harvestsu/seeed-local-voice-rk-artifacts
Add Piper en_US/lessac 512-mel artifacts (split_piper_vits --seq-len 128 --mel-len 512)
Add Piper en_US/lessac 512-mel artifacts (split_piper_vits --seq-len 128 --mel-len 512)
Add Piper en_US/lessac 512-mel artifacts (split_piper_vits --seq-len 128 --mel-len 512)
Add Piper en_US/lessac 512-mel artifacts (split_piper_vits --seq-len 128 --mel-len 512)
Add missing rk3588 w8a8 ASR decoder (referenced by rk_manifest since 2026-06-10)
release: add external Japanese dictionary resource
release: add RK3588 Kokoro ConvOnly v1.0 bundle
release: add RK3576 Kokoro ConvOnly v1.0 bundle
Add PROVENANCE.md (Qwen3-4B RK1828 w4a16)
Add Qwen3-4B.weight (Qwen3-4B RK1828 w4a16)
Add Qwen3-4B.embed.bin (Qwen3-4B RK1828 w4a16)
Add Qwen3-4B.rknn (Qwen3-4B RK1828 w4a16)
Add Qwen3-4B.tokenizer.gguf (Qwen3-4B RK1828 w4a16)
feat(manifest): add ASR_ENCODER_SIZES to all qwen3_asr artifact sets
Update RK Qwen3 ASR optimized manifest
Add RK3576 Qwen3 ASR W8A8 decoder
Add RK Paraformer hybrid artifacts
Update RK artifact manifest
Add kokoro tokens.txt for RK slim image (#44)
Upload rk3588/kokoro-hybrid-v1/style.npy with huggingface_hub
Upload rk3588/kokoro-hybrid-v1/default.npy with huggingface_hub
P7b: bucket-8 tail-rest static MM+Gemm INT8 (audio PASS, perf -48% vs FP32 / -30% vs P7a dyn)
feat(kokoro): bucket-32 tail-rest INT8 (P7a)
feat(kokoro): bucket-16 tail-rest INT8 (P7a)
feat(kokoro): bucket-8 tail-rest INT8 (P7a)
Add kokoro bucket-16 4-stage RKNN/ONNX artifacts
Add kokoro bucket-8 4-stage RKNN/ONNX artifacts
Upload rk3588/kokoro-hybrid-v1/kokoro-vocoder-tail-rest-cpu.onnx with huggingface_hub
Upload rk3588/kokoro-hybrid-v1/rk3588/kokoro-vocoder-front-half.native.fp16.rknn with huggingface_hub
M2: add kokoro-bert-encoder.fp16.rknn (BERT subgraph FP16, RK3588, parity PASS_RELAXED)
add kokoro-decoder-front.int8.rknn (hybrid v1, seq32)
add kokoro-generator-tail-cpu.onnx (hybrid v1, seq32)
add kokoro-prefix-cpu.onnx (hybrid v1, seq32)
Add Paraformer RKNN hybrid artifacts
Upload rk_manifest.json with huggingface_hub
Upload shared/opt/asr/models/embd/decoder_hf.embed_tokens.npy with huggingface_hub
Remove stale RK host-path duplicates
Add RK deploy sidecars and fixed manifest
Add RK fixed-path Matcha ONNX
Add RK fixed-path Matcha ONNX
Update RK manifest fixed runtime paths
Upload RK generated artifacts 2026-05-17
initial commit