# Known Limitations Things that do not work the way you might expect yet. Many have a planned fix on the [[Roadmap]]. ## Content > [!bug] HTML comments are read aloud > `<!-- HTML comments -->` are not stripped and are read as literal text, symbols included. Obsidian comments (`%% ... %%`) work and have their own settings. - Sentence and word level highlighting are not available. - Images are silently skipped. They are not described. - A `%%` comment containing heading-like text (`# ...`) can throw off section boundaries. ## Saved audio - **Staleness is whole-note.** The hash covers the full note, so editing anything, even outside what was read (for example after reading a selection), marks the audio outdated. - **Replace keeps the old filename.** With "Replace existing file", a narrator change keeps the original filename (with the old voice name in brackets) and only replaces the contents. - **Multi-part files are concatenated bytes.** Chunks are joined without re-muxing the MP3 stream. This works with ElevenLabs output but is not strictly spec-correct MP3 concatenation. - **Part navigation can fall back.** During Play saved, if chunk-affecting settings changed since the audio was made, or the byte lengths do not add up, the file plays as a single non-navigable piece even though the note still says "up to date". Regenerate to fix it. - **Clear has no undo** for the properties. The file goes to trash per your vault setting. ## Background generation - **Outdated jobs still play the old version.** A job whose note was edited (or whose narrator changed) after it was generated is marked **Outdated**, but playing it still plays the old text. Pressing **Read** or the **Background** button replaces it with a fresh job. **Move to background** on a read of an edited note moves the old read as it is. ## Playback and highlighting - No scrubbing to an arbitrary time, only relative rewind and skip. - Chunk and section highlight positions are proportional estimates and can be a word or two off at a boundary. - Highlighting needs Editing view. It cannot draw in Reading view. - Neither highlighting nor scroll-to-current applies to selection reads. ## ElevenLabs - Only the first **100 voices** on a provider's account are listed in a profile's voice dropdown. - The rate limit fallback to one chunk at a time only lasts for the read in progress. Parallel generation is set per provider. - Eleven v3 is a research preview and can mispronounce or invent words. Professional Voice Clones are not fully optimized for it yet. - **Auto-generate on open** spends credits on every open of a missing or outdated note. ## Platform - Requires Obsidian 1.13.1 or newer. - Requires an ElevenLabs account. ElevenLabs is the only provider type so far, though you can add several ElevenLabs providers. No offline or local voice yet. > [!question] Found something not listed? > Open an issue on the [GitHub repository](https://github.com/pablokvitca/note-narrator).