Last updated:
Open Source Licenses
Applies to: LYR (Android and iOS). In addition to frameworks provided by Apple, the iOS app uses the libraries listed under “iOS” in 1. The server-side models in 4 and 5 are shared by both versions
LYR uses the software, fonts, and machine learning models listed below. Copyright in each remains with its respective holder, and we use each item under the terms of its license.
1. Software libraries
Android
| Name | License | Source |
|---|---|---|
| ONNX Runtime | MIT License | microsoft/onnxruntime |
| OkHttp | Apache License 2.0 | square/okhttp |
| Gson | Apache License 2.0 | google/gson |
| ICU4J | Unicode License | unicode-org/icu |
| AndroidX / Jetpack Compose / Media3 | Apache License 2.0 | Android Open Source Project |
| Kotlin / kotlinx.coroutines | Apache License 2.0 | JetBrains/kotlin |
| Okio | Apache License 2.0 | square/okio |
| Guava | Apache License 2.0 | google/guava |
| Checker Framework (annotations) | MIT License | typetools/checker-framework |
| Error Prone (annotations) | Apache License 2.0 | google/error-prone |
| J2ObjC (annotations) | Apache License 2.0 | google/j2objc |
| JSR 305 (annotations) | Apache License 2.0 | findbugsproject/findbugs |
| javax.inject | Apache License 2.0 | javax-inject/javax-inject |
iOS
| Name | License | Source |
|---|---|---|
| Firebase Apple SDK (FirebaseCore / FirebaseCoreInternal / FirebaseInstallations / FirebaseAnalytics) | Apache License 2.0 | firebase/firebase-ios-sdk |
| GoogleUtilities | Apache License 2.0 | google/GoogleUtilities |
| Promises (FBLPromises) | Apache License 2.0 | google/promises |
| nanopb | zlib License | nanopb/nanopb |
| RevenueCat (purchases-ios) | MIT License | RevenueCat/purchases-ios |
Google App Measurement (the measurement part of Firebase Analytics) is distributed as a binary and is used under the Google APIs Terms of Service. Full text of the Apache License 2.0: https://www.apache.org/licenses/LICENSE-2.0
Common
In addition to the above, the app includes components provided under the Apache License 2.0 as transitive dependencies of these libraries. Please contact us if you need the individual names and full license texts.
Google Play services, Firebase and the Google Play Billing Library are used under the Google APIs Terms of Service and the terms of each service.
2. Fonts
| Name | License | Source |
|---|---|---|
| Bangers | SIL Open Font License 1.1 | googlefonts/bangers |
| Comic Neue | SIL Open Font License 1.1 | crozynski/comicneue |
The full license texts are bundled with the app.
3. On-device models, including optical character recognition (OCR)
PaddleOCR (PP-OCRv5) The detection, recognition and angle classification models derive from PaddleOCR. Some recognition models (for example, for vertical text) were further trained by us starting from PaddleOCR models. License: Apache License 2.0 / Source: PaddlePaddle/PaddleOCR
Speech bubble detection model The model that detects speech bubbles in manga was further trained by us on synthetic data only, using the YOLOX training source code and starting from the pretrained YOLOX-Nano weights. YOLOX license: Apache License 2.0 / Source: Megvii-BaseDetection/YOLOX
Silero VAD (voice activity detection) In Audio mode, this model detects which parts of the audio are speech, on the device (Android and iOS). We use the weights of silero-vad 6.2.3. The iOS version uses these weights converted by us to the Core ML format (the weights themselves are not modified or further trained). License: MIT License (Copyright (c) 2020-present Silero Team) / Source: snakers4/silero-vad
Other on-device models (for example, for tracking text regions) were designed and trained by us.
4. Speech recognition (ASR)
Speech recognition for Audio mode runs on servers operated by the Developer. The current app does not contain any speech recognition model (see the Privacy Policy for what is sent).
The servers use the following models and software. They are not distributed as part of the app, but are listed for reference.
| Name | License | Source |
|---|---|---|
| Whisper large-v3 (OpenAI) | MIT License | openai/whisper |
| faster-whisper | MIT License | SYSTRAN/faster-whisper |
| CTranslate2 | MIT License | OpenNMT/CTranslate2 |
5. Translation
Translation is performed either by a translation model on servers operated by the Developer or by an external translation API. The server-side translation model uses part of Qwen3.6-27B (Qwen team, Alibaba Cloud) to read the input, with the part that generates translations trained by us. It is not distributed as part of the app, but is listed for reference. Qwen license: Apache License 2.0 / Source: Qwen/Qwen3.6-27B
Attribution
The current app contains no components licensed under Creative Commons Attribution.
Included in earlier versions
Earlier versions of the app (versionCode 5–8, versions 0.2.0–0.2.3, distributed for testing) performed speech recognition for Audio mode on the device, downloading and using the speech recognition models below. We keep the attribution for anyone who still has those versions.
| Language | Model | License | Source |
|---|---|---|---|
| Japanese | Trained and owned by us | — | Training data: ReazonSpeech |
| English | NVIDIA NeMo FastConformer | CC BY 4.0 | nvidia/stt_en_fastconformer_hybrid_large_streaming_multi |
| Korean | Trained and owned by us | — | Training data: YODAS (CC BY 3.0) |
| Portuguese | Trained and owned by us | — | Training data: YODAS2 (CC BY 3.0) |
The Japanese model is trained and owned by us. Training was carried out as information analysis under Article 30-4 of the Japanese Copyright Act (use not intended for the enjoyment of the thoughts or sentiments expressed in the work).
The Korean and Portuguese models are likewise trained and owned by us. Because their training data (YODAS / YODAS2) is distributed under CC BY 3.0, attribution is given below. The Korean and Portuguese models were also trained with the MUSAN corpus for noise robustness. MUSAN is distributed under CC BY 4.0, so it is likewise credited below. Those versions used sherpa-onnx (Apache License 2.0, k2-fsa/sherpa-onnx) to run recognition.
English speech recognition model
- Author: NVIDIA Corporation
- Title:
stt_en_fastconformer_hybrid_large_streaming_multi - Source: https://huggingface.co/nvidia/stt_en_fastconformer_hybrid_large_streaming_multi
- License: Creative Commons Attribution 4.0 International (CC BY 4.0)
- Changes: We converted the model to ONNX format and applied int8 quantization. No changes were made to the model weights themselves.
Training data for the Korean speech recognition model
- Author: Xinjian Li et al. (YODAS)
- Title: YODAS (
ko000= manual-caption subset) - Source: https://huggingface.co/datasets/espnet/yodas
- License: Creative Commons Attribution 3.0 (CC BY 3.0)
- Changes: We used this dataset to train a speech recognition model. The dataset itself is not redistributed. The resulting model is owned by us.
Training data for the Portuguese speech recognition model
- Author: Xinjian Li et al. (YODAS)
- Title: YODAS2 (
pt000= manual-caption subset) - Source: https://huggingface.co/datasets/espnet/yodas2
- License: Creative Commons Attribution 3.0 (CC BY 3.0)
- Changes: We used this dataset to train a speech recognition model. The dataset itself is not redistributed. The resulting model is owned by us.
Noise data used to train the Korean and Portuguese speech recognition models
- Author: David Snyder, Guoguo Chen, Daniel Povey
- Title: MUSAN: A Music, Speech, and Noise Corpus (arXiv:1510.08484)
- Source: https://www.openslr.org/17/
- License: Creative Commons Attribution 4.0 International (CC BY 4.0)
- Changes: We used the noise portion of this corpus for noise augmentation during training. The corpus itself is not redistributed.
6. Artwork and trademarks on the website (lyr.jp)
The website lyr.jp uses the following artwork and names to indicate supported operating systems.
Android robot
- Where it is used: The icon indicating the Android version on the lyr.jp home page and the early tester page
- Attribution: The Android robot is reproduced or modified from work created and shared by Google and used according to terms described in the Creative Commons 3.0 Attribution License.
- License: Creative Commons Attribution 3.0 Unported (CC BY 3.0)
- Changes: Only the head is redrawn as a single-color shape
Trademarks
- Android is a trademark of Google LLC.
- iPhone is a trademark of Apple Inc., registered in the U.S. and other countries.
- App Store is a service mark of Apple Inc. Google Play is a trademark of Google LLC.
The Service is not provided or endorsed by Google LLC or Apple Inc.
Contact
If you find an error or omission on this page, please let us know and we will review and correct it promptly.
Revision history
- 2026-11-04: Reflects using Silero VAD to detect speech in Audio mode, and adds it to “3. On-device models, including optical character recognition (OCR)”
- 2026-10-29: Added “6. Artwork and trademarks on the website (lyr.jp)”, with the attribution for the Android robot and trademark notices for Android, iPhone and others
- 2026-10-27: Reflects adding Firebase Analytics and RevenueCat to the iOS version, and adds the list for the iOS version to “1. Software libraries”
- 2026-10-26: Reflects moving speech recognition for Audio mode to servers and no longer distributing on-device speech recognition models or sherpa-onnx. Server-side models are listed for reference, and the attribution for speech recognition models distributed in earlier versions has moved to “Included in earlier versions”. Describes accurately how the speech bubble detection model was trained, and adds the iOS version.
- 2026-08-06: Initial version