[XB] Text to Speech 1.1.2

• Add configurable OpenAI chunk prefetching
• Added a speech generation queue that reuses in-progress requests for identical text instead of starting duplicate OpenAI generations

Smarter OpenAI Speech, Better Performance!​

This update brings significant improvements to the OpenAI text-to-speech experience with smarter caching, faster playback, more flexible permissions, and enhanced administration tools.

New Features​

• Replaced the content type option with separate permissions for each supported content type.
• Added content context to OpenAI speech requests and audio cache records.
• Added a separate OpenAI TTS character limit option.
• Added linked content details and a content type filter to the audio cache log.
• Added an option to select a default OpenAI voice for users who have not chosen one.
• Added an option to choose the OpenAI audio format (MP3, AAC, or Opus), defaulting to AAC for smaller cached files with the same broad browser support as MP3.

Improvements​

• Audio cache entries are now scoped to their source content_type and content_id.
• Increased the default OpenAI TTS character limit from 160 to 3500 (maximum: 4096), significantly reducing the number of requests required for long posts.
• OpenAI audio is now prefetched while the current chunk is playing, eliminating playback gaps for uncached content.
• Added connection and request timeouts for OpenAI API requests to prevent indefinitely hanging requests.

Bug Fixes​

• Fixed rapid play/pause actions while OpenAI audio was still being generated causing duplicate requests for the same audio chunk.
• Fixed the hourly rate limit permission being applied to cached audio instead of only newly generated OpenAI speech.
• Fixed long sentences or unbroken text exceeding the OpenAI character limit being truncated instead of split into multiple chunks.

​

More voices, more content, and more control!​

This update introduces OpenAI-powered text-to-speech with server-side audio caching, expands support for additional content types, and includes several playback and highlighting improvements.

New Features​

• Added optional OpenAI text-to-speech support as an alternative to browser voices, with per-voice selection, server-side audio caching, configurable hourly rate limiting, and a loading indicator on the speech button.
• Added an ACP page (Content → Text-to-Speech Audio Cache) to browse, filter, and batch-delete cached OpenAI audio.
• Added automatic pruning of unused cached OpenAI audio after a configurable number of days (daily cron).
• Added support for XenAddons AMS Articles, Comments, and Reviews.
• Added support for XenAddons UBS Blog Entries, Comments, and Reviews.
• Added support for Report Center comments.
• Added an option to enable or disable text-to-speech for individual content types.

Bug Fixes​

• Fixed Speech All skipping the first element on the initial run.
• Fixed Speech All incorrectly replaying the first element after normal playback.
• Fixed incorrect word highlighting offsets for messages split into multiple speech parts.
• Fixed the highlight option not reflecting the saved cookie state on page load.

• Added option to hide TTS elements for browsers without SpeechSynthesis support
• Fixed "Speech all" speaking the first element twice