soloterm/koko is the Solo-maintained fork of WismutHansen/kokorox, itself based on lucasjinreal/Kokoros.
This repository is the source of truth for koko binaries distributed to Solo users. Solo builds, signs, and publishes all binaries from this repo.
- Original Kokoro/Kokoros work by Lucas Jin and contributors
- Rust fork and major extensions by the kokorox maintainers
- Ongoing distribution/maintenance in this fork by Solo Terminal, Inc.
This fork is licensed under the GNU General Public License v3.0 (GPL-3.0-only). See LICENSE.
For third-party attribution details, see NOTICE and tracked local modifications in CHANGES.md.
Prerequisites:
- Rust stable toolchain
pkg-config- Opus development libraries
Examples:
# macOS
brew install pkg-config opus
# Ubuntu/Debian
sudo apt-get update
sudo apt-get install -y pkg-config libopus-dev libasound2-devBuild:
cargo build --release# Generate speech from text
./target/release/koko text "Hello world" -o test.wav
# List voices as JSON
./target/release/koko voices --format json
# Stream from stdin
printf "Hello from stdin\n" | ./target/release/koko stream > output.wav
# Pipe mode
printf "Hello from pipe mode\n" | ./target/release/koko pipe -o pipe.wavRelease workflows produce per-platform archives containing:
koko/koko.exeLICENSENOTICEespeak-ng-data/alongsidekokoin the macOS archive, copied from theespeak-rs-sysbuild so it matches the statically linked eSpeak library
Checksums are published alongside artifacts in GitHub Releases.
The macOS archive extracts to this layout:
koko
espeak-ng-data/
phontab
...
LICENSE
NOTICE
Keep koko and espeak-ng-data/ together when installing the macOS binary.
At startup, koko uses the data beside its canonical executable path. You can
instead set PIPER_ESPEAKNG_DATA_DIRECTORY to the parent of
espeak-ng-data/; ESPEAK_DATA_PATH also accepts that parent or the data
directory itself. koko exits with an error when it cannot find phontab.
Model files are distributed separately. just model-assets-prepare downloads
kokoro-v1.0.onnx and voices-v1.0.bin from the
model-files-v1.0 upstream release
to target/model-v1.0/, verifies their SHA-256 hashes, and writes SHA256SUMS.
just model-assets-release performs the same verification and uploads those
three files to a draft model-v1.0 release. The model and voices are Apache-2.0;
see NOTICE.
Use manual upstream review; do not auto-merge upstream into main.
# Review upstream drift
just upstream-check
# Create a dated sync branch from main
just upstream-sync-branch
# If your shipping remote is not 'origin'
SHIP_REMOTE=soloterm just upstream-sync-branch
# Cherry-pick reviewed commits onto upstream-sync/*
just upstream-pick <sha> [<sha> ...]The Upstream Report GitHub workflow runs monthly (and on manual dispatch) to report upstream commit drift and update a tracking issue. It does not modify code.