What's Changed

• feat: opt-in VAD run gate for the Ultra/Redux heads, and the real-recording VAD evidence by @localai-org-maint-bot in #96
• fix: use 64-bit seek/tell that MSVC has in the bundle loader by @localai-org-maint-bot in #97

Full Changelog: v0.6.0...v0.6.1

What's Changed

• mel: skip zero-weight filterbank entries in the mel projection by @haydonryan in #64
• perf: move the mel buffer into the 1-item MelBatch instead of copying it by @sylvestre in #67
• mel: precompute FFT twiddle factors and reuse scratch by @haydonryan in #65
• fix: bound the streaming mel buffer to a sliding window (fixes #63, the streaming RSS leak) by @simobenziane in #69
• fix: include cstdint in the backend header by @localai-org-maint-bot in #72
• feat(diarization): add the model card's low-latency streaming modes by @localai-org-maint-bot in #73
• Add sound-event detection and a combined scene stream by @localai-org-maint-bot in #75
• fix: GCC 16 build, and TDT beam search failing on parakeet-tdt-0.6b-v3 by @EnacheB in #74
• chore: pin ced.cpp to main by @localai-org-maint-bot in #76
• feat: identify speakers in the scene stream (voice-detect.cpp, C-API v9) by @localai-org-maint-bot in #78
• feat(capi): raw-embedding enroll and diarize-only speaker naming, ABI v10 by @localai-org-maint-bot in #79
• feat(capi): export clean speaker profiles by @localai-org-maint-bot in #80
• feat: support Moondream Parakeet Ultra and Redux by @localai-org-maint-bot in #77
• feat: exact batched decode on CPU by @localai-org-maint-bot in #81
• perf: decode VAD segments in batches by @localai-org-maint-bot in #82
• feat: standalone VAD (model head and Silero), streaming, and external VAD for transcribe by @localai-org-maint-bot in #83
• docs: link the published Ultra, Redux and Silero GGUF files by @localai-org-maint-bot in #84
• fix: explicit licences in publish_hf.py (no CC-BY default), docs/licenses.md by @localai-org-maint-bot in #86
• feat: VAD-only GGUF files (head, front end, subsampler) and a slicer by @localai-org-maint-bot in #87
• docs: VAD benchmarks (Silero vs the Parakeet head vs whisper.cpp) with reproducible scripts by @localai-org-maint-bot in #88
• feat: bundle GGUF format (ASR, VAD, diarization, sound events and voice in one file) by @localai-org-maint-bot in #85
• docs: why the VAD head fires on noise (root-cause study, mitigations) and a Silero plus head fusion experiment by @localai-org-maint-bot in #89
• feat: fingerprint the speaker encoder in the registry and refuse a family mismatch before naming by @localai-org-maint-bot in #90
• fix: Silero VAD must not leave the global backend at one thread by @localai-org-maint-bot in #91
• feat: load ced and voice bundle components from memory (no temp file), bump ced.cpp and voice-detect.cpp by @localai-org-maint-bot in #92
• docs: rework the README (features first, quick start, demos) by @localai-org-maint-bot in #93
• feat: trim VAD segments to speech (default 0.3 s) and an opt-in local-confidence word filter by @localai-org-maint-bot in #94
• docs: the Redux trim regression was noise (enlarged measurement), keep the 0.3 default by @localai-org-maint-bot in #95

New Contributors

• @haydonryan made their first contribution in #64
• @sylvestre made their first contribution in #67
• @simobenziane made their first contribution in #69
• @localai-org-maint-bot made their first contribution in #72
• @EnacheB made their first contribution in #74

Full Changelog: v0.5.0...v0.6.0

What's Changed

• fix(batching): stop shorter clips in a batch from failing to transcribe by @richiejp in #49
• build(arm64): cross-compile parakeet-cli for the Raspberry Pi 5 by @richiejp in #32
• fix: release host weight staging buffer after device upload (47% memory saving) by @masc-it in #53
• feat(tdt): add opt-in N-best beam decoding by @inlanger in #56
• Extends workflows to build an release cuda12/linux based images by @derjohn in #52
• Add parakeet_capi_transcribe_pcm_logits: expose CTC log-probs by @aanchan in #57

New Contributors

• @masc-it made their first contribution in #53
• @inlanger made their first contribution in #56
• @derjohn made their first contribution in #52
• @aanchan made their first contribution in #57

Full Changelog: v0.4.0...v0.5.0

What's Changed

• ci(release): package parakeet-server alongside parakeet-cli by @quartzjer in #30
• ci(release): build the linux vulkan bundle for arm64 by @quartzjer in #31
• feat(server): emit per-word confidence in verbose_json by @quartzjer in #34
• docs: add Community projects section (iOS streaming demo) by @Kashif-E in #37
• Make server e2e readiness timeout configurable by @quartzjer in #43
• Add version flags to parakeet-cli and parakeet-server by @quartzjer in #41
• Support stdin WAV input in parakeet-cli by @quartzjer in #42
• fix(tokenizer): strip special tokens from offline decode text by @localai-bot in #48

New Contributors

• @quartzjer made their first contribution in #30
• @Kashif-E made their first contribution in #37

Full Changelog: v0.3.2...v0.4.0

What's Changed

• ci(release): ship the C-API shared library and header (#25) by @localai-bot in #29

Full Changelog: v0.3.1...v0.3.2

What's Changed

• fix(server): build the OpenAI server on Windows (MSVC) by @localai-bot in #28

Full Changelog: v0.3.0...v0.3.1

What's Changed

• feat(capi): ABI v5 - surface vs distinction across the C boundary by @richiejp in #24
• build(cuda): enable GGML_CUDA_GRAPHS on CUDA builds by @localai-bot in #26
• OpenAI-compatible HTTP server + docker image by @mudler in #8

New Contributors

• @richiejp made their first contribution in #24

Full Changelog: v0.2.0...v0.3.0

What's Changed

• feat: banded local (Longformer) attention — fix O(T^2) long-audio OOM by @localai-bot in #9
• Add nemotron-3.5-asr-streaming-0.6b (multilingual prompt-conditioned streaming) by @localai-bot in #10
• Batch mode for nemotron: batched causal subsampling + batched target_lang C-API by @localai-bot in #11
• docs: supported-models table in the README by @localai-bot in #12
• docs(bench): nemotron GPU numbers on the GB10 by @localai-bot in #14
• fix: reset the streaming decoder on / so transcription continues (#13) by @localai-bot in #15
• feat(capi): segment-timestamp support (frame_sec + streaming JSON, ABI v4) by @localai-bot in #16
• fix: define M_PI for MSVC builds by @mvanhorn in #18
• fix: tile subsampling for long audio to avoid ggml 2^31 tensor overflow on GPU by @localai-bot in #19
• fix: select integrated GPUs and allow PARAKEET_DEVICE to name a device by @localai-bot in #20
• ci: pre-built release binaries for linux, macos and windows by @localai-bot in #22

New Contributors

• @localai-bot made their first contribution in #9
• @mvanhorn made their first contribution in #18

Full Changelog: v0.1.2...v0.2.0

What's Changed

• Batched encoder: run N clips through one fused ggml graph by @mudler in #1
• fix: consolidate metal support by @mudler in #4
• ci: build and publish the parakeet-cli container image to ghcr by @mudler in #7

Full Changelog: v0.1.1...v0.1.2

Full Changelog: https://github.com/mudler/parakeet.cpp/commits/v0.1.1