How do I add a spoken tagline and sound sting to the end of an ad?

To add a spoken tagline and sound sting to the end of an ad, generate the tagline as its own short TTS job, generate a one- to three-second sting with Music Router, and join them to the ad read with a Timeline audio concat. A 60-character tagline is $0.01, the sting is $0.125 and each concat is $0.01, so $0.145 the first time, and then 1 cent for the concat each time you attach the same tag and sting to a new ad read.

The end tag is the part of an ad that never changes, which makes it the best candidate to generate once, approve once and then reuse across every version.

Write the tag

Keep it to four to eight words, ending on the brand. Say it the same way every time, which is the main reason to generate it as a separate file instead of as the last sentence of each read. Voices vary a little between generations; one approved file does not.

Name pronunciation matters most here. If the brand name is read oddly, add a pronunciation dictionary and send its id with the request. Test it alone, since a one-second file is a cheap place to find a problem.

Generate the tag and sting

Send the tag as wav at 44.1 kHz so it matches the ad reads. A tight, slower read at generation_config.speed 0.9 sounds more final. For the sting, put the length in the Music Router prompt, since duration is rejected: 'A 2-second brand sting: a rising two-note motif ending on a clean chord, no vocals.' Router prompts are 1 to 5,000 characters, and a non-empty negative_prompt returns a 400, so describe what you want instead of what to avoid.

curl -X POST https://api.sume.com/v1/music-router/generate \
  -H "Authorization: Bearer $SUME_API_KEY" \
  -H "Content-Type: application/json" \
  -H "Idempotency-Key: brand-sting-v1" \
  -d '{
    "model": "sume/music-auto",
    "prompt": "A 2-second brand sting: a rising two-note motif ending on a clean, resolved major chord. Warm synth and soft bell. Instrumental, no vocals."
  }'

Attach it to any ad read

Timeline audio concat joins up to 20 parts, gapless and without re-synthesis, for a flat $0.01 a job. Put the ad read first, the tag second and the sting last, or the sting under the tag by rendering with a soundtrack. If you want the sting to land with the last word, trim the read's tail with duration, and keep the parts the same channel layout; a mismatch fails with audio_parts_channel_mismatch.

curl -X POST https://api.sume.com/v1/timeline-1.0/audio \
  -H "Authorization: Bearer $SUME_API_KEY" \
  -H "Content-Type: application/json" \
  -H "Idempotency-Key: ad-v7-with-tag" \
  -d '{
    "operation": "concat",
    "parts": [
      { "url": "'"$AD_READ"'" },
      { "url": "'"$TAGLINE"'" },
      { "url": "'"$STING"'" }
    ]
  }'

Check the end of the ad by ear

Listen to the join with the volume up and with it low. The tag should be as loud as the read, not louder, and the sting should end cleanly rather than ring into silence. If the sting is too long, regenerate it with a shorter prompt; it is a flat $0.125 whatever its length.

Play the ad on a phone speaker as well. A tag that sounds fine in studio headphones can be lost where the ad actually runs, and the fix is cheap: a slower read of the same words, or a one-second gap before the tag.

Reuse across versions

Once the tag and sting exist, each new ad costs its own read plus one concat. Ten versions of an ad, each 450 characters, cost 3 cents of speech apiece because every read rounds up to a whole cent, plus 1 cent for each concat, so 40 cents for all ten with the tag and sting already made. Mp3 output adds encoder padding, so choose wav for the parts and encode once at the end.

• Approve the tag file once, in the loudest and quietest places it will play.
• Keep the tag and sting in a folder named for the brand, with the idempotency keys written down.
• If the tagline changes, the sting can stay; only the 1-cent tag is regenerated.