0.7.0

feature

• Added speech-aware chunking that plans chunks around where people actually talk.

  • Silent stretches are never uploaded or sent to the model.
  • A recording broken up by long silences is split into speech islands and chunked per island, so a late remark keeps its own chunk without transcribing the gap before it.
  • Skipped ranges keep a numbered slot in the transcription file and can be transcribed later.
  • The detail log shows a speech activity sparkline with chunk boundaries, skipped ranges, and a per-chunk timeline.
  

• Added per-chunk retry from the progress log, rewriting only that chunk's region in the transcription file and leaving manual edits to other chunks intact.

• Added uploaded file reuse on chunk retry so most retries skip the upload and go straight to the request.

• Added chunk numbers and time ranges to progress log lines so parallel chunks can be told apart.

fix

• Fixed a trailing chunk shorter than two minutes costing a full upload and request that returned nothing useful.
• Fixed NotFoundError from removeChild when the plugin was disabled with the progress view open.
• Fixed the progress view's sidebar placement being discarded when the plugin was disabled.
• Fixed transcript text containing $& or $` being mangled when written to the transcription file.