Locate the Imbalance Before Changing Channels
One silent earcup alone does not establish a damaged video. Test familiar working audio on the same device, then compare the current export on another headset or playback device. If other audio also leans to one side, investigate playback first. If this file alone differs, inspect its inputs and mix. Both faults can coexist.
In an audio editor, inspect and listen to left and right separately. Compare the selected narration input with the same complete sentence in the export. Does every sound lean, or just the voice? Silence, a quieter channel, and different content call for different decisions. A stereo label establishes two channels, not speech in both.
This editorial diagnostic draws on actual channel conversion, input selection, and loudness methods. The skill does not automatically diagnose hardware or fill the right channel. Separate guides cover an entirely silent export and uneven segment loudness.
Choose Between Preserving Stereo and Correcting a Voice Input
| Actual input | Decision |
|---|---|
| Genuine mono voice | The cited full-sound path places it left and right at 1/sqrt(2) each. This is a path rule, not a universal standard. |
| Valid stereo | That path preserves both channels independently, including intentional spatial voices, music, and ambience. |
| Complete voice on the left, empty right; centered delivery intended | Confirm all required speech is on the left, save a new mono voice from that channel in an audio editor, and select the new input. |
| Different speakers or useful content on each side | Assign their story roles before choosing a channel; do not discard another speaker. |
| More than two channels | The cited full-sound v1 path rejects this input and offers no arbitrary mapping. Complete an explicit conversion separately first. |
Changing a label to mono does not create correct samples. Copying the empty channel cannot recover speech. A new file must retain complete required words. If two voice channels cancel each other when combined, a simple sum can weaken the result; inspect them before combining.
Spatial placement can be intentional. The repair addresses an accidentally displaced narrator, without flattening every musical, environmental, and character position in the film.
Complete Paper Example: Voice on the Left, Music on Both Sides
An original eight-second recap says, “She opened the empty cabinet and found last night’s records missing.” The author wants a centered narrator and valid stereo music. The values below represent one illustrative instant, not measured recording samples.
| Step | Input, operation, and result |
|---|---|
| Original voice | Illustrative stereo (left 0.20, right 0.00): an empty right channel, not intentional placement. |
| Direct full-sound mix | Stereo remains independent. With a bed of (0.02, 0.02) and segment and master gains of 1, the illustrative sum is (0.22, 0.02). Music on the right does not establish voice there. |
| Double the voice globally | (0.40, 0.00) still has an empty right channel. RMS normalization also applies a global gain rather than copying a valid channel. |
| Correct the input | Save the complete useful left channel as a new mono file in an audio editor. Preserve the sentence and duration; keep the original. |
| Select it and remix | Mono 0.20 becomes about 0.1414 on each side through this path. Adding the unchanged bed gives about 0.1614 per side. |
| Actual review | Render a new candidate; listen to each side and both together through the complete sentence, then watch the eight seconds at normal speed. Illustrative values establish neither intelligibility nor freedom from distortion. |
Update the actual voice path in the segment metadata and recheck selected words, order, and tempo policy. Confirm that the current mix consumed the new file. Old snapshots, paths, and exports are not a repaired result. The full-sound path will not cut a sentence ending or automatically shrink a placement that does not fit.
The paper example establishes only that gain cannot fill silence and how selected mono is distributed on this path. No example audio was produced or heard; headset behavior, voice quality, and final loudness remain unverified.
Review the Actual Path and Output
The cited explicit full-sound path decodes voice directly to 48 kHz stereo float. It distributes mono at equal power and preserves stereo channels. It bypasses the older 44.1 kHz mono conversion and provides no automatic normalization or channel repair. The legacy narration entry behaves differently; its records or listening results do not describe the explicit path.
adopted-packet-copy preserves selected AAC packets without channel changes or gain. A one-sided input can remain one-sided. Repair requires an explicitly changed sound version, rather than a simultaneous claim of frozen audio. A successful copy check does not establish centered speech or listening review.
Review the current actual output: the whole voice sentence on both sides as intended, preserved dialogue and spatial cues, and no added clipping, cancellation, or missing ending. Binding records establish consumed inputs and conversion; waveforms reveal signal; complete listening evaluates the experience. These evidence layers do not substitute for each other.
FAQ
Can loudness normalization restore the right channel?
No. The cited TTS method multiplies samples by one gain; zero remains zero. Correct the actual input channels, then review the new mix.
Should every stereo file become mono?
No. Preserve intentional positioning and distinct useful content. This example corrects a narrator intended at center without discarding other speakers, music, or ambience.
Do it with the skill
Use video-assemble to investigate this one-sided narration. Check the selected input and actual channels, distinguish the device from the file, preserve the original, and select a corrected input before rendering and listening to a short candidate.
Read the method: Actual full-sound channel matrices · Selected inputs, snapshots, and path differences · Global TTS RMS gain does not fill silence · Processed mixes and stream-copy boundaries