CriCodecs and CriStudio

CriStudio is the desktop interface for inspecting, previewing, extracting, editing, and building CRI middleware files. It is powered by CriCodecs, a modern C++23 codec and container library with a standalone CLI and Python bindings built with nanobind.

Most users should download CriStudio. The CLI, Python package, and C++ SDK are available for automation, scripting, and application integration.

Inspect and play muxed VP9 and ADX streams directly from a USM container.

Browse nested CPK entries and inspect structured binary data without extracting it first.

Download

The latest GitHub release provides platform-specific downloads:

Download	Intended use
CriStudio portable	Recommended. Extract and run; Qt and FFmpeg are included.
CriStudio slim	Smaller advanced-user build; requires compatible Qt and FFmpeg on the system.
CriCodecs CLI	Optimized command-line executable for scripts and batch work.
C++ SDK	Headers, CMake package metadata, and static/shared libraries.
Python wheels	Install from PyPI with python -m pip install cricodecs.

Portable CriStudio is non-installer software. Windows packages expose only CriStudio.exe and an _internal runtime directory; Linux uses an AppImage; macOS uses a self-contained application bundle.

CriStudio

CriStudio provides two complementary workspaces:

• Browse loads files and archives, searches nested entries, previews audio, video, images, tables, and raw bytes, and extracts selected content.
• Editor exposes format-aware inspection and mutation tools, archive entry management, build workflows, media muxing, encryption controls, and local key handling without changing the global CRI key implicitly.

Key recovery is available for supported HCA, ADX, AHX, USM, AWB, and ACB inputs. Multi-file recovery runs in the background and ranks a bounded set of candidates rather than blocking the interface.

Format coverage

Area	Formats
Audio and audio containers	ADX, AHX, HCA, WAV/PCM, AAX, AIX
Sound banks and tables	ACB, AWB, ACX, CSB, UTF
Archives and disc containers	AFS, CPK, CVM/ROFS
Video and stream containers	USM, SFD/SofDec

Available operations differ by format. CriStudio only exposes editing, encoding, muxing, or key-recovery actions where the corresponding native API supports them.

Detailed CLI, Python, and C++ examples are in USAGE.md.

Installation

Requirements

• CMake 4.2 or newer
• A recent C++23 compiler
  • GCC 16.1 or newer on Linux
  • Visual Studio 2026 / MSVC v145 or newer on Windows
  • A current Apple Clang/Xcode toolchain on macOS
• Python 3.9 or newer for Python bindings

Use a fresh build directory for each generator and compiler. If you change compilers, generators, Python versions, or CMake versions, delete the old build directory first.

Python From Source

From the repository root:

python -m pip install .

For verbose build logs:

python -m pip install -v .

For local development with an already prepared build environment:

python -m pip install --no-build-isolation -ve .

Python From A Wheel

Download a wheel matching your Python version, operating system, and CPU architecture, then install it directly:

python -m pip install cricodecs-1.1.0-*.whl

C++ Core Library

Build and install the optimized static C++ library:

cmake -S . -B build -G Ninja \
  -DCMAKE_BUILD_TYPE=Release \
  -DCRICODECS_BUILD_TESTS=OFF \
  -DCRICODECS_BUILD_CLI=OFF \
  -DCRICODECS_INSTALL_CPP=ON
cmake --build build --parallel 4
cmake --install build --prefix ./install

Set -DBUILD_SHARED_LIBS=ON to build and install the shared-library variant. Both variants install the same headers and CMake package metadata.

Consume an installed package with:

find_package(CriCodecs CONFIG REQUIRED)
target_link_libraries(my_app PRIVATE CriCodecs::CriCodecs)

Build the C++ library and Python bindings with CMake:

cmake -S . -B build-python -G Ninja -DCMAKE_BUILD_TYPE=Release -DCRICODECS_BUILD_PYTHON=ON
cmake --build build-python --parallel 4

On Windows with Visual Studio 2026:

cmake -S . -B build-vs -G "Visual Studio 18 2026" -A x64 -DCRICODECS_BUILD_PYTHON=OFF
cmake --build build-vs --config Release --parallel 4

If the Visual Studio generator is missing, update CMake first and run cmake -G to list the generators available on your machine.

CLI

Release builds provide a standalone cricodecs executable. Python installs also provide a cricodecs console command backed by the same native implementation.

Common usage is input-driven:

cricodecs input.hca
cricodecs input.cpk

Audio inputs such as ADX, AHX, HCA, and AAX decode to WAV by default. Archive and container inputs extract to a sibling directory by default. Use -m for metadata, -m --json for JSON metadata, -f to force the input format, and -o to choose the output path or extraction root.

Recover the effective low-56 key used by HCA cipher type 56 with one or more HCA files, directly from AWB/ACB/USM containers, or recursively from folders:

cricodecs --recover-key -f hca music.hca
cricodecs --recover-key -f hca cue_01.hca cue_02.hca --json
cricodecs --recover-key -f hca music.awb
cricodecs --recover-key -f hca music.acb
cricodecs --recover-key -f hca movie.usm
cricodecs --recover-key -f hca audio_folder

Multiple inputs are assumed to use the same effective table and compatible HCA frame grammar. Folder scans ignore unrelated files and pool every cipher-56 HCA found in HCA, AWB, ACB, or USM inputs. Every matching audio channel in a multi-audio USM is included. Recovery always returns the best guess and its validation score; a low score is not treated as a command failure.

Recover the effective low-56 USM stream-mask key with -f usm:

cricodecs --recover-key -f usm movie.usm
cricodecs --recover-key -f usm movie_folder --json

Each USM is recovered independently because separate containers may use different keys. Recovery can use video, encrypted HCA, or USM-masked ADX evidence. Text and JSON results include the input path, key, score, sampled video blocks, and contributing video/ADX chunk counts. Recovery reports the candidate without applying it or rewriting the container.

Recover standalone ADX or AHX encryption triplets with their explicit format domains:

cricodecs --recover-key -f adx music.adx
cricodecs --recover-key -f adx same_key_adx_folder --json
cricodecs --recover-key -f ahx voice_a.ahx voice_b.ahx

Multiple ADX or AHX inputs are pooled under the contract that they use the same encryption type and effective triplet. Results include the directly usable start,mult,add triplet, structural score, and frame evidence. AHX output also keeps per-component candidate counts visible because sparse files can leave one component ambiguous.

-f adx targets ADX's own header-declared type-8/type-9 frame encryption. ADX carried inside a USM can additionally have the separate USM repeating audio mask; -f usm evaluates that audio-mask evidence together with SFV video and embedded HCA evidence.

Recover CRI's effective low-52 AAC key from encrypted M4A waveforms in an AWB or its ACB cue sheet with an explicit container domain:

cricodecs --recover-key -f awb BGM.awb
cricodecs --recover-key -f acb BGM.acb --json

The ACB path selects only EncodeType 19 waveforms. The AWB path selects only entries sharing the encrypted CRI M4A header pattern, so unrelated HCA, ADX, or other bank entries are not passed to the AAC solver. Recovery is unavailable when the selected input contains no encrypted AAC/M4A evidence.

When building a USM, --audio accepts repeatable ADX or HCA inputs. Supplying --key masks the video and ADX audio by default, converts plain HCA audio to cipher type 56, and preserves HCA that is already encrypted:

cricodecs --build -f usm --key 0x165CF4E2138F7BDA \
  --audio dialogue.adx --audio music.hca -o movie.usm movie.ivf

Python Usage

Import the package as cricodecs. The top-level cricodecs.load() helper detects supported formats from a path or bytes and returns the matching object. Individual modules also expose format-specific helpers such as load(), decode(), encode(), extract(), demux(), or save_bytes().

Top-Level Load

import cricodecs

obj = cricodecs.load("input.cpk")
print(type(obj))
print(repr(obj))

ADX Decode And Encode

from pathlib import Path

from cricodecs import adx

wav_bytes = adx.decode("input.adx")
Path("output.wav").write_bytes(wav_bytes)

encoded = adx.encode(Path("source.wav").read_bytes())
Path("output.adx").write_bytes(encoded)

HCA Decode, Encode, Encrypt

from pathlib import Path

from cricodecs import hca

wav_bytes = hca.decode("input.hca", keycode=0xCF222F1FE0748978)
Path("decoded.wav").write_bytes(wav_bytes)

config = hca.HcaEncodeConfig()
config.sample_rate = 48000
config.channel_count = 2
config.quality = hca.HcaQuality.HIGH

hca_bytes = hca.encode(Path("source.wav").read_bytes(), config)
Path("encoded.hca").write_bytes(hca_bytes)

encrypted = hca.encrypt(hca_bytes, cipher_type=56, keycode=0xCF222F1FE0748978)
Path("encrypted.hca").write_bytes(encrypted)

CPK Inspect, Extract, Build

from pathlib import Path

from cricodecs import cpk

archive = cpk.load("input.cpk")
print(archive.info())
archive.extract("input_extracted")

new_archive = cpk.create(cpk.CpkPreset.FILENAME)
new_archive.add_file("data/voice.adx", "voice/voice.adx")
new_archive.add_bytes(b"hello", "text/readme.txt")
Path("output.cpk").write_bytes(new_archive.save_bytes())

ACB/AWB And USM

from cricodecs import acb, usm

cue_sheet = acb.load("sound.acb")
print(cue_sheet.info())
cue_sheet.extract("sound")

movie = usm.load("movie.usm")
print(movie.info())
movie.extract("movie_streams")

config = usm.UsmMuxConfig(
    video_path="movie.264",
    audio_tracks=[usm.UsmMuxAudioTrack("movie.adx")],
    subtitle_tracks=[
        usm.UsmMuxSubtitleTrack(
            "subtitles_en.srt",
            language_id=0,
            format=usm.UsmSubtitleFormat.SRT,
        ),
        usm.UsmMuxSubtitleTrack(
            "subtitles_alt.ass",
            language_id=1,
            format=usm.UsmSubtitleFormat.ASS,
        ),
    ],
)
usm.mux(config, "movie_with_subtitles.usm")

USM SBT language ids are numeric subtitle slots. The checked SDK docs do not define a universal locale mapping; pass the intended slot with UsmMuxSubtitleTrack(..., language_id=N). ASS imports can also preserve per-cue ids from Dialogue names like lang5.

Objects and small data structs implement useful repr() output, so interactive inspection is intended to be practical:

from cricodecs import adx

print(repr(adx.AdxEncodeConfig()))

C++ Usage

Include the complete API and link the namespaced CMake target:

#include <cricodecs/cricodecs.hpp>

target_link_libraries(my_app PRIVATE CriCodecs::CriCodecs)

The source tree remains organized by CRI format under CriCodecs/src. Installed module headers preserve that organization under <cricodecs/...>, so a project can include only the formats it uses. Public APIs use value types and std::expected<T, std::string> for recoverable failures.

ADX

#include <filesystem>
#include <fstream>
#include <vector>

#include <cricodecs/adx/adx_codec.hpp>

int main() {
    auto adx = cricodecs::adx::Adx::load("input.adx");
    if (!adx) {
        return 1;
    }

    auto decoded = adx->decode();
    if (!decoded) {
        return 1;
    }

    const auto& pcm = decoded->pcm_data;
    (void)pcm;
}

HCA

#include <cstdint>
#include <vector>

#include <cricodecs/hca/hca_codec.hpp>

std::vector<int16_t> decode_hca(std::span<const uint8_t> bytes) {
    auto decoded = cricodecs::hca::decode(bytes, 0xCF222F1FE0748978ULL);
    if (!decoded) {
        return {};
    }
    return std::move(*decoded);
}

CPK

#include <cricodecs/cpk/cpk_container.hpp>

int main() {
    auto archive = cricodecs::cpk::Cpk::load("input.cpk");
    if (!archive) {
        return 1;
    }

    return archive->files().empty() ? 1 : 0;
}

Credits

CriCodecs builds on public knowledge and prior open-source work around CRI formats. Many thanks and credits to:

• vgmstream for HCA and CRI audio research.
• VGAudio for ADX and HCA codec behavior references.
• bnnm for CRI audio-format research and tooling.
• Nyagamon for CRI format research.