hanzoai/speech
2
0
Fork 0
mirror of https://github.com/hanzoai/speech.git synced 2026-10-04 07:37:19 +00:00
Code Issues Packages Projects Releases Wiki Activity

Commit Graph

  • d5de7c9e94 fix(transcript): a window committed inside unbroken speech is cut in its quietest lull, never mid-word main zeekay 2026-10-04 00:37:08 -07:00
  • 39abc2b228 feat(speech): max_seconds holds a transcription to a length, and the decode stops just past it zeekay 2026-10-03 01:09:28 -07:00
  • 0a0f42c9ea docs(speech): RTF and memory as measured in the pod zeekay 2026-10-02 14:00:05 -07:00
  • 4f6eff020e feat(speech): rebuild on sherpa-onnx in Go — parakeet, whisper turbo, kokoro zeekay 2026-10-02 13:36:47 -07:00
  • 68ea213010 ci: a runner label names the platform, not the org zeekayandHanzo Dev 2026-09-08 12:09:47 -07:00
  • 9631675dff Publish both architectures, not just the pool's own zeekay 2026-09-05 19:07:42 -07:00
  • 71f43202b0 chore: symlink AGENTS.md + GEMINI.md + CLAUDE.md -> LLM.md zeekayandHanzo Dev 2026-09-02 01:14:23 -07:00
  • abcd9b9faa Say where the sustained-rate number comes from zeekayandHanzo Dev 2026-08-20 13:06:29 -07:00
  • 1be055905d speech 0.1.13 — declare timestamp_granularities instead of digging it out zooqueen 2026-08-13 18:22:18 -07:00
  • 77502bbfda speech 0.1.12 — return the word timings the decode already found zooqueen 2026-08-13 18:12:13 -07:00
  • 7830dedfb2 speech 0.1.11 — a KMS that is restarting is not a KMS that is down zeekayandHanzo Dev 2026-08-12 20:00:28 -07:00
  • 15a656cce8 speech 0.1.10 — each gate runs its own suite, and says which zeekayandHanzo Dev 2026-08-12 18:52:42 -07:00
  • 1dd157eefa speech 0.1.9 — withdraw the numbers, keep the code zeekayandHanzo Dev 2026-08-09 12:28:24 -07:00
  • 19fe03ca24 speech 0.1.8 — do not size the decoder's thread pool by hand zeekayandHanzo Dev 2026-08-09 11:51:01 -07:00
  • 5eb25e2af8 speech 0.1.6 -> 0.1.7 zeekayandHanzo Dev 2026-08-09 11:45:49 -07:00
  • fd6d7ed375 speech: every session gets a turn, and the decodes are really parallel zeekayandHanzo Dev 2026-08-09 11:31:26 -07:00
  • 6d9aca2b3a speech: decode the talking, not the room zeekayandHanzo Dev 2026-08-09 11:33:49 -07:00
  • a48646613c speech: the live fixture resamples in numpy, so say that zeekayandHanzo Dev 2026-08-08 15:31:04 -07:00
  • 6c9e104b4f speech: say when a decode fails, instead of transcribing nothing quietly zeekayandHanzo Dev 2026-08-08 15:23:27 -07:00
  • 6388fd9ab0 speech: decode in a pool, so the process can exit while one is running zeekayandHanzo Dev 2026-08-08 15:14:41 -07:00
  • ed0bf45113 speech: record why the transport is HTTP and why the second gate exists zeekayandHanzo Dev 2026-08-08 15:05:07 -07:00
  • 81071b1d83 speech: a transcript that grows, metered by the audio it received zeekayandHanzo Dev 2026-08-08 14:59:28 -07:00
  • 7dc4714dcc ci: read the pin token at the path KMS serves zeekayandHanzo Dev 2026-08-07 21:02:51 -07:00
  • da3529e4a9 ci: the build ends by pinning, so a published image is a deployed one zeekayandHanzo Dev 2026-08-07 20:41:51 -07:00
  • 005298be18 ci: the gate is a build stage, so it fails only for reasons in the code zeekayandHanzo Dev 2026-08-07 16:23:13 -07:00
  • e66ed069f5 ci: the test gate depends on docker, which this runner has zeekayandHanzo Dev 2026-08-07 16:09:33 -07:00
  • 6f516b97e5 speech: return the duration it measured, and the container it produced zeekayandHanzo Dev 2026-08-07 16:00:10 -07:00
  • d519c1e460 speech: CPU STT (faster-whisper) + TTS (kokoro) behind one OpenAI-shaped surface zooqueenandHanzo Dev 2026-08-06 15:03:58 -07:00