Skip to content

v0.89.0

Choose a tag to compare

View all tags
@sboily sboily released this 24 Sep 04:35
· 2142 commits to main since this release
v0.89.0
ffb6c20

What's Changed

  • docs(changelog): promote 0.89.0
  • docs(examples): a local voice assistant with Vui, sherpa-onnx and Ollama (RMK-203)
  • feat(pipeline): EventType.INSTRUCTION directs an agent without a participant's line (RMK-200)
  • fix(voice): bound Vui's cache by the 6 minutes of audio it was trained on (RMK-194)
  • refactor(voice): tidy the Vui decoder re-seed per review (RMK-199)
  • fix(voice): re-seed Vui's audio decoder at the start of each reply (RMK-199)
  • fix(voice): a handoff greeting is the agent's line on every channel (RMK-200)
  • fix(voice): a configured greeting is the agent's line on every realtime provider (RMK-200)
  • fix(voice): a handoff's cue to introduce the new agent is an instruction (RMK-200)
  • fix(voice): bound and release Vui's cache, cut at the heard frame, per review (RMK-194)
  • fix(voice): close the TTS stream once playback stops (RMK-194)
  • feat(voice): Vui Nano TTS provider with AUDIO conversation context (RMK-194)
  • fix(voice): stitch ElevenLabs requests within one voice, per review (RMK-193)
  • feat(voice): ElevenLabs request stitching from the TTS context (RMK-193)
  • fix(voice): keep a continuous-mode verdict for the whole utterance per review (RMK-196)
  • fix(voice): SEMANTIC waits for the words of transcribed speech (RMK-196)
  • fix(conference): barge-in position leaves out synthesis latency (RMK-192)
  • fix(conference): the roster writes the status of the records it homes only (RMK-195)
  • fix(voice): tighten the SEMANTIC hold and the playback measure per review (RMK-190, RMK-192)
  • fix(voice): SEMANTIC waits for words instead of judging an empty onset (RMK-190)
  • fix(voice): barge-in hooks and threshold measure the audio heard (RMK-192)
  • fix(voice): label speech audio with the pipeline's rate, not the transport's (RMK-191)
  • fix(voice): tighten the TTS context per review (RMK-187)
  • fix(voice): an utterance already played is not recorded as interrupted (RMK-187)
  • feat(voice): conversation context for TTS providers (RMK-187)
  • fix(voice): discard a backchannel speech segment during playback (RMK-187)
  • fix(voice): a stop never starts a tool, closes unrun calls, keeps row order (RMK-189)
  • fix(voice): let a running tool finish on a barge-in, keep an in-flight segment (RMK-189)
  • fix(voice): a barge-in closes the streamed response and stores it cancelled (RMK-189)
  • test(voice): join sessions with kit.join in the fan-out tests (RMK-188)
  • fix(voice): pull the fanned-out response on demand, surface source errors (RMK-188)
  • fix(voice): play a non-streamed response to every session in parallel (RMK-188)
  • fix(voice): stream a response to every session of the room, in parallel (RMK-188)
  • Begin 0.89.0.dev0 development

Full Changelog: v0.88.0...v0.89.0

Assets 3
Loading