The Reply Came From the Wrong Side of the Table

The customer on the right says, “Yes, please.”

Through headphones, the answer comes from the left—the barista’s side. The mouth movement is correct, but the scene still feels confusing.

This is a useful listening step for a Kling 4.0 AI Video Generator project or any dialogue workflow that returns stereo audio. It is not necessarily a lip-sync problem. The picture and the sound may simply be pointing to different people.

Confirm the Audio Format

First, find out whether the exported file has two audio channels:

ffprobe -v error -select_streams a:0 -show_entries stream=codec_name,channels,channel_layout -of default=noprint_wrappers=1 dialogue-draft.mp4

A stereo file may report:

codec_name=aac
channels=2
channel_layout=stereo

Some files show channels=2 without reporting a layout. That does not mean the audio is damaged. Listen to both channels before making a decision.

If the file is mono, there is no left-right placement to diagnose. Focus on which person appears to speak and whether the line arrives at the correct time.

Map the Screen

Write down where each person appears:

LEFT: barista
RIGHT: customer
BACKGROUND: café ambience

Then mark the speaking order:

0.0–2.4s: barista
2.8–4.6s: customer

This map does not mean every voice must be pushed toward one ear. It only identifies who owns each line.

Listen in Stereo

Use headphones and start each line from a moment of silence.

Ask:

• Which person is visibly speaking?
• Is the voice centred or leaning to one side?
• Does its position change during the line?
• Do the background sounds move at the same time?

Note the first moment when the voice and picture disagree. Avoid changing the mix before you know whether the problem affects one line or the entire soundtrack.

Make a Mono Copy

A stereo effect may narrow or disappear on a phone or small speaker.

Create a mono test file:

ffmpeg -i dialogue-draft.mp4 -af "pan=mono|c0=0.5*c0+0.5*c1" -c:v copy dialogue-mono-check.mp4

Compare the two versions:

The mono copy is for diagnosis. It does not have to become the final export.

Do Not Swap Everything for One Bad Line

If the entire soundtrack is reversed, a left-right swap may be reasonable.

If only the customer’s reply is misplaced, swapping the full mix may move footsteps, ambience, and the barista’s voice to the wrong side.

Use a narrow fix:

• Whole mix reversed: inspect a channel swap
• One line misplaced: isolate that line
• Position moves during a line: repair or replace the section
• Dialogue weak in mono: inspect the mix before delivery
• Voices centred and clear: leave them centred

Centred Dialogue Is Often Fine

A person standing on the right does not require a voice placed hard right.

Dialogue is often kept near the centre so it remains clear on phones, laptops, televisions, and headphones. Gentle positioning can support the scene, but clarity matters more than exact screen matching.

The problem is not a centred voice. It is a voice that clearly seems to belong to the other person.

Leave One Useful Note

Save a compact observation with the draft:

3.1s | speaker on screen: right | stereo: left-weighted | mono: clear

This tells the next editor what happened without asking them to diagnose the soundtrack again.

Confirm permission before uploading or imitating a recognizable voice, including internal recordings that were not cleared for reuse. Do not present synthetic dialogue as a real person’s statement.

When the words sound right but the conversation still feels wrong, listen to where the reply comes from.