Why Zoom MP4 audio is tricky
A Zoom cloud or local recording is usually an MP4 with one mixed audio track. That means the host, guests, system sounds, and room noise have already been blended. Cleaning the audio can reduce hiss, hum, fan noise, keyboard clatter, and room tone, but it cannot fully separate two people talking over each other or undo clipping. Treat the MP4 as your source, extract the audio, clean a copy, and reattach it. Keep the original untouched until you are happy.
Step 1: Inspect the MP4 before you touch it
Use VLC or ffprobe to see what streams are inside. This tells you whether you have one audio track or several and whether the file is stereo or mono.
ffprobe -v error -show_entries stream=index,codec_type,codec_name,channels,sample_rate -of default=noprint_wrappers=1 input.mp4
Most Zoom MP4s show one AAC audio stream, often stereo at 48 kHz. If you see multiple audio tracks, pick the one with the actual meeting audio. If the file is mono or duplicated stereo, work in mono for speech; it simplifies denoising and avoids weird phase issues. Keep stereo only if there is music, panning, or sound effects you need to preserve.
Step 2: Extract the audio to WAV
Do not keep re-encoding AAC while you clean. Extract once to a lossless WAV, edit that, then encode back to AAC when you rebuild the MP4.
ffmpeg -i input.mp4 -vn -acodec pcm_s16le -ar 48000 -ac 2 audio.wav
For a single voice track, change -ac 2 to -ac 1. For more headroom, use pcm_s24le if your editor supports 24-bit. The tradeoff is file size; a one-hour 48 kHz stereo WAV is large, but it is temporary. If you do not use command-line tools, any editor that can import MP4 and export WAV will do. In a pinch, VLC can convert, but a real editor gives you better control over the audio stream.
Step 3: Do basic repair before noise removal
Denoisers work better on a clean, well-leveled file. Start with these steps in order:
- Trim dead air at the start and end. Keep a half-second of noise-only room tone if your denoiser needs a noise profile.
- High-pass filter at 80 to 100 Hz. This removes rumble, desk thumps, and HVAC low end. Do not go higher unless the voice sounds thin; 120 Hz is usually too much for male voices.
- Fix obvious hum with a notch filter. Use 50 Hz in Europe and many other regions, 60 Hz in North America. If you hear a buzz, notch the fundamental and maybe the first harmonic.
- Do not normalize or compress hard yet. Leave headroom. A peak around -6 dBFS before denoising is fine.
If the audio is clipped, noise removal will not fix it. You can lower the level, use a de-clipper sparingly, and accept that distorted consonants stay distorted. For phone recordings and gaming clips, clipping is common; pull down the loudest moments before any enhancer.
Step 4: Apply noise removal conservatively
Set the denoiser to reduce steady noise by about 6 to 10 dB. Going past 12 dB often creates metallic, underwater, or warbling artifacts, especially on Zoom audio where the codec has already removed detail. If the tool has a noise profile, sample 0.5 to 1 second of noise-only audio from the beginning or a pause. If there is no clean noise section, use a lower setting and rely on a high-pass filter plus a gentle low-pass filter around 12 to 15 kHz to tame hiss.
Always A/B against the original. Listen for tail-off artifacts after words, hollow room tone, and softened sibilance. A little remaining room tone sounds more natural than a completely silent background. For interviews and webinars, preserve breath sounds unless they are distracting; aggressive gating makes edits sound choppy.
Step 5: Recombine the cleaned audio with the video
Once the WAV sounds good, put it back into an MP4 without re-encoding the video:
ffmpeg -i input.mp4 -i cleaned.wav -c:v copy -map 0:v:0 -map 1:a:0 -c:a aac -b:a 192k -ar 48000 output.mp4
Use 192 kbps AAC for stereo or 128 kbps for mono speech. If the final destination is a podcast, export a WAV or high-quality MP3 from the cleaned file instead. If the video is for YouTube or a learning platform, check that the audio stays in sync. Reattaching a WAV should not change timing; if it does, confirm the sample rate matches and that you did not trim the audio separately unless you also trimmed the video.
Zoom-specific problems and realistic fixes
- Keyboard clicks and mouse clicks: these are short transients. Spectral repair or a light de-clicker can help, but heavy processing makes speech lisp. Fix the worst clicks manually and leave the rest.
- Overlapping speakers: one mixed track cannot be fully separated. Noise removal will not unmix voices. Use the original video and edit around crosstalk, or re-record if possible.
- Room reverb: de-reverb tools can reduce it, but they also thin the voice. Use less than you think you need.
- Internet dropouts and robotic audio: these are codec damage, not noise. Denoising can make them worse. Cut around the glitch or accept it.
- Music or system audio: skip the high-pass filter and use a lighter denoise setting. Voice presets can mangle music.
A practical starting preset for Zoom voice
If you need a conservative baseline, try this: high-pass at 90 Hz, de-hum at 60 or 50 Hz if needed, denoise at 8 dB, low-pass at 14 kHz only if hiss is obvious, then normalize to about -16 LUFS for spoken-word video. Add a limiter at -1 dBTP for safety. Listen on laptop speakers and headphones; laptop speakers hide hiss, while headphones reveal artifacts. Adjust one setting at a time and bypass the chain often.
If you do not want to install a DAW, a free online option like ClearVoiceHQ (clearvoicehq.com) can handle the denoise step on an extracted WAV or MP3. Upload the file, run the cleanup, listen for artifacts, and download. ClearVoice is best for steady hiss, hum, fan noise, and room tone; it will not rescue clipped audio or separate two people talking over each other. Use it as one step in the workflow, not as a replacement for careful editing.
The reliable routine is simple: extract lossless audio, fix rumble and hum, denoise gently, check the result, and reattach the cleaned track to the original video. That gets you cleaner Zoom MP4 audio without making the speaker sound like a robot.