aufklarer/Kokoro-82M-CoreML

Kokoro-82M CoreML

3-stage CoreML pipeline for Kokoro-82M text-to-speech, optimized for Apple Neural Engine. Requires iOS 18+ / macOS 15+.

Pipeline

Stage	Model	Input	Output	Size
1. Duration	duration.mlmodelc	Phoneme tokens + voice + speed	Durations, prosody features, text encoding	39 MB
2. Prosody	prosody.mlmodelc	Aligned prosody features + style	F0 (pitch) + noise predictions	17 MB
3. Decoder	decoder_*.mlmodelc	Aligned text + F0 + noise + style	24 kHz audio waveform	107 MB

Swift builds an alignment matrix between stages 1 and 2 from predicted durations.

Decoder Buckets

Bucket	Max Frames	Max Audio
decoder_5s	200	5.0s
decoder_10s	400	10.0s
decoder_15s	600	15.0s

Voices

54 preset voices across 10 languages: English (US/UK), Spanish, French, Hindi, Italian, Japanese, Korean, Portuguese, Chinese.

Usage

Welcome to Swift!

Subcommands:

swift build Build Swift packages swift package Create and work on packages swift run Run a program from a package swift test Run package tests swift repl Experiment with Swift code interactively

Use swift --version for Swift version information.

Use swift --help for descriptions of available options and flags.

Use swift help <subcommand> for more information about a subcommand.

Conversion

WARNING: Defaulting repo_id to hexgrad/Kokoro-82M. Pass repo_id='hexgrad/Kokoro-82M' to suppress this warning. Loaded Kokoro-82M (81.8M params)

3-Stage Verification: Reference audio: 42000 samples 3-stage audio: 117600 samples Diff: max=0.9962, mean=0.0493 Duration diff: max=12.0 PASS: 3-stage pipeline matches reference

=== Converting Duration Model === Phoneme buckets: [16, 32, 64, 128] Tracing...

License

• Model weights: Apache-2.0 (hexgrad/Kokoro-82M)
• CoreML conversion + Swift inference: Apache-2.0
• Dictionaries and G2P: Apache-2.0

---

• Guide: soniqo.audio/guides/kokoro
• Docs: soniqo.audio
• GitHub: soniqo/speech-swift