VoxCPM ANE 2 Voices
Preset voice caches for seba/VoxCPM2ANE-Preview / VoxCPMANE2.
The cache files live under caches/:
*.embed.npy: reference voice feature embeddings, NumPy little-endian float16, shape (T, 2048).
*.prompt.embed.npy: high-similarity prompt feature embeddings, NumPy little-endian float16, shape (T, 2048).
*.prompt.cond.npy: high-similarity prompt LocDiT condition, NumPy little-endian float16, shape (4, 64).
*.prompt.decode_context.npy: high-similarity VAE decoder warmup context, NumPy little-endian float16, shape (N, 4, 64).
*.lm_prefix.npz: LM prefix cache arrays; existing arrays are already float16.
*.txt: prompt transcription text for high-similarity mode.
These files were migrated from the caches/ folder previously bundled inside seba/VoxCPM2ANE-Preview.