The Kokoro-82M text-to-speech model converted to Core ML in seven stages (Albert, PostAlbert, Alignment, Prosody, Noise, Vocoder, Tail), for on-device speech on iPhones and iPads with a Neural Engine (A12 and later, iOS 17+).
manifest.json lists every file with its size and SHA-256; the app downloads,
verifies and compiles the stages on the device.These files are unmodified conversions of the original weights and carry the original Apache 2.0 licence.
KokoroVocoder.mlpackage/Data/com.apple.CoreML/weights/weight.bin 49 MBKokoroPostAlbert.mlpackage/Data/com.apple.CoreML/weights/weight.bin 14 MBKokoroProsody.mlpackage/Data/com.apple.CoreML/weights/weight.bin 8 MBKokoroAlbert.mlpackage/Data/com.apple.CoreML/weights/weight.bin 6 MBKokoroNoise.mlpackage/Data/com.apple.CoreML/weights/weight.bin 5 MBKokoroTail.mlpackage/Data/com.apple.CoreML/weights/weight.bin 81.1 kBKokoroAlignment.mlpackage/Data/com.apple.CoreML/weights/weight.bin 4.1 kBhexgrad/Kokoro-82M