequency components. This approach helps restore the brightness and clarity of an audio track, making it sound more natural and pleasant to listeners.

Bandwidth extension is particularly helpful in scenarios such as working with archival speech recordings, low-bitrate voice calls, and compressed audio streams where high frequencies are lost due to bandwidth constraints.
The Auphonic bandwidth extension is specifically optimized for speech and does not enhance music or environmental sounds, like noise, reverb, etc.

Listening Examples

The following example from the Lex Fridman Podcast #457 demonstrates how it is possible to take a high-quality recording, downsample it to 4 kHz bandwidth, and then use our bandwidth extension algorithm. The result comes remarkably close to the original audio with a full 20 kHz bandwidth.
Hear the difference yourself as the audio regains its natural clarity and depth:


Original Audio:

Low Bandwidth:

Bandwidth Extended:

Bandwidth extension of historic recordings (male and female speaker)

In practice, bandwidth extension is useful for restoring historic recordings that have been downsampled due to bandwidth constraints. This example is a record of the historic “Margaret Thatcher's Iron Lady speech” and shows the difference between the original recording and the bandwidth-extended version:

Low Bandwidth:
Bandwidth Extended:
BWE and Denoise:

We added a version with and without denoising here, to hear that our algorithm does not add frequencies to environmental sounds like noise, reverb, etc.

The second example is an archive recording of Richard Nixon's “End Of Vietnam War speech”:

Low Bandwidth:
Bandwidth Extended:
BWE and Denoise:

How to use it?

To activate the new bandwidth extension feature, just select the option “Voice AutoEQ + Bandwidth extension” as “Filtering Method” in the section “Filtering”:

Or use the Auphonic preset “Bandwidth Extension and Voice AutoEQ (keep noise)”:

Bandwidth extension is always applied in combination with the Voice AutoEQ, where we also released some updates ...

Voice AutoEQ Updates

We took some time to further refine our Voice AutoEQ algorithm, please also take a look at our previous blog post.
Now the Voice AutoEQ works more effectively with reverberant speech and can adapt even when speech is mixed with background music.
The de-plosive/de-pop feature that reduces strong plosives in speech is now integrated into our Noise Reduction Algorithm making Voice AutoEQ even more effective when combined with our Speech Isolation or Dynamic Denoiser.

Voice AutoEQ Listening Examples

Voice AutoEQ with background music

This is an example from the Climate Of Joy podcast intro. The enhanced version improves speech intelligibility without affecting the music balance.

Original:
AutoEQed:

Voice AutoEQ with reverberant speech

This recording by conduitministries.com is recorded in a very reverberant room. The original audio sounds slightly dull, but the Voice AutoEQ version preserves the reverberation while adding more clarity. However, if you want to decrease the reverb, you could also select the Remove Reverb amount option in our Noise Reduction section.

Original:
AutoEQed:

Reducing plosives in speech

This snippet extracted from a video about Pop Filter vs Windscreen shows how our Noise Reduction algorithm — here in combination with the Voice AutoEQ — can be used to reduce plosives in speech.
(Please listen with headphones or good loudspeakers to hear the difference.)

Original:
Processed:

What do you think?

Your input helps us make our audio features even better. Give these improvements a try and let us know what you think!
Whether it is a game-changer for you or there is still room to improve, we would love to hear from you in the production feedback or through our support channels.

Speaker Identification in Single Track Productions

lukast@auphonic.com (Lukas) — Mon, 05 Aug 2024 09:34:04 +0000

We're excited to announce the release of our new speaker diarization feature! With our latest update, we can now detect individual speakers within a single audio recording, allowing users to see precisely who said what and when. This information is then reflected in the transcript.
Previously, this feature was only available in Multitrack Productions.

What is Speaker Diarization?

Speaker diarization is the process of identifying different speakers and segmenting an audio file accordingly, making it easier to accurately determine who said what. This is particularly useful in scenarios like meeting recordings, interviews, and podcasts involving multiple speakers.

What's New

Until now, Auphonic users had to use our Multitrack Production Workflow to gain access to speaker identification. With our recent update, this feature is now also accessible for single track productions, thereby further simplifying our workflow.

How It Works

When selecting Auphonic Whisper ASR as your speech recognition engine on our production page, the option to use “Speaker Detection” appears. You can select the number of speakers in the recording yourself, or have the AI identify and label them automatically.

Select "speaker detection" in the speech recognition menu

The identified speakers will then be shown next to what they said in the captions of our audio player and within the generated transcript. We know exactly who is saying what at any given time.

If you need to change speakers' names, you can edit them in our Transcript Editor. By changing the name of the speaker, the changes will be applied automatically within the whole transcript.

Click here for an example transcript to play around with!

The transcript editor also lets you edit the generated transcript's content and correct any mistakes. This gives our users the final say in quality control.

For more information on how to use our transcript editor, click here!

Conclusion

With our newly released speaker detection feature for single track productions, we simplified the automatic transcription process even further. You can drop a single audio file into our algorithms, the AI automatically detects and transcribes its contents and attributes speakers correctly. And if the model should unexpectedly be wrong at any time, you can correct it in our Transcript Editor.

Feedback

We're looking forward to having you utilize our new speaker diarization feature!
Please, feel free to send us feedback through our Contact Form or submit it directly in the status page of your production.
Happy detecting!

Independently control Noise, Reverb and Breath Reduction Amounts

mpagavino@auphonic.com (Manuel) — Thu, 16 May 2024 13:20:55 +0000

Responding to your feedback, we are now proud to present new separate parameters for noise, reverb, and breath reduction to give you more flexible control for your individual, best output results.
Find all the new parameters below and listen to the Audio Examples to get a closer impression of the upgrade.

What's the update about?

Before

Previously, you could only set the Denoising Method and one reduction amount, that was used for all elements.
Depending on the selected method, you were already able to decide whether music, static, or changing noises should be removed, but there was no setting to keep the typewriter sound effects while removing the reverb, for example.

Now

With our latest upgrade, you can now set the reduction amounts separately for noise, reverb, and breathing sounds.
For example, you could completely remove the background noise while reducing the reverb just a little to enhance speech intelligibility but keep the atmosphere. Like we did in Audio Example 1.
Many of you have also asked about the possibility of slightly reducing breath sounds rather than eliminating them completely. In Audio Example 2 we demonstrate how you can prevent your audio from sounding strange and unnatural by reducing instead of eliminating all breathing sounds.

To all of you who are happy with the results and don't want anything to change, relax:
If you don't change the default settings, the noise reduction algorithms work exactly the same as before.

Note: As the 'Static Denoiser' removes only stationary noise, there are no 'Remove Reverb' and 'Remove Breathings' parameters available for this denoising method.

New Parameters

Screenshot of the new Noise Reduction Parameters in the production form.

In order to use the new noise reduction features, you may separately set the following parameters:

Denoising Method: (unchanged) Select what kind of noise you want to remove.
[Dynamic Denoiser (default), Speech Isolation, Static Denoiser]
Note that the parameters 'Remove Reverb' and 'Remove Breathings' are NOT available for Static Denoiser!
Remove Noise: Select the amount of noise you want to remove.
[100 dB (default), Disable Denoise, 3 dB, 6 dB, ..., 100 dB (full)]
Remove Reverb: Select the amount of reverb you want to remove.
[100 dB (default), Disable Deverb, 3 dB, 6 dB, ..., 100 dB (full)]
Remove Breathings: Select the amount of breathings you want to remove.
[Off (default), 3 dB, 6 dB, ..., 100 dB (full)]

Feel free to experiment with all the options to find your preferred parameter settings! Editing and reprocessing existing productions does not cost any additional credits as long as you don't change the input file.

Listen to the results:

1. Reverb reduction with full noise elimination

For the first audio example by conduitministries.com we set the 'Remove Noise' amount to 100 dB (full) and varied the 'Remove Reverb' amount starting from 0 dB (Off) to 12 dB (medium) and to 100 dB (full). Listen to how first the noise is gone and then step by step the reverb is lower:

Original
-100dB Denoise -0dB Deverb
-100dB Denoise -12dB Deverb
-100dB Denoise -100dB Deverb

2. Breathing sound reduction

In the breathing reduction audio example by LibriVox.org we used the 'Remove Breathing' amounts increasing from the original audio with 0 dB (Off) to 12 dB (medium) and to 100 dB (full) reduction.
In the result files you can hear, that the 100 dB (full) elimination leads to weird, unnatural-sounding pauses, that can be prevented by just reducing the breathing sounds:

Original
-12dB Debreath
-100dB Debreath

Try it now on auphonic.com!

Feedback

We hope you like our upgraded version of the Noise Reduction Algorithms with new parameters for more control.
If you have more feature requests or feedback for us, please let us know! You can also leave a comment in the feedback section on the status page of your specific production. We're looking forward to hearing from you!

Eliminate Breathing Sounds and Mouth Noises

mpagavino@auphonic.com (Manuel) — Wed, 29 Nov 2023 14:05:53 +0000

No matter how good your technical equipment might be, it is almost impossible to avoid capturing unwanted in-/exhaling sounds and mouth noises during voice recordings. After some users asked for an automatic removal of such sounds to improve the audio quality, we got to work and are now proud to present a major upgrade to our Denoiser, including the automatic removal of mouth noises and a new “Remove Breathings” option!
Check out our Audio Examples and the Getting Started Guide below.

What is new?

Remove Breathings: When the new “Remove Breathings” option is enabled, all the inhalation and exhalation sounds will be muted like all the other noises.
Eliminate Mouth Noises and Eating Sounds: With our upgrade to the Denoiser, we have integrated the elimination of all types of mouth noises, such as smacking, chewing and clicking. These sounds are set to silence in your output file.

Breath sounds are traditionally either manually silenced or removed via level-based gating algorithms using a pre-defined threshold to discriminate breaths from speech. However, depending on the speaker and recording setup, levels of speech and breath sounds may be very similar, making accurate breath removal difficult. If the threshold is set too low, some unwanted breaths may pass. Conversely, if the threshold is set too high, quiet speech may be unintentionally removed. In contrast to these traditional approaches, our new AI-based “Remove Breathings” algorithm is trained on a huge collection of breath sounds to automatically discriminate breaths from speech and to reliably remove the former while keeping the latter intact.

NOTE:
By default, mouth noises and breathing segments are muted but not cut. If you want to cut out all silent segments, just add the “Cut Silence” option for “Automatic Cutting” in the audio algorithms.

Audio Examples

1. Example: Breathing Removal

In our first audio example (Freak Show 136), a speaker breathes directly into the microphone, creating a Darth Vader-like effect. This heavy breathing is removed using the “Remove Breathings” algorithm and the Dynamic Denoiser:

Original:
Denoised:

2. Example: Mouth Noise and Eating Sounds Elimination

The following example is an excerpt from the Joe Rogan Experience #2054 where Elon Musk and Joe Rogan are eating pizza and talking about how terrible it is for the sound. Listen to how the chewing and smacking sounds are removed after processing with the new Dynamic Denoiser:

Original:
Denoised:

Getting Started Guide

For the Auphonic Web Service:

To remove breathing sounds in the Auphonic Web Service, simply create a production or preset as usual and check the “Remove Breathings” checkbox for “Noise Reduction” in the “Audio Algorithms” section.
“Remove Breathings” is only available for Speech Isolation and the Dynamic Denoiser, but NOT for the Static Denoiser.

For the removal of mouth noises, no special settings are required. Just enable the Speech Isolation or the Dynamic Denoise method for Noise Reduction.

For API Use:

To use “Remove Breathings” with the API, set the "denoisemethod" to dynamic or speech_isolation and the "debreath" flag to true, as in these two examples:

"algorithms": {
    "denoisemethod": dynamic, "denoiseamount": 100, "debreathamount": 100,
}

"algorithms": {
    "denoisemethod": speech_isolation, "denoiseamount": 12, "debreathamount": 100,
}

For more information about the API, please visit our API help page, where you will also find all other API Audio Algorithm Settings.

Conclusion

Breath and mouth noise removal is essential for podcasters, as it enhances audio clarity and ensures a more professional, polished presentation. By removing distracting sounds, you can maintain listener engagement and provide a seamless listening experience.

Anyway, we hope you enjoy the new features!
If you have any feedback for us on how the breath and mouth noise elimination is working for you, please feel free to contact us via email or directly comment in our production interface!

Export Cuts for Audio/Video Editors and Filler Cutting Updates

lukasm@auphonic.com (Lukas) — Wed, 15 Nov 2023 14:09:43 +0000

For those of you who like to be in control of every applied cut, we are introducing an update for our Automatic Silence and Filler Word Cutting Algorithms today: The export of Cut Lists allows you to import cuts into your favorite audio/video editor to check and apply the cuts to your files manually.
Thanks to your great feedback, we were able to update our “Filler Word Cutting” algorithm as well.

Cut Lists Export

We now provide the export of various formats of “Cut Lists” in the Auphonic Web Service.
You can use these formats to modify and apply cuts in your post production audio or video editing program:

Screenshot of the Reaper DAW with imported “ReaperRegions.csv”: pale red regions display filler words and pale gray regions mark silence.

To generate the cut lists, choose “Cut List (Filler and Silence)” in the “Output Files” section. By selecting the correct “Ending” you can export the cut lists to a format that is suitable for your post production audio or video editor:

We support regions and EDL (edit decision list) formats:

Region formats (“AudacityRegions.txt”, “AuditionRegions.csv” and “ReaperRegions.csv”) list filler and silence cut regions with their respective start and end timestamps.
You can import those cut region lists into your existing Audacity, Audition, or Reaper projects. There you can edit the cut positions to your liking by deleting, moving, or adding individual cut positions. After that, you may apply the cuts manually.
EDL (edit decision list) formats, “SamplitudeReaper.edl”, “FinalCutProLegacy.xml”, “FinalCutProX.xml” and “DaVinciResolve.edl” (also known as “cmx3600” format and compatible with various editors), can be imported in your software to automatically apply cuts from the EDL file. Depending on the software you use, you may also edit individual cuts manually before applying them. Be aware though that in some editors it may not be possible to add them to an existing project. If this is the case, simply open a new project with the EDL file instead.

Please just let us know if we do not yet provide a cut list format for your favorite audio or video editing program!

Export Uncut Audio

To only detect fillers/silence but NOT cut them, choose the “Export Uncut Audio” mode for “Automatic Cutting” in “Audio Algorithms”. Then you can query cutting times via our API or select your favorite Cut List format as “Output File” to only export cut times for later editing, but without applying cuts during the Auphonic audio post production.

Using the option “Export Uncut Audio”, it is also possible to use our cutting algorithms for video productions. In this case, we will just detect and report cuts, and you can import and apply them in your favorite video editing program afterwards.

Filler Word Cutting Updates

Thanks to your great feedback, we've been able to further optimize our Automatic Filler Word Cutting algorithm.
By now you probably know that our filler word cutter removes all types of “uhm”, “uh”, “mh”, “ähm”, “euh”, “eh” and similar sounds. If you haven't heard about it yet, our release blog post might be a good starting point.

Enhanced Fade Curves
We noticed that due to reverb or background noise in input files, some filler cuts did not sound as smooth as they should have. To prevent such rough cuts, we trained our filler detection models with more diverse speech data and compared different fade curve approaches. Now the filler detection is much more robust against noise. Additionally, we developed an asymmetrical fade solution, which, in our opinion, makes the cuts almost inaudible.

We nonetheless recommend, though, that you use filler cutting in combination with our Dynamic Denoising or Speech Isolation algorithms. These algorithms remove heavy reverb or noise and thus prevent noise artifacts during cutting.
Handling of Affirmative Sounds
Some of you were bothered by the fact that affirmative sounds, like “mhm”, “aha”, and “yeah” were cut along with filler words. Driven by your feedback, we tweaked our filler detection algorithm in a way that will remove considerably fewer of these sounds.
Multitrack Filler Silencing
We also received requests to optimize filler cutting for crosstalk in Multitrack productions.
Let's assume that two speech tracks are simultaneously active and one of the two tracks contains a filler word. In this case, we can not cut the filler word in one track without destroying parts of the speech in the other track.
Our original solution was to skip such a filler. Now, however, we replace it with silence instead. This removes the filler words while preserving all the speech content.

Feedback

If you want to provide us feedback on how you like the updated filler cutting algorithm or about any other features that are missing, you are very welcome to contact us. You can also leave a comment in the feedback section on the status page of your concrete production.

Improve your Audio with our new Automatic Filler Word Cutter

isabell@auphonic.com (Isabell) — Wed, 04 Oct 2023 08:44:33 +0000

We all know the problem: the content is perfectly prepared, and everything is in place, but the moment you hit the record button, your brain freezes, and what pops out of your mouth is a rain of “ums”, “uhs”, and “mhs” that no listener would enjoy.
Cleaning up a record like that by manually cutting out every single filler word is a painstaking task.

So we heard your requests to automate the filler word removing task, started implementing it, and are now very happy to release our new Automatic Filler Cutter feature. See our Audio Examples and Usage Instructions below.

What is removed?

While the definition of filler words is not the same, depending on who you ask, some words can be used as filler as well as content. For example, “like”, “well”, “you know”, etc. cannot be removed without the risk of removing also content and destroying sentences, even if those words are used as filler words in some cases.

Therefore, we decided to focus on the removal of the obvious fillers, namely any kind of “ums”, “uhs”, “mhs”, German “ähm”, “äh”, “öh”, French “euh”, “euhm” and similar.

Audio Examples

1. English Male Speaker

The first audio example is an excerpt from the interview “From Racing Failure to Red Bull Champion: The Untold Christian Horner Story”. Our algorithm found and removed a remarkable ten filler words in this 45-second snippet:

Screenshot of the Auphonic Audio Inspector: each pale red shaded area corresponds to a cut-out filler word.

Original:
Cut:

2. Austrian-German Female Speaker

The following example is an interview with the Austrian Ex-Foreign Minister, Karin Kneissl, who uses seven filler words within 26 seconds:

Original:
Cut:

Usage Instructions

To use the Auphonic Automatic Filler Cutter feature, you just have to create a production or preset as you are used to and select “Cut Fillers” for “Automatic Cutting” in the section “Audio Algorithms”:

When your production is done, all cut-out filler words will appear as pale red shaded areas in the Auphonic Audio Inspector on the production status page, as you can see in the upper screenshot of the Audio Inspector.

If you want to remove silent segments from your audio as well, please also enable our Automatic Silence Cutting feature.

NOTE: Our Automatic Cutting features (for filler and silence) are not available for video files!

Behind the Scenes

For the training of our Automatic Filler Cutter AI-Algorithm, we created datasets that contain manually labeled audio files, collected from 'real world' audio data. So far, we have labeled, trained, and tested the system with English, German, Spanish, and French data.

However, in the Auphonic Web Service, you can activate and test the Automatic Filler Word Cutter for all languages. We would be very happy to hear how the filler removal works out for completely different-sounding languages from, e.g., the Asian, African, or Slavic language families.

Please send us feedback on any problems or error patterns you discover! This will help us generate specific data for the training to improve the algorithm and eliminate your problems.

Conclusion

Automatic filler word cutting is a powerful tool for podcasters looking to enhance the quality of their content. It boosts clarity and professionalism, all while making your editing process more efficient. Some users, however, see a touch of authenticity in filler words within podcasts. So, we leave it up to you to enable or disable the Automatic Filler Cutter feature for your next Auphonic production, depending on your desired style.

We are currently working on filler word cutting optimizations for more languages, so watch our channels to get all the news on our upgrades!

If you have any feedback for us – how the filler cutter is working in your language, what you do or don't like, what you miss, what else you would want to remove from your audio besides silence and filler words, etc. – you are welcome to contact us via email or directly comment on our production interface!

Automatic Silence Cutting

lukasm@auphonic.com (Lukas) — Tue, 09 May 2023 10:13:21 +0000

Cutting audio can be a rather tedious task. It requires a decent amount of time and is quite repetitive. Often silence segments, like speech breaks or breathing pauses, make cutting audio necessary in the first place.

Photo by Daniel Schludi on Unsplash

We introduce our new automatic silence cutting feature! It will make your life easier by saving you the time you would normally require to cut silence segments in your recordings.

Why do you need to cut silence segments?

Silence segments occur in your audio recordings naturally. They can be due to short speech breaks or breathing pauses. It's also possible, that at the beginning of a recording equipment needs to be re-adjusted, which also may result in a few seconds of silence.

Usually, listeners do not want to hear silence segments. The reason is easy: silence is redundant filler content. Hence, silence segments need to be cut to achieve a high-quality listening experience. Locating the segments and cutting them manually is tedious though and takes a decent amount of time for longer recordings.

This is why we developed and released our own automatic silence cutting feature.

How do we cut silence for you?

Our automatic silence cutting algorithm detects and cuts silence segments reliably. All you need to do is to enable the algorithm in your production - no further settings are required! This is the easiest possible way for you to cut silence in an audio file.

How to enable silence cutting for your production.

In our audio processing system, silence cutting is a multi-stage algorithm. We use our sophisticated voice activity detection algorithms to detect what is useful content and what is silence. Then we cut the silence parts and crossfade the remaining audio segments to make sure no audio artifacts are introduced. When cutting we ensure that intended speech breaks, e.g. between two sentences, remain untouched.

When we cut any audio, we make sure that chapter marks and speech recognition transcripts are adapted accordingly. Additionally, when exporting to other external services we make sure the cut tracks are exported and correctly labeled.

Audio Examples

Let's listen to two audio examples.

Example 1 (Singletrack production)

Here we have a singletrack production with three silence segments. The segments are located at the beginning of recording, between the first and the second "Hi" as well as at the end. Each silence segment is cut down to a length which still sounds natural, but does not annoy the listener.

The breathing onsets, which happen straight before voice kicks in again, are not cut. This makes sure that the character of the recording remains natural.

For this example, we also activated our new AutoEQ feature to remove pops in the audio.

Original:
Cut:

Here is also a screenshot of how the result looks in our audio player. The gray areas show where our audio processing system cut the silence.

The result of example 01, as displayed by our Audio Inspector.

Thx to the Feel Free to Deviate podcast for providing this recording.

Example 2 (Multitrack production)

Let's also look at silence cutting in a multitrack production. This small excerpt of the TVEye podcast contains a music track and three speaker tracks. Before the intro music kicks in, there are a few seconds of silence.

Un-cut:
Cut:

The few seconds of silence at the beginning of the file are cut. After that, no further cuts are applied though to make sure the background music remains intact.

This is how the result of example 02 looks in our Audio Inspector.

Silence Cutting in our Audio Inspector

As you may have seen already in the examples, the cut segments are displayed as checked grey areas in the Audio Inspector. This is how they will be displayed by default.

Cut segments are displayed as checked gray areas by default.

During playback, the Audio Inspector will automatically skip these cut segments on the master track and play the silence segments if you activate the input track. This way, you can check each segment that was cut.

It's also possible to hide the cut segments. The following picture shows you how to do that.

You can also hide cut segments.

First, you must click onto the "?" icon in the bottom right corner to show the Audio Inspector options. Then you must toggle the "Silence Cut Region" switch. After that, the cut regions are not displayed anymore, and also won't be played back on the input track anymore.

If you click onto "Show Stats", you can activate the audio processing statistics. These statistics show you how much of your audio was cut, as well as the resulting track length after the cuts were applied.

The processing statistics tell you more about how much audio was cut.

Conclusion

With our automatic silence cutting feature, we went one step further towards the perfect audio assistant. Audio editing has been a tedious-but-necessary task for a long time, but with our automatic silence cutting feature, we just made it easier for you!

Feel free to send us your feedback - how do you like our new feature? Also, make sure to follow us, as we will release more automatic cutting algorithms for you in the future!

New Auphonic AutoEQ Filtering (Beta)

grh@auphonic.com (Georg) — Tue, 24 Jan 2023 09:35:03 +0000

In addition to our Leveler, Denoiser, and Adaptive 'Hi-Pass' Filter, we now release the missing equalization feature with the new Auphonic AutoEQ.
The AutoEQ automatically analyzes and optimizes the frequency spectrum of a voice recording, to remove sibilance (De-esser) and to create a clear, warm, and pleasant sound - listen to the audio examples below to get an idea about what it does.

Screenshot of manually adjusted example settings for the equalizer plug-in 'Pro-Q3' by fabfilter.

What is Equalization and why is it difficult?

Equalization (EQ) in audio recording and reproduction is the process of adjusting the volume of different frequency bands within a signal.
The following vocal EQ cheat sheet, published by Producer Hive, will give you a small impression, of what can be influenced by equalizing:

Vocal EQ Cheat Sheet by Producer Hive.

On the other hand, it is very easy to ruin a good voice recording with manual heavy-handed equalization, resulting in voices, that sound very sharp or muddy or even like the speaker had a blocked nose.
Besides the skill and experience of an audio engineer, manual adjustments of frequencies also require a very good and linear studio playback device. For example, performing manual equalization with strongly bass-heavy speakers would most likely lead to a very sharp, unpleasant listening experience using treble-heavy headphones.

For singletrack productions with more than one speaker, equalizing is also a very complex and time-consuming process, as every voice has its unique frequency spectrum and needs its own equalization. One could separate speakers with cuts or create a track envelope to fade from one speaker to another, however, any solution is a very tedious job, if you do it by hand.

That is where the Auphonic AutoEQ comes in! All those steps are now available in just one click!

How does the Auphonic AutoEQ work?

The Auphonic Web Service analyzes your audio content and classifies the audio file into small and meaningful segments like music, silence, different speakers, etc. to process every single segment with the best matching algorithms.
All our features like the Adaptive Leveler, Dynamic Denoising, Adaptive 'Hi-Pass' Filtering, and now the new AutoEQ filter option are built on top of this basic processing.

Using Auphonic AutoEQ, spectral EQ profiles are created for each speaker separately and permanently changing over time. The aim of those time-dependent EQ profiles is to create a constant, pleasant sound in the output file even if there are slightly changing voices in the record, for example, due to modified speaker-microphone positions.

Audio Examples

Here are two short audio examples, which demonstrate some features of our AutoEQ.
We recommend listening with headphones so you can hear all the details.

Example 1. Female Speaker with Background Music

In the following example (BCB: The Voices of Bainbridge Island) of a female narrator speaking while background music is playing, you can easily recognize quite sharp 'sss' sounds in the female voice. This sharpness in the female voice is removed by the so-called De-essing feature of the Auphonic AutoEQ, while the background music is not changed.

Original:
AutoEQed:

Example 2. Dialog of Male and Female Speakers

The next example (BCB: The Voices of Bainbridge Island) shows how the AutoEQ optimizes a singletrack record containing two speakers with different voice characteristics. Our AutoEQ algorithms analyze each voice separately and calculate the matching frequency adjustments to optimize the voice of every single speaker.

Original:
AutoEQed:

AutoEQ Beta Integration in the Auphonic Web Service

To use the Auphonic AutoEQ, you just have to create a production or preset as you are used to, toggle “Advanced Parameters” on the top right in the section “Audio Algorithms” and select “Voice AutoEQ” within “Filtering”:

For a first test period, the AutoEQ will only be available for Beta and paying users, to incorporate your feedback and finalize an optimized version.
If you are a free user but want to try Auphonic AutoEQ: please just ask for access!

Practical Tips

For best results using Auphonic AutoEQ, however, it is still necessary, that your audio content is of sufficiently good quality, as no equalizer can make up frequencies that are not there in the first place. Audio files with low bitrates often lack important frequencies, that cannot be recovered by equalizing. AutoEQ is just a feature to boost or cut individual frequency bands, not a bandwidth extension. For more information about required audio quality, see the former blog post: Audio File Formats and Bitrates for Podcasts.

Another important topic is the definition of the 'best result'. Equalizing is a very subjective task, that differs a lot depending on every personal opinion. So Auphonic AutoEQ is set up to follow quite conservative rules of equalizing and rather apply subtle tweaks and remove obvious problems, than support personal preferences. This also means, your record will experience no significant changes from Auphonic AutoEQ if it sounds reasonably OK or pretty good already.

Conclusion

Auphonic audio post production algorithms keep getting better and better in leaps and bounds lately, offering you new Beta Features: Beta Auphonic Denoiser, Beta Auphonic Speech Recognition, and Beta Auphonic AutoEQ.
Right now we are fine-tuning all our current Beta Features with high intensity to release a new upgraded version of our Auphonic Web Service as soon as possible.
Please watch this channel for further updates – soon to come.

If you have any feedback for us or want to become a Beta user, you are very welcome to comment directly in our production interface or to contact us via email!

New Noise Reduction Algorithms (Beta)

grh@auphonic.com (Georg) — Fri, 10 Jun 2022 08:06:02 +0000

Our classic noise reduction algorithms remove broadband background noise and hum in audio files with slowly varying backgrounds.
We released the first beta version of the new dynamic noise reduction algorithms now, which work much better with fast-changing and complex noises. Listen to the audio examples below, they demonstrate some of the new features and use cases!

Glitch While Streaming by Michael Dziedzic.

How to try out the Beta Denoiser

At the moment, only users with access to our advanced algorithm parameters can try the beta noise reduction algorithms (free users: please just ask for access).

To try them out, create a (singletrack) production in our web interface, toggle the Advanced Algorithm Parameters and change the Denoising Method to one of these options:

Dynamic:: This keeps speech and music signals, but removes everything else in your audio.
Speech Isolation:: Only keeps speech and removes everything else, including music, from your audio.

Warning:
These algorithms might delete signals from your audio, like certain sound effects, because they are classified as noise!
We will provide other settings in the future to specify exactly which signals should be kept or removed - in the meantime, please double-check the results in case you process very complex productions.

Real-World Audio Examples and Use Cases

Here are real-world audio examples from our users to demonstrate some features of the new beta denoising algorithms. More real-world examples will be available after the beta phase.
We recommend listening with headphones, so that you can hear all the details!

1. Extreme Breathing, like Darth Vader

In this example (Freak Show 136), one speaker is breathing directly into the microphone, which creates a Darth Vader-like effect:

Noisy:
Denoised:

2. Recording in a Van, Noise Reduction Amount

This is a recording in a van, with changing engine and wind noises in the background:

Noisy:
Denoised:

You can also use the parameter Noise Reduction Amount, to not remove all the ambiance/noise but just improve the speech intelligibility:

-3dB Denoise:
-9dB Denoise:
-15dB Denoise:

3. Keep Music, Static Background Noise

Our Dynamic denoising method keeps speech and music signals, but removes everything else in your audio (Speech Isolation would remove music as well).
In this example (HeadRightOut Podcast), the intro music and fade out are preserved but static background noises (>30s in the audio sample) are removed:

Noisy:
Denoised:

Noise Reduction Test Files

As additional examples, we added some short files from test datasets, to demonstrate further features and use cases of the new denoiser.

1. Non-Stationary Noises

Male speaker on the left (stereo audio), non-stationary noises in the center:

Noisy:

Denoised:

Male speaker, computer game shooting in the background:

Noisy:

Denoised:

Female speaker on the left (stereo audio), drilling machine in the background:

Noisy:

Denoised:

Female speaker, animals (sheep, birds) in the background:

Noisy:

Denoised:

2. Stationary Noises

Female speaker on the left (stereo audio), broadband noise:

Noisy:

Denoised:

Male speaker on the left (stereo audio), hum in the background:

Noisy:

Denoised:

Male speaker, babble noise in the background:

Noisy:

Denoised:

Female speaker, low-frequency hum background:

Noisy:

Denoised:

3. Periodic / Music-like Noises

Male speaker with background music (denoising with Speech Isolation method):

Noisy:

Denoised:

Female speaker and periodic beeps:

Noisy:

Denoised:

Feedback

Please try out our new denoising algorithms and send us some feedback (directly in the production interface or via email) if something works well or if you find any problems.
We are currently tuning the algorithms to more practical use cases. For that any feedback helps a lot!

Stay tuned for further updates and more new audio restoration algorithms :)