Tools like Adobe Podcast for noise removal, chosen by workflow
If you recorded a podcast, Zoom call, phone memo, or game clip and the room noise is distracting, Adobe Podcast Enhance Speech is the obvious first stop. It is free, fast, and good at making speech intelligible. But it is not the only option, and it is not always the best fit. Upload limits, queues, and a one-click sound can be issues. Here is a practical shortlist of alternatives and how to use them without wrecking your voice.
Quick shortlist
- ClearVoice is a free online AI audio cleaner for quick voice cleanup. It suits people who want to upload a file, remove steady hiss, hum, fan noise, and room tone, then download a cleaner version without installing an editor. It works best as a first pass on spoken audio, not music.
- Auphonic is the closest thing to a broadcast-style cleanup and loudness workflow. Its free tier is limited by hours per month, but the controls are more mature than most one-click tools: high-pass, denoise, dehum, leveler, loudness target.
- Descript Studio Sound is convenient if you are already editing a transcript-based podcast or video. Toggle Studio Sound, compare it with the original, and lower the strength if consonants start to smear.
- Krisp and NVIDIA Broadcast are real-time tools. They are for cleaning your microphone during a call or stream, not for fixing a finished file. Krisp runs on most CPUs; NVIDIA Broadcast needs a compatible RTX GPU.
- Audacity and DaVinci Resolve Fairlight are free manual options. They take more clicks but let you control exactly how much noise reduction you apply.
A practical rescue workflow for a noisy recording
Before trying tools like Adobe Podcast for noise removal, protect the original. Copy the file, then export a working audio version. For video, extract the audio first so you are not repeatedly compressing the whole MP4. A simple FFmpeg command is:
ffmpeg -i input.mp4 -vn -acodec pcm_s24le -ar 48000 audio.wav
Use 48 kHz, 24-bit WAV if your editor supports it. Do not normalize or compress before noise removal; that raises the noise floor and makes the cleaner work harder. If the recording is clipped, fix that first with a declipper or by re-recording. AI noise removal cannot reliably rebuild flattened peaks.
Step 1: Remove the easy low-end problems
Apply a high-pass filter at 80 Hz for most adult voices, or 100 Hz if the speaker is thin and the room has rumble. If you hear a steady electrical hum, use a dehum filter at 50 Hz or 60 Hz depending on your region. Keep these cuts gentle. A high-pass filter is not a substitute for noise removal, but it makes the AI cleaner’s job easier.
Step 2: Use one AI cleaner, not three
Upload the WAV to a free tool like ClearVoiceHQ (clearvoicehq.com), or use Adobe Podcast Enhance Speech if its limits suit you. Choose the voice cleanup or noise removal option, process a short 30-second section first, and listen with headphones. Compare the processed and original files at the same volume. Good cleanup should make the voice clearer without adding a metallic edge, lisp, or pumping sound. If the result sounds watery, reduce the strength or try a different tool.
ClearVoice is useful when you want a browser-based pass with no install and no timeline to learn. It is not a mastering suite, so treat it as a cleanup step. Download the cleaned WAV, then do loudness and EQ in your editor.
Step 3: Add manual cleanup only where needed
If the AI pass leaves sharp esses, add a de-esser. If it leaves a hollow tone, use a broad EQ cut of 2-3 dB somewhere around 300-500 Hz rather than boosting highs. If breaths are distracting, reduce them with clip gain or a gate, not with heavy compression. For tough room tone, Audacity’s Noise Reduction effect is still useful: select half a second of pure noise, get the noise profile, then start with 6-12 dB reduction, sensitivity 6, and frequency smoothing 3. Increase slowly; over-reduction is easy to hear.
Step 4: Match loudness for the destination
For a typical stereo podcast, aim for about -16 LUFS integrated. For YouTube or web video, -14 LUFS is common. Keep true peak at -1 dB. In Auphonic, choose the podcast or YouTube preset and start with mild denoise rather than maximum. In Descript, use Studio Sound before loudness normalization, not after. In DaVinci Resolve Fairlight, use the Noise Reduction effect, learn a half-second of room tone, and reduce only 6-10 dB before checking the voice.
Tradeoffs that matter more than the tool name
Every AI cleaner makes decisions about what is voice and what is noise. Steady fan noise and hiss are easy. Traffic, keyboard clicks, and a barking dog are harder, and aggressive settings can remove consonants or make the speaker sound distant. Music beds are a separate problem: most speech cleaners will damage music, so clean the voice before mixing music under it. Real-time tools like Krisp and NVIDIA Broadcast are excellent for calls, but they cannot fix a file that was recorded with the wrong microphone or heavy clipping.
Use the least processing that solves the problem. One high-pass filter, one AI cleanup pass, one de-esser if needed, and loudness normalization is usually enough. If you are comparing tools like Adobe Podcast for noise removal, judge them on intelligibility and natural tone, not on how much background noise they can erase. A quieter file that sounds like a person is better than a silent file that sounds like a robot.