Choose Auphonic when a spoken recording needs automated level balancing and a repeatable preparation workflow. Choose Audacity when you want to inspect a specific noise and control what is removed. For transcription, the winning result is the version that preserves the words, not simply the version with the quietest background.
Describe the fault before choosing an effect
Listen to a passage without changing anything. Is one person much quieter than another? Does a steady hum continue during pauses? Does a chair scrape cover a particular answer? Write down the problem in ordinary language. These are different faults, and a single instruction to clean up the audio hides the distinction.
Also identify what must survive. A podcast editor might remove a long hesitation to improve pacing. A researcher studying that hesitation may need it intact. Establish whether you are producing a listening copy or preserving evidence of the original conversation. Keep an untouched source in either case.
A useful processing note might say: reduce the constant background hum while retaining the quiet final word of each answer. That gives you something audible to check. Make the criterion specific enough that you can reject a polished result if it removes information you actually need.
Auphonic automates several separate jobs
Auphonic's feature page documents a leveler for differences between speakers, adjustable noise and reverberation reduction, and optional cutting of pauses or filler words. Those are separate decisions. Start with the feature that matches the fault you identified. Do not enable removal of pauses merely because you also want more even speech levels.
The pricing page provides two free processing hours per month with an Auphonic jingle. Speech recognition is excluded from Free. Processing has a three-minute minimum charge; rerunning the same production with unchanged input files does not cost additional credits. These conditions make the free allowance useful for checking audio treatment, but not a free transcription allowance.
The page did not expose a reliable paid euro or dollar amount in the retrieved price table, so no paid subscription price is quoted here. Distinguish a confirmed allowance from an unverified checkout total. The useful first decision is whether the output and workflow fit, before deciding how many paid processing hours you need.
For repeated work, write down the settings that produced an acceptable copy. Automation is valuable when it reduces decisions you have already made well. It is less helpful if you cannot explain why material disappeared. Review a quiet passage and a louder passage before applying the same treatment to a whole collection.
Audacity lets you examine what gets removed
Audacity's own site describes a free, open-source desktop editor for Windows, macOS and Linux, with formats including WAV, MP3, FLAC and Ogg. This makes it a practical option when you want to prepare a file yourself rather than build your workflow around processing credits.
Its Noise Reduction manual targets steady noise, requires a noise-only profile and warns against treating irregular traffic or audience sounds with this effect. Preview lets you hear a trial; Residue lets you hear what would be removed. Recognisable speech in the residue is a reason to reconsider the settings.
That last check changes how you judge the effect. A quiet output alone can conceal an overaggressive choice. Listening to the rejected material gives you a second question: did the process classify part of the wanted voice as noise? If so, preserving a little background may be preferable to weakening a name or sentence ending.
Do not select a noise profile that contains somebody speaking. Find a genuine pause with the unwanted background present. If your sample is not representative of the problem, stop and choose another passage rather than repeatedly increasing the controls. There is no single setting in this article that can responsibly be applied to every room and microphone.
Match a real problem to a limited intervention
| Problem you can hear | First route to consider | What to inspect afterward |
|---|---|---|
| Large differences between speakers | Auphonic's level balancing | Quiet answers as well as loud ones |
| Steady background hum or hiss | Audacity's profile-based reduction | Wanted speech in the removed material |
| A wish to shorten pauses | A separate editorial decision | Whether timing is evidence you must preserve |
| A word masked by another sound | Revisit the original first | Whether the word is actually recoverable by listening |
| No meaningful intelligibility problem | Keep the original workflow | Whether processing adds any useful benefit |
Consider a hypothetical remote interview in which the interviewer is consistently louder than the guest. Begin by evaluating level balancing. If the guest's final syllables remain clear and the conversation becomes easier to follow, the copy may be useful for review. That is a proposed listening criterion, not a claim about a measured Auphonic result.
Now imagine a permitted lecture recording with a constant fan audible between sentences. Audacity's documented noise-profile method is a relevant experiment. Select a sample containing the fan without speech, then judge both the output and what the effect would remove. A sudden cough covering a word is a different case; do not claim that this method reconstructs the missing word.
A ten-minute comparison that can reject the edit
For the first minute, choose a short passage containing clear speech and the actual fault. In the next minute, make a working copy and record the original filename. Spend two minutes deciding on one treatment and preparing it. Leave optional cutting disabled when the original timing matters.
Use the next two minutes to listen to the original and candidate at comparable playback volume. Alternate at the same sentence rather than relying on a memory of how the whole file sounded. Write down a name, a number and a quiet ending that you can hear reliably in the source. Check whether each remains intelligible.
Spend two minutes examining the least convincing part. With Audacity, include the residue check described above. With Auphonic, review the chosen feature settings before rerunning the same production. Processing wait time sits outside this listening exercise; ten minutes is not a promised service turnaround.
Use the final two minutes to choose the original, the processed copy or a more conservative retry. Record why. A valid result can be that processing made the file worse. If you later compare transcripts, hold the transcription language and other available settings constant so you are not changing two variables together.
Notibo belongs after the audio decision
For audio repair itself, choose the appropriate Auphonic or Audacity workflow. If the original is already understandable, you can skip repair. Notibo is relevant when the allowed recording should become a transcript with structured notes, a summary and key points. It is not an audio editor.
Notibo accepts audio or video uploads up to 50 MB and processes in the cloud, with EU storage and US processors. Choose one main language. Its transcript has no timestamps, so keep your original recording and any separate review references if exact locations matter to your work.
Use the source to check important statements even when the notes read smoothly. A transcript should not turn a guess about an obscured word into an established fact. The purpose of preparation is to make review easier while preserving information, not to make uncertainty disappear from view.
Related comparisons
- Dolby On vs Voice Memos: cleaner audio or simpler capture?
- WAV vs MP3 for transcription: quality, size and backups
This is a documentation-based comparison by Notibo, checked on 20 October 2026, not a measured accuracy benchmark. The examples are suggested trials, not reported test results.