elevenlabs-voices

The value here is breadth: named personas, many languages, sound effects, batch processing and voice design, all behind one toolkit. The first-run setup wizard that writes config.json is a good decision for an interactive session, but in an unattended context it is a blocker, so provision the key by configuration rather than expecting to walk the wizard. The genuine gotcha is the environment variable: the skill requires ELEVEN_API_KEY and only accepts ELEVENLABS_API_KEY as an alias. Setting only the more obvious name yields an authentication failure that reads like a bad key rather than a naming mismatch.

Use case: Multilingual voice generation and sound effects where persona consistency across languages matters.

# Voice synthesis: setup, then repeatable runs Nearly every early problem with this skill comes from skipping its first-run wizard. The wizard collects the credential, a default voice, a language, an audio quality level, optional cost tracking and an optional monthly budget cap, and writes them to a local config file that the scripts then read. Set it up once and the rest is scripting. ## What the wizard decides for you Choosing a default voice and language up front means later calls can omit both. The budget cap is worth setting on any shared machine, because the skill tracks character usage and estimated cost, and a cap converts an unbounded spend into a clean failure. ## Credentials stay local The key is stored only in the local config file and kept out of version control by the project ignore file. That is the intended handling: do not paste keys into prompts or commit them to a repository. Reconfigure any time by re-running the wizard. ```bash # first run only, then reconfigure whenever settings change python3 scripts/setup.py ``` ## Tips - Run the wizard once and set a budget cap if the machine is shared; it turns over-spend into a clear error. - Set the default voice and language during setup so batch and streaming calls stay terse. - Use the pronunciation dictionary for names and acronyms rather than respelling them in every source line. - gotcha: without a local config file the scripts have no defaults, so a clean checkout appears broken until the wizard has run once. - gotcha: streaming changes how you consume output, so decide per job which mode you want before writing the caller. - Batch related lines of text together to keep character accounting meaningful.