murtaza-nasir/speakr v0.8.11-alpha

v0.8.11 - Video Retention, Parallel Uploads & Speaker Improvements

on GitHub

Video Retention, Parallel Uploads & Speaker Improvements

Big feature release. Video playback, faster uploads, duplicate detection, and a bunch of speaker identification improvements.

New Features

• Video Retention - Set VIDEO_RETENTION=true to keep video files for in-browser playback instead of extracting audio and throwing away the video. Great for lectures, presentations, and screen recordings. Audio gets extracted to a temp file for transcription and cleaned up after. Seeking works properly via HTTP Range requests. There's a collapsible toggle in the player controls so you can hide the video when you just want the transcript.

• Parallel Uploads - Files now upload concurrently instead of one at a time. Makes batch uploads way faster. Control how many go at once with MAX_CONCURRENT_UPLOADS (default 3).

• Duplicate Detection - SHA-256 hashing on upload catches duplicate files. You get a warning toast and a clickable indicator in the sidebar showing existing copies.

• Speaker API Endpoints - REST API for speaker identification and assignment with bearer token auth. Extracted the identification logic into a shared service so the API and UI use the same code.

• Speaker Identification Improvements - Split button UI for identify/apply, bulk "Apply Suggested" for filling in names without calling the LLM, name sanitization to strip LLM commentary, and AUTO_IDENTIFY_RESPONSE_SCHEMA for local LLM compatibility.

• Volume Controls - Volume slider popups and mute indicators on all audio and video players.

Changes

• Speaker profiles preserved by default - Deleting all recordings for a speaker no longer auto-deletes their profile. Voice embeddings are aggregated and can't be reconstructed, so we keep them now. Set DELETE_ORPHANED_SPEAKERS=true if you want the old cleanup behavior.

Bug Fixes

• ASR_DIARIZE=false was being ignored, diarization always ran
• Bulk delete failed with integrity error on speaker snippets
• File monitor stability check capped at 2s regardless of config
• clean_llm_response was too aggressive, broke markdown formatting
• Null transcription crash on certain processing paths
• Safari iOS recording view not updating
• Missing translations for folders, API tokens, recording recovery, events, and speakers

Compatibility

Backwards compatible with v0.8.x. Video retention and parallel uploads are opt-in via env vars. The speaker profile preservation change is the only default behavior change. See docs for DELETE_ORPHANED_SPEAKERS if you need the old behavior.

Thanks to ItsMly for the markdown fix (#232)