SkillPartly tested (needs an account)

Voiceover scripts for ElevenLabs TTS

tts-director

Rewrites short-video scripts for the ear, adds ElevenLabs audio tags within a budget, checks them for free and tests risky lines before a full voice run.

Works with
Claude Code
Good for
Graphic designers, Daily work and research

What it does

TTS Director prepares a script before it goes to a text-to-speech engine, so the voice sounds like a person talking rather than reading. It is built for Persian (Farsi) first and supports English. It labels each line with a beat (hook, setup, tip, joke, call to action) and an energy level, rewrites the text for speech while keeping every fact, and adds ElevenLabs v3/v4 audio tags such as [whispers] or [laughs] only where the text cannot carry the delivery on its own, within a strict tag budget.

A free linter checks the beat map, over-tagging, written-style phrasing and model-specific mistakes, then suggests which lines to A/B test. After you approve the text, an optional script generates the risky lines on your real voice and transcribes them, so you can see whether a tag was performed or read aloud before you pay for the whole script. It can also convert the result for other TTS engines (OpenAI, Azure, Google).

Example prompts

No example prompts are published as quotes. The README says the skill triggers whenever a script is headed to a TTS engine, or when you ask for a more expressive or human-sounding voice. SKILL.md also lists asks about tone, emotion, laughter, pauses or pacing in TTS.

How to install

git clone https://github.com/aligh98/tts-director ~/.claude/skills/tts-director

How to use it

Give the assistant a script that is going to a TTS engine, or ask for a more expressive voice. It maps the beats, rewrites for speech, adds tags, runs the free linter and asks you to approve the exact text before anything is sent to ElevenLabs.

Check it works

After cloning into ~/.claude/skills/tts-director, start a new Claude Code session, paste a short script and say it is headed to ElevenLabs. The assistant should add beat labels (for example # HOOK 5) and run python scripts/tag_lint.py before it suggests any paid generation.

Common problems

Other assistants

What to watch out for

  • No disable-model-invocation, and the description says to use it for any script headed to ElevenLabs or another TTS engine, so the assistant can start it on its own
  • scripts/ab.py sends your text, and the --prev and --next context lines, to api.elevenlabs.io text-to-speech, then sends each clip back to its speech-to-text. Both are billed to your ElevenLabs account
  • Cost: ElevenLabs' API price page lists v3 and v4 text-to-speech at $0.08 per 1,000 characters and speech-to-text by audio time. --dry-run prints the character count only, not the speech-to-text cost
  • ab.py reads ELEVENLABS_API_KEY from the environment, then from a .env file in the current folder or any parent folder, so a key in a parent project's .env can be picked up
  • tag_lint.py --approve writes a SCRIPT.approved.json stamp next to your script; --check-approved exits 3 until the stamp matches the text
  • ab.py writes the mp3 clips and an ab.json file into the --out folder. The scripts are standard-library Python 3.9+; ffmpeg/ffprobe are optional and only used for durations and calibration
  • Repo created on 2026-11-03, 5 commits, no releases or tags

Licence and author

Licence
MIT
Author
aligh98
Source code
github.com/aligh98/tts-director

How we tested it

We installed it on an empty test computer on 8 Nov 2026. Partly tested: installed in a clean test system, but its main features need an account or key we did not use.

Test level

The git clone install worked. The 19 offline tests pass. tag_lint.py on the English example printed the beat map and A/B plan and exited 0. ab.py --dry-run printed 4 clips, 116 characters and wrote nothing. --check-approved exited 3, then 0 after --approve. ab.py without a key stopped with "ELEVENLABS_API_KEY is not set (env var or .env)." We did not call ElevenLabs.

Technical details of the test
Date
8 Nov 2026
Method
Reviewed by our team against the repository at commit d97af42 (README.en.md, SKILL.md, scripts/ab.py, tag_lint.py, reading.py), then ran the free parts in a clean test container with no ElevenLabs key.

Checked again on 8 Nov 2026

Alternatives

  • No catalog entry does the same job. Gemini API CLI can make voiceovers, but it does not direct the script; HyperFrames builds the video the voiceover goes into

FAQ

  • Q: Does it cost money? A: The beat map, the linter and the A/B plan are free and run on your computer. Only ab.py and your full voice run use ElevenLabs, billed to your account. --dry-run prints how many characters would be sent.
  • Q: Is it only for Persian? A: No. It is Persian first and supports English with its own delivery reference.
  • Q: Is it affiliated with ElevenLabs? A: No. The README says it is not affiliated.

Sources

  • github.com/aligh98/tts-director
  • github.com/aligh98/tts-director/blob/d97af42bc26fcf22302eb785dc5577a47699bb52/README.en.md
  • github.com/aligh98/tts-director/blob/d97af42bc26fcf22302eb785dc5577a47699bb52/SKILL.md
  • github.com/aligh98/tts-director/blob/d97af42bc26fcf22302eb785dc5577a47699bb52/scripts/ab.py
  • elevenlabs.io/pricing/api