mattmirelesCommunity·23 Agent Skills Included

kokoro-coreml

On-device text-to-speech on Apple Neural Engine

Converts the Kokoro 82M text-to-speech model into Core ML packages that run natively on the Apple Neural Engine. Splits the pipeline so dense math runs on the Neural Engine while variable-length logic stays on CPU, avoiding silent fallbacks and slow GPU round-trips. Includes export scripts, a Swift inference package, benchmarks, and audio-quality validation to ship fast on-device speech.

57|18

↗

npx skills add mattmireles/kokoro-coreml --all -g -y

Available:

Gives the agent operating rules for this repo: prioritize simplicity, route Core ML conversion and validation work to the right sub-skills, and require measured evidence before claiming Neural Engine performance.

All Skills in This Repository (23)

Pure Emerald Level Indicators

📦 In Repo

mattmireles

git-commit

Automates git commits and optionally audits projects with cucumber-style tests and linting.

documentation

Generate and review Python docstrings to enforce inline documentation standards.

phase-audit

Audit a completed plan phase against active plans and changed files.

ilya-sutskever

Convert PyTorch models to Core ML for Apple Neural Engine optimization.

audio-judge

Assess text-to-speech audio quality for intelligibility and acceptability.

write-notes

Create and update notes in a repository's README/Notes/ directory.

execute-plan

Execute implementation plans phase by phase with audits and git commits.

deploy

Coordinate release engineering for kokoro-coreml with Git operations and CI/CD practices.

markdown

Edit and standardize markdown files in a code repository.

coreml-profile

Analyzes CoreML model execution on Apple Silicon to detect compute units, silent fallbacks, and bottlenecks.

create-plan

Create structured implementation plans with phases, verification steps, and dependencies.

audit-fix-loop

Run audit cycles, fix issues until criteria are met, then commit via git-commit.

Community

Advanced

Frequently Asked Questions

FAQPage Schema

How to install kokoro-coreml?▼

Run `npx skills add mattmireles/kokoro-coreml --all -g -y` in your terminal to install all skills in this suite globally.

How to run Kokoro TTS on the Apple Neural Engine?▼

This repo splits the Kokoro pipeline so dense decoder and vocoder math runs as fixed-shape Core ML models on the Neural Engine, while variable-length duration logic stays on CPU. The included Swift package chains five compiled models with zero Python at inference time.

How fast is Kokoro on Apple Silicon?▼

An M1 Mac Mini synthesizes 15 seconds of speech in about 691 ms, roughly 22x realtime, and an M2 Ultra finishes 30 seconds of audio in 422 ms. That is 1.8 to 3.4x faster than PyTorch on the GPU.

Does kokoro-coreml work with Claude Code and Cursor?▼

Yes. All skills follow the universal SKILL.md standard and run in Claude Code, Cursor, Codex, and other compatible agents.

Can I use pre-converted Kokoro Core ML models?▼

Yes. Pre-converted .mlpackage files are published on Hugging Face at mattmireles/kokoro-coreml, so you can skip the export scripts and go straight to the Swift pipeline.