Results

Kokoro-Hexagon 0a03be39Updated 2026-09-27

Dated measurements from Kokoro-Hexagon. Each result records one narrow measurement on named hardware with its inputs and hashes. Results that used QNN are reference evidence, not product gates.

Date	Measurement
2026-09-27	Stock scalar-reference profile
2026-09-27	Model weight admission and synthesis handoff
2026-09-27	Learned generator composition
2026-09-27	Hexagon linear-tile emission gate
2026-09-27	AdaIN convolution runspace scheduling audit
2026-09-26	Upstream source study inventory
2026-09-26	Text encoder FP32 reference
2026-09-26	Style-conditioned duration branch
2026-09-26	Stock AdaIN residual block differential
2026-09-26	Sherpa-onnx Kokoro streaming-method audit
2026-09-26	Same-session CDSP interface inventory
2026-09-26	S23 CDSP node inventory
2026-09-26	Recovered device DSPQueue diagnostics
2026-09-26	ReadyToRun payload audit
2026-09-26	LSTM and duration FP32 references
2026-09-26	Live DSPQueue diagnostic reproduction
2026-09-26	Learned generator primitives
2026-09-26	Kokoro methodology study mirrors
2026-09-26	Independent build-location migration
2026-09-26	Generator source and STFT reference
2026-09-26	F0/N bounded FP32 reference
2026-09-26	DSPQueue upstream boundary
2026-09-26	DSPQueue capability comparison
2026-09-26	Decoder prelude and core reference
2026-09-26	Connected acoustic branches
2026-09-26	CDSP transport evidence ledger
2026-09-26	Bounded ALBERT FP32 reference
2026-09-25	Private model-root and signed load gate
2026-09-25	Native-binding audio smoke
2026-09-25	Model-less NativeActivity appliance
2026-09-25	Model-less appliance with Kokoro managed namespace
2026-09-25	Model-less appliance rebuild before managed-namespace change
2026-09-25	Direct FastRPC native-open probe
2026-09-24	Windows AOA protocol result
2026-09-24	Speaker baseline
2026-09-24	Resident provider host control
2026-09-24	Persistent QNN and AAudio pipeline result
2026-09-24	Persistent AAudio stream reuse
2026-09-24	Native AAudio speaker path
2026-09-24	Model-lowered R0Sub0 versus LLVM
2026-09-24	Model assembly Windows load result
2026-09-24	Kokoro managed lowering and archive
2026-09-24	Float playback handoff
2026-09-24	Direct V73 W4A8 contraction and INT32 readback
2026-09-24	Cross-SoC result: model-lowered R0Sub0 versus LLVM
2026-09-24	Corpus float playback
2026-09-23	Which op names V73 accepts
2026-09-23	What our context binaries are actually made of
2026-09-23	What is actually inside a prepared context
2026-09-23	What a serialized QNN context actually charges for
2026-09-23	Verification result: Exhaustive HMX & Pcycle Named Encoders vs SDK LLVM Assembler Oracle
2026-09-23	Snake Sine Physical Lowering Ratchet
2026-09-23	R0.0 HVX Physical-Substrate result (R0SUB0-HVX-POINTWISE-R001)
2026-09-23	Qualcomm Hexagon V73 Hardware Capability & VTCM Geometry
2026-09-23	PowerShell-emitted Hexagon ELF executes on the S23
2026-09-23	Physical HMX Accelerator Acquisition & Instruction Execution
2026-09-23	Physical Hexagon V73 Benchmark — Pure PowerShell vs LLVM Clang
2026-09-23	Lowering a model written in PowerShell
2026-09-23	Kokoro ResBlock R0Sub0 Aligned 7712 Padding & Post-Affine Masking
2026-09-23	Kokoro affine subgraph emitted as V73 instructions
2026-09-23	Generator resblock r0, emitted from the checkpoint
2026-09-23	Executable mapping on the DSP is closed
2026-09-23	Delayed shared-buffer execute mapping smoke test
2026-09-23	Channel tiling is exact; the naive phase fold is not
2026-09-23	Channel tiling at the generator's real tensor size
2026-09-23	Breath groups from SMA's tokenizer
2026-09-23	Authoring a prepared context's weight payload
2026-09-22	What HTP V73 will and will not do with tensor datatypes
2026-09-22	Reading the Kokoro checkpoint in PowerShell
2026-09-22	R009 F0/N second pair: Resize in the shortcut
2026-09-22	FastRPC from PowerShell on the cDSP
2026-09-22	Corpus benchmark
2026-09-22	Corpus benchmark
2026-09-22	Capacity bucket: 64 vs 160 frames, same phrase
2026-09-22	Canonical fp16 baseline — frozen 2026-09-22