Skip to content

SpeechKit v0.75.0

Choose a tag to compare

View all tags
@kombify-release-publisher3 kombify-release-publisher3 released this 03 Oct 14:16
· 11 commits to main since this release
v0.75.0
e07ce25

Highlights

  • Long live dictations stay whole: Deepgram live dictation keeps transcribing past the first pause, and the overlay returns to idle when the text is in. If the live connection drops mid-dictation, the rest of the recording is transcribed from the local capture.
  • Ready before recording: Android and web voice clients wait for the server to be ready before sending microphone audio, and end stalled setup with a clear failure.
  • Stop means stop: Canceling a conversation cuts queued speech and releases audio resources. Bounded microphone and playback queues report overload instead of accumulating old audio.
  • Clear session endings: Voice clients preserve safe failure reasons and end expired authorization promptly. Start a fresh authenticated session to continue; interrupted turns are not replayed.
  • Android keyboard migration: Keyboard embedders must replace ImeVoiceAgentController.audio with consumeAudio(flush, play). This public source API change brings bounded playback and cancellation to the keyboard host.

⚠ BREAKING CHANGES

  • voice: IME source embedders replace the removed audio property with consumeAudio(flush, play).

Features

  • android: run attested hosted Voice co-install sessions (#720) (741d92b)
  • voice: bound readiness, media and session authority (#730) (f8ca761)

Bug Fixes

  • android: bind application artifacts to prepared release version (#722) (4974d11)
  • android: stamp the delivery version into Firebase tester builds (#705) (d71053e)
  • android: start coinstalled Voice from the visible caller (#724) (0992ba5)
  • dictation: keep long Deepgram live dictations whole (#733) (ea738fa)
  • models: download Gemma 4 E4B Q4_K_M from a live, pinned source (#714) (4435a08)
  • oss: export only standalone open-source content to kombifyio/SpeechKit (#717) (c8a5a2d)
  • pipeline: preserve pending audio across recording restart (#726) (641f6b4)
  • server: derive WS origin from public_url and free the session slot on rejected upgrade (#707) (3c7d8f8)
  • voiceagent: deliver agent speech as 24 kHz PCM and keep spaces between streamed words (#708) (2bc0a18)
  • voiceagent: report failed turns as error frames with the agent's typed reason (#712) (49cf642)

Notes cover changes after VERSION 0.74.0 (f78a55815bf3e5f96ba4b1262aab303f7200efc6), the earliest available contiguous 0.74.x VERSION anchor, through selected source ea738fa16d6ade7b012c3591376332f341be790f.

Full Changelog: v0.74.6...v0.75.0

Assets 8
Loading