1. What this lesson is for

In this lesson you build a stable text-to-speech (TTS) pipeline: from a script draft in a text file to a voiceover clip sitting on your CapCut timeline, in sync and ready to export. Keep the module principle in mind: script quality drives TTS quality more than any model setting. A clean, well-punctuated script with one idea per sentence will beat any amount of slider tweaking on a messy one.

By the end of the module you will deliver synchronized English and Polish voiceover tracks built from one shared script structure. This lesson sets up the plumbing; lessons 5.2 and 5.3 tune each language, and lesson 5.4 ties both versions into one edit.

2. Account setup and the voice library

ElevenLabs is a web app (with an API for automation). You sign up, pick a plan, and work mainly in its text-to-speech area. The interface and plan limits change often, so treat the points below as concepts to look for rather than exact buttons:

• Plan and quota - every plan gives you a monthly character (or credit) allowance. Each re-render of a line spends it again. Check your usage page before a big project and note the reset date.
• Commercial rights - check whether your plan allows commercial use of generated audio. Client and monetized work usually needs a paid tier.
• Voice library - premade voices plus a community library of shared voices. Filter by language, accent, gender, age and use case (narration, conversational, and so on). Save the few you like to your own list so you do not hunt for them again.
• Models - ElevenLabs offers several speech models (multilingual, low-latency, more expressive). Multilingual models are the ones you want for Polish. Names change between releases - pick the current multilingual model and write its name in your project notes.
• Voice cloning - you can create a voice from recordings. Only do this with your own voice or with written consent from the speaker (see the warning below).

3. The pipeline at a glance

Every voiceover in this course follows the same six stages. The loop back from QC is normal - most lines pass first time, a few need a script fix or a re-render.

4. Script preparation for TTS

TTS reads exactly what you write. It does not know which word is important, where you wanted a breath, or that "v2" should be "version two". Prepare the script for the ear, not the eye:

• One idea per sentence. Split anything longer than about 20-25 words.
• Punctuation is your pacing control: a comma is a short pause, a full stop a longer one, a paragraph break longer still.
• Spell out numbers, units and symbols the way they should be spoken.
• Expand or respell acronyms and tricky names (more in lesson 5.2).
• Remove anything visual only: URLs, bullet symbols, emoji, "see below".
• Keep a scene ID at the start of each line in your master file, but strip it before you paste text into ElevenLabs.

BEFORE (written for reading)
S03: Our app v2.1 cuts export time by ~40% (see chart) & supports 4K@60fps!

AFTER (written for listening)
S03: Version two point one of our app cuts export time by about forty percent.
     It also supports 4K video at sixty frames per second.

5. Export formats and naming conventions

ElevenLabs typically offers MP3 at several bitrates and, on some plans, higher quality formats such as WAV/PCM. For editing, pick the highest quality your plan allows (a lossless format or a high-bitrate MP3 at 44.1 kHz) and use the same choice for the whole project. Mixing sample rates and formats is a common source of subtle sync drift and level jumps.

Pattern:  <project>_<scene>_<LANG>_v<NN>[_APPROVED].<ext>

promo26/
  script/
    promo26_master_v03.txt
    promo26_EN_v03.txt
    promo26_PL_v03.txt
  vo/
    EN/  promo26_S01_EN_v01.mp3  promo26_S02_EN_v02_APPROVED.mp3 ...
    PL/  promo26_S01_PL_v01.mp3  promo26_S02_PL_v01_APPROVED.mp3 ...
  settings/
    voice_settings.txt   (voice name, model, slider values, date)

• Zero-pad numbers (S01, v01) so files sort correctly.
• Never overwrite a file - bump the version. You will want to go back.
• Rename downloads immediately; the default names from the browser are useless later.

6. Importing voiceovers into CapCut

1. Import the approved files from your vo/EN (or vo/PL) folder into the CapCut media panel. Importing the whole folder keeps them together.
2. Drag them onto a dedicated audio track and keep that track for voiceover only. Music and SFX go on their own tracks (see module 4).
3. Place each scene clip at the start of its matching picture section. Leave the natural tail of silence on each clip; trim it only when you sync.
4. Normalize or adjust the VO level consistently, then check it against music with ducking as in module 4.
5. Generate captions from the voiceover with CapCut's auto-captions feature if you use them, then proofread against the script.

7. Checklist and risk controls

• Voice profile selected and written down (voice name, model, settings, date) in settings/voice_settings.txt.
• Export format standardized for the whole project.
• Re-render strategy defined - per-scene files, versioned names, and who approves a retake.
• Quota tracking - estimate characters (script length x expected retakes, roughly 1.5-2x) before you start.
• Backup voice - shortlist a second voice in case your main voice is removed from the library or changes.
• Local archive - download and keep every approved file. Do not rely on the generation history in the web app as your only copy.

8. Exercise

1. Take a 5-scene script (about 60 seconds). Rewrite it for listening using the rules in section 4.
2. Choose one English voice and write down the voice, model and settings.
3. Render one file per scene and name them with the pattern from section 5.
4. Import them into CapCut on a dedicated voiceover track and sync them to a rough picture edit.
5. Change one line in the script, re-render it as v02, replace it on the timeline and note what else moved.
6. Retrospective: how many characters did you spend, and which stage slowed you down?

Further reading

• ElevenLabs documentation
• ElevenLabs - Text to speech product guide
• ElevenLabs prohibited use policy
• CapCut Help Center