Skip to content

v0.9.0

Choose a tag to compare

View all tags
@sboily sboily released this 10 Jun 20:53
· 46 commits to main since this release
v0.9.0
5415c07

What's Changed

  • docs(changelog): fold codec floor and thinking-coalescing into unreleased 0.9.0
  • docs(changelog): describe thinking-delta coalescing and aiortp 0.3.2 codec floor
  • refactor(ai): split oversize thinking flushes, dedupe coalescer setup
  • perf(ai): coalesce thinking-stream deltas into time/size windows
  • chore(deps): require aiortp>=0.3.2 for vectorised RTP codecs
  • docs(changelog): stamp 0.9.0
  • chore: reword history-narration comments to current-state
  • fix(fastrtc): skip sends on non-open RTCDataChannel
  • perf(voice): vectorise mu-law encode and rms_db, keep numpy lazy
  • perf(core): size RoomContext.recent_events to what bound channels read
  • feat(realtime): warn when falling back to the pure-Python sinc resampler
  • perf(realtime): split tool-call loop step and instrument its segments
  • refactor(aec): collapse duplicated reference log into logger.log(level)
  • fix(examples): honor the documented MUTE_MIC override in the Gemini example
  • fix(local): feed playback-time AEC reference continuously, silence included
  • fix(local): prebuffer realtime speaker playback to absorb provider jitter
  • perf(aec): vectorise energy diagnostics and move them off the AEC lock
  • perf(realtime): skip context build for hookless partials and speech events
  • refactor(realtime): extract _reset_outbound_resampler helper
  • feat(sip): expose outbound pacer prebuffer/headroom on SIPVoiceBackend
  • perf(realtime): move per-session resampling off the event loop

Full Changelog: v0.8.0...v0.9.0

Assets 2
Loading