Kitten TTS

Kitten is the compact option in the OfflineTTS engine switcher. The approximate model size can reduce first-load bandwidth and cache use compared with Kokoro, but it does not predict generation speed or compatibility by itself. Browser memory, the selected backend, device thermal limits, extensions, and other open tabs can still affect a run. Test the intended browser and keep a short fallback script before relying on it in a live workflow.

The model produces English speech from eight preset embeddings rather than a broad catalog of named speakers. Choose it when a lightweight draft, a simple expression comparison, or a smaller cached asset is the priority. Use Kokoro or Piper when their speaker selection better fits the text, and use Supertonic when a supported non-English language is required.

Cheer, Serious, Sad, Whisper, Excited, Gentle, Calm, and Neutral are interface labels attached to fixed voice embeddings. The application does not read the semantic meaning of a paragraph and choose an emotion, detect a real speaker, or guarantee that listeners will perceive the label. Punctuation, wording, speed, and the model can make the same preset sound different across passages.

Compare all candidate presets with identical text and settings. Listen beyond the opening sentence for intelligibility, level, artifacts, and whether the delivery stays usable. Record the preset ID and output rate with production files. The page generates locally after assets load and does not send the text to an OfflineTTS synthesis API; normal model delivery and website requests remain documented separately.