How to Extract and Clean Voice Audio from an MP4 File

You have an MP4 with a voice track you need to reuse—maybe a podcast recorded in Zoom, a phone video with a ceiling fan in the background, a gaming clip where your mic is buried under keyboard clicks, or a camera take with wind noise. The workflow is almost always the same: extract the audio cleanly first, then clean the voice as a separate step. Do not try to fix everything while the audio is still inside the compressed MP4.

Step 1: Extract the audio track from the MP4

An MP4 is a container. The video might be H.264 and the audio AAC at 44.1 or 48 kHz. Your first job is to get that audio into an editing format without generation loss. Use WAV, not MP3.

Option A: ffmpeg for exact control

If you are comfortable with a terminal, this is the cleanest method. It copies the original audio into a WAV file and avoids the video entirely:

ffmpeg -i input.mp4 -vn -acodec pcm_s16le -ar 48000 -ac 2 audio.wav

Use -ac 1 instead of -ac 2 if the source is a single speaker and you want a mono editing file. Keep -ar 48000 when the source is 48 kHz. If the MP4 has several audio tracks, list them first with ffprobe input.mp4, then choose one with -map 0:a:1—for example, when a Zoom cloud recording has separate mixed and isolated tracks.

Option B: VLC or Audacity

In VLC, choose Media > Convert/Save, add the MP4, click Convert/Save, and pick an Audio - CD or WAV profile. Set the sample rate to 48 kHz and channels to stereo or mono. In Audacity, install FFmpeg if needed, open the MP4, select the audio track, and use File > Export > Export Selected Audio as WAV. Both are fine for a quick one-off. VLC is faster; Audacity is better if you plan to edit immediately.

Tradeoff: a 10-minute stereo WAV at 48 kHz/16-bit is roughly 115 MB. That is normal. Use 24-bit if your editor supports it, but 16-bit is enough for most voice cleanup.

Step 2: Make a clean working copy

Save the extracted WAV as your untouched master. Create a second copy for processing. If the recording is mono, work in mono. If it is stereo with one voice centered, you can process in stereo but check for phase issues after noise reduction. Set your editor to 48 kHz and keep peaks around -6 to -3 dB before cleanup. Avoid normalizing to 0 dB first; noise reduction works better with a little headroom.

Step 3: Clean the voice in the right order

Order matters. Fix obvious problems before broad processing, and use the least amount of treatment that solves the issue.

1. Trim and repair. Cut dead air, remove coughs and clicks, and repair obvious pops. Do not gate yet.
2. High-pass filter. Set 80–100 Hz for most male voices and 100–120 Hz for most female voices. This removes rumble, HVAC thumps, and handling noise without thinning the voice.
3. Noise reduction. Select one second of room tone—no speech—and capture a noise profile. Start with 6–8 dB reduction and 6.0 sensitivity. If the result sounds watery or robotic, back off to 4–6 dB. Never go to 20 dB just because the slider allows it.
4. De-ess and de-click. Use a de-esser around 5–8 kHz only if sibilance is sharp. For keyboard clicks, use spectral repair or a very short manual fade rather than a heavy gate.
5. EQ. Cut 200–400 Hz by 1–2 dB if the voice sounds muddy. Add 1–2 dB around 2–5 kHz for intelligibility. Avoid boosting above 8 kHz unless the mic was dull.
6. Compression. Use a 2:1 to 3:1 ratio, threshold around -18 to -12 dB, attack 5–10 ms, and release 80–150 ms. Aim for 3–6 dB of gain reduction on loud phrases, not constant squashing.
7. Limiter. Set the output ceiling to -1 dBTP. This prevents clipping when you export.

If you are working from a phone recording, camera clip, or voice memo and do not want to install a full editor, a free online option like ClearVoiceHQ (clearvoicehq.com) is built for exactly those sources: it handles voice cleaning for mobile recordings, cameras, and voice memos. Upload the extracted WAV or a short excerpt, compare the cleaned version against the original, and keep the one that sounds natural.

Step 4: Export and put the audio back

Export the cleaned voice as WAV for archiving, then choose AAC if you need to put it back into an MP4. For spoken voice, 192 kbps AAC is plenty; 256 kbps is safer if the file will be re-edited. To replace the original audio while copying the video stream:

ffmpeg -i video.mp4 -i cleaned.wav -map 0:v:0 -map 1:a:0 -c:v copy -c:a aac -b:a 192k output.mp4

Check sync at the start, middle, and end. If the audio drifts, confirm both files use the same sample rate and that you did not cut the WAV without trimming the video to match. For social clips, export a separate WAV and let the editor mux it rather than re-encoding the video twice.

Source-specific settings that save time

• Phone recording or voice memo: mono, 100 Hz high-pass, 6–8 dB noise reduction, light de-ess, 2:1 compression. Phone mics often boost upper mids, so cut 3–5 kHz by 1 dB if it sounds harsh.
• Zoom call: the audio is already compressed. Use 4–6 dB noise reduction and skip aggressive EQ. Fix plosives manually; a gate will make the compression pump.
• Camera clip: handle wind with a 120 Hz high-pass and a gentle low-cut, not a huge EQ scoop. If wind is intermittent, edit those sections by hand.
• Gaming clip: keyboard clicks sit above the voice. Spectral repair is better than gating. If game audio is on the same track, extraction cannot separate it—use a version with separate mic and game tracks if available.

ClearVoice is also useful as a first pass when you need a quick cleaned file for review before doing detailed edits in your DAW. The key is to compare A/B and stop before the voice sounds processed.

Common mistakes to avoid

• Cleaning the AAC track in place and re-encoding it multiple times.
• Using 20 dB noise reduction to remove a fan; it usually creates metallic artifacts.
• Normalizing before noise reduction or compression.
• Gating every pause so the room tone disappears and the edits sound unnatural.
• Forgetting to keep the original MP4 and the untouched WAV.

Extract to WAV, clean in a sensible order, export at -1 dBTP, and remux only once. That gets you a natural-sounding voice track from almost any MP4 without wrecking the original.