Gaps between joined MP3 clips: priming padding and Sume's wav default

The symptom

You render ten lines of speech, join them, and hear a small hitch at each seam. The cause is usually that the clips were MP3. The Timeline audio page says the mp3 output option re-adds priming padding at every edge, and recommends keeping wav when the file will be joined again or will drive lip-sync.

Compressed formats carry a bit of encoder padding at the start and often at the end. Join compressed files naively and the padding becomes silence between clips.

What Timeline audio does

Timeline audio's concat operation joins Sume-hosted audio in the sample domain. The docs say there is no re-synthesis and no silence at the seams. The default output is wav with pcm_s16le, described as sample-exact; mp3 is the smaller option.

A pipeline that avoids gaps

Keep everything lossless until the last step.

• Ask Sume TTS for a wav output_format for each line.
• Import any outside files first with POST /v1/media-imports, because Timeline audio accepts only this workspace's media.sume.com audio.
• Concat with operation: concat and the default wav output.
• Encode to mp3 once, at the very end, or ship the wav.

Details that trip people up

Parts must share one channel layout, or the job returns audio_parts_channel_mismatch. A concat takes one to 20 parts, and produced audio is capped at 1,800 seconds. The price is a flat $0.01 per job whether you join two clips or twenty.

There is no GET resource for a timeline audio job. Poll the jobs envelope at /v1/jobs/:id/status and read /v1/jobs/:id/result, whose segments[] give you the offsets of each joined part.

How to confirm the problem is padding

Take two clips, join them as wav, and listen. Then encode each clip to mp3, join those, and listen at the same spot. If the first is clean and the second ticks, you have found it. Sume's own join does not insert silence at the seams, so the difference is the format.

The fix is order of operations: lossless join first, lossy encode last.

Takeaway

If your joins click, check the format first. Use wav through the whole chain and encode once. See wav vs mp3 for the general tradeoff.