Textify

Third Party Notices

Current desktop dependencies and bundled notices are documented in electron/THIRD_PARTY_NOTICES.md.

The records below retain upstream attribution for the shared signed catalog and historical model evaluation. Runtime descriptions refer to that historical implementation unless listed in the desktop notices. They do not imply that the current app ships every runtime or exposes every catalog entry.

Textify does not bundle speech model artifacts. Curated models are downloaded from immutable Textify release assets or exact commit-pinned upstream files and verified against the signed model catalog before installation. User-imported Whisper models remain local and their licenses are not verified by Textify.

whisper.cpp

Textify’s desktop build downloads and verifies a pinned whisper.cpp source archive. The repository retains its license and historical vendor provenance.

• Upstream repository: https://github.com/ggml-org/whisper.cpp
• Upstream tag: v1.7.6
• Upstream commit: a8d002cfd879315632a579e73f0148d06959de36
• License: MIT
• Copied license text: THIRD_PARTY_LICENSES/whisper.cpp.txt
• Vendored license source: Vendor/whisper.cpp/LICENSE
• Vendor provenance: Vendor/whisper.cpp/UPSTREAM.md

The desktop worker is built with CMake. Model binaries remain external curated downloads and are not included in this repository snapshot.

OpenAI Whisper models

Whisper model assets are optional external downloads and are not included in the application bundle or this repository snapshot.

• Whisper small.en q5_1: mirrored as an immutable Textify release asset
• Whisper Large V2 q5_0 and Whisper Large V3 q5_0: exact commit-pinned GGML artifacts from https://huggingface.co/ggerganov/whisper.cpp
• Whisper Large V3 Turbo q5_0: exact commit-pinned GGML artifact from https://huggingface.co/ggerganov/whisper.cpp
• Whisper Large V3 Turbo MLX: exact commit-pinned safetensors artifact from https://huggingface.co/mlx-community/whisper-large-v3-turbo with immutable tokenizer assets from the original OpenAI repository
• Original models: https://huggingface.co/openai
• Model and conversion license: MIT
• Copied license texts: THIRD_PARTY_LICENSES/OpenAI-Whisper.txt, THIRD_PARTY_LICENSES/whisper.cpp.txt, and THIRD_PARTY_LICENSES/ggml-small.en-q5_1.LICENSES.txt
• Every artifact revision, byte size, and checksum is pinned by the signed Textify model catalog

CrisperWhisper 2.0 models and CrisperWhisper.cpp

CrisperWhisper model assets are optional external downloads and are not included in the application bundle or this repository snapshot.

• Textify catalog entries: crisperwhisper-2-large-f16 and crisperwhisper-2-turbo-f16
• Original models: https://huggingface.co/nyralabs/CrisperWhisper2.0_large at fe94128aee7cdf1ad43ab59c4b1f7ed6362c1106 and https://huggingface.co/nyralabs/CrisperWhisper2.0_turbo at 16424a0a0dad3d7bcc08927451718da49bce8ce0
• GGML FP16 conversions: https://huggingface.co/drbaph/CrisperWhisper2.0-GGML at 4807002cb64c57f78b37ffb4c48e986815a7c885
• Native compatibility implementation: https://github.com/Saganaki22/CrisperWhisper.cpp at 13c7b3efdafd8bd20bd6361e7354ce2aa9bec464
• CrisperWhisper.cpp code license: MIT
• Copied CrisperWhisper.cpp license text: THIRD_PARTY_LICENSES/CrisperWhisper.cpp-LICENSE.txt
• Exact copied Nyra model terms: THIRD_PARTY_LICENSES/CrisperWhisper-2.0-Nyra-License.md
• Every selected artifact, revision, byte size, and checksum is pinned by the signed Textify model catalog.

Textify does not ship or link the CrisperWhisper.cpp runtime. Inference uses Textify’s existing pinned ggml-org whisper.cpp v1.7.6 Metal integration at commit a8d002cfd879315632a579e73f0148d06959de36. Focused decoder logic in The retired Swift app’s shim was adapted from the pinned CrisperWhisper.cpp implementation identified above.

The standard public CrisperWhisper 2.0 repositories publish noncommercial terms for the model weights and outputs generated by them; their software is covered separately by MIT terms. Those public upstream terms remain visible offline in Textify and are not replaced or relaxed by this notice. Textify’s release of these two catalog entries depends on its separate commercial grant from nyra health GmbH.

MLX Whisper Large V3 Turbo model

The MLX Whisper model assets are optional external downloads and are not included in the application bundle or this repository snapshot.

• MLX conversion: https://huggingface.co/mlx-community/whisper-large-v3-turbo
• Exact MLX revision: a4aaeec0636e6fef84abdcbe3544cb2bf7e9f6fb
• Original model and tokenizer: https://huggingface.co/openai/whisper-large-v3-turbo
• Exact tokenizer revision: 41f01f3fe87f28c78e2fbf8b568835947dd65ed9
• Original model and converted checkpoint license: MIT
• Every runtime artifact byte size and checksum is pinned by the signed Textify model catalog; all tokenizer files are installed locally before model load

FluidAudio

The historical runtime pinned FluidAudio 0.15.5 for Core ML speech-model loading and Parakeet and Paraformer inference on Apple Silicon.

• Upstream repository: https://github.com/FluidInference/FluidAudio
• Pinned version: 0.15.5
• Pinned revision: 19600a485baa4998812e4654b70d2bab8f2c9949
• License: Apache-2.0
• Copied license text: THIRD_PARTY_LICENSES/FluidAudio.txt
• Linked fastcluster notice: THIRD_PARTY_LICENSES/FluidAudio-fastcluster.txt
• Linked VBx notice: THIRD_PARTY_LICENSES/FluidAudio-VBx.txt

SwiftNIO and Swift Crypto

The historical runtime linked SwiftNIO Core 2.101.3 and Swift Crypto 4.5.1.

• SwiftNIO revision: 0b18836bd8b0162e7e17a995a3fbee20ed8f3b2b
• Swift Crypto revision: 47d3869a7291f085c1fb9fb1e6d3b97a793f45c6
• Exact copied notices: THIRD_PARTY_LICENSES/SwiftNIO-NOTICE.txt and THIRD_PARTY_LICENSES/SwiftCrypto-NOTICE.txt

NVIDIA Parakeet models

Parakeet model assets are optional external downloads and are not included in the application bundle or this repository snapshot.

• Parakeet TDT 0.6B V3: https://huggingface.co/nvidia/parakeet-tdt-0.6b-v3
• Parakeet TDT 0.6B V2: https://huggingface.co/nvidia/parakeet-tdt-0.6b-v2
• Parakeet TDT-CTC 110M: https://huggingface.co/nvidia/parakeet-tdt_ctc-110m
• Parakeet TDT-CTC 0.6B Japanese: https://huggingface.co/nvidia/parakeet-tdt_ctc-0.6b-ja
• Parakeet RNNT 1.1B: https://huggingface.co/nvidia/parakeet-rnnt-1.1b
• Core ML conversions: https://huggingface.co/FluidInference
• MLX RNNT conversion: https://huggingface.co/mlx-community/parakeet-rnnt-1.1b
• MLX TDT V2 conversion: https://huggingface.co/mlx-community/parakeet-tdt-0.6b-v2 at 8ae155301e23d820d82aa60d24817c900e69e487
• MLX TDT V3 conversion: https://huggingface.co/mlx-community/parakeet-tdt-0.6b-v3 at ed2b7e8c15f9aaa0b5772e2efb986255eaef7e15
• GGUF TDT V2 conversion: https://huggingface.co/handy-computer/parakeet-tdt-0.6b-v2-gguf at 07cee0616125a08ef619729bb47f40ef747e4bc4
• GGUF TDT V3 conversion: https://huggingface.co/handy-computer/parakeet-tdt-0.6b-v3-gguf at 85ac09ea12fc4b1112fa76810059364bc6adc9de
• Model license: CC-BY-4.0 for all catalog entries
• Copied model license text: THIRD_PARTY_LICENSES/CC-BY-4.0.txt
• Every runtime artifact revision and file checksum is pinned by the signed Textify model catalog

NVIDIA Canary-Qwen model

Canary-Qwen model assets are optional external downloads and are not included in the application bundle or this repository snapshot.

• Original model: https://huggingface.co/nvidia/canary-qwen-2.5b
• Q4_K_M GGUF conversion: https://huggingface.co/handy-computer/canary-qwen-2.5b-gguf at 3370d4e2f28cc70eea79dfc9f2f43fb91eef3163
• Canary-Qwen model weights and conversion: CC-BY-4.0
• Copied model license text: THIRD_PARTY_LICENSES/CC-BY-4.0.txt
• Qwen 2.5 language-model component: Apache-2.0
• transcribe.cpp runtime and conversion tooling: MIT
• Every selected artifact, revision, byte size, and checksum is pinned by the signed Textify model catalog

MLX and MLX Audio Swift

Textify pins MLX Audio Swift at commit d302a5c6080d2bb97bae38c7418f82abb76013b6 and MLX Swift 0.31.3 at commit 61b9e011e09a62b489f6bd647958f1555bdf2896 for eligible local Metal speech inference. Textify calls the local-directory model loader and does not use the package’s network-backed pretrained-model loader.

• MLX Audio Swift: https://github.com/Blaizzy/mlx-audio-swift (MIT)
• MLX Swift: https://github.com/ml-explore/mlx-swift (MIT)
• MLX Swift LM: https://github.com/ml-explore/mlx-swift-lm (MIT)
• Exact historical direct and transitive revisions remain in Git history.
• Exact copied license texts for MLX, Hugging Face Swift, Swift project, EventSource, and yyjson dependencies: THIRD_PARTY_LICENSES/
• Linked MLX Swift nested component texts: THIRD_PARTY_LICENSES/MLXSwift-fmt.txt, THIRD_PARTY_LICENSES/MLXSwift-nlohmann-json.txt, THIRD_PARTY_LICENSES/MLXSwift-metal-cpp.txt, and THIRD_PARTY_LICENSES/MLXSwift-MLX-Core.txt

LiteRT-LM

The historical runtime linked a pinned arm64 slice of Google’s LiteRT-LM for eligible local model paths.

• Upstream repository: https://github.com/google-ai-edge/LiteRT-LM
• Pinned release: v0.14.0
• SwiftPM binary artifact: https://github.com/google-ai-edge/LiteRT-LM/releases/download/v0.14.0/CLiteRTLM_mac.xcframework.zip
• SwiftPM artifact checksum: 450615483509aaa6d34b321fdc6862e41a224b674468ab10aff64ebe113d21b7
• License: Apache-2.0
• Copied license text: THIRD_PARTY_LICENSES/LiteRT-LM.txt

MossFormer2 SE voice-cleaning models

MossFormer2 SE model assets are optional external downloads and are not included in the application bundle or this repository snapshot.

• FP32 conversion: https://huggingface.co/starkdmi/MossFormer2-SE at 8744c59f925154f4ba2e9f15ae7eeaa870f80118
• FP16 conversion: https://huggingface.co/starkdmi/MossFormer2-SE-fp16 at dd04b1b736b9f49951433b7f051cd8d32eb024b6
• 8-bit conversion: https://huggingface.co/starkdmi/MossFormer2-SE-8bit at 694e69b58f2457e02d96f4ba7fa151a28b07805a
• Original model: https://huggingface.co/alibabasglab/MossFormer2_SE_48K
• Converted checkpoint license: Apache-2.0
• Every config and weight revision, byte size, and checksum is pinned by the signed Textify model catalog

Qwen3-ASR models

Qwen3-ASR model assets are optional external downloads and are not included in the application bundle or this repository snapshot.

• Original models: https://huggingface.co/Qwen/Qwen3-ASR-0.6B and https://huggingface.co/Qwen/Qwen3-ASR-1.7B
• MLX 8-bit conversions: https://huggingface.co/mlx-community/Qwen3-ASR-0.6B-8bit at 89e96d92ba34aca20b3e29fb10cc284097d1219f and https://huggingface.co/mlx-community/Qwen3-ASR-1.7B-8bit at a8379a2e2f9e313c9292cdf1af4055ab56d50d55
• GGUF BF16, Q8_0, and Q5_K_M conversions: https://huggingface.co/handy-computer/Qwen3-ASR-0.6B-gguf at e4e16599b900eb0cb36e524514756bb92eb092b7 and https://huggingface.co/handy-computer/Qwen3-ASR-1.7B-gguf at 92282af1610a2db19d66f2bef1e260f5deca782d
• Original and converted checkpoint license: Apache-2.0
• MLX Audio Swift, MLX Swift, and transcribe.cpp runtime licenses: MIT
• Every selected runtime artifact, revision, byte size, and checksum is pinned by the signed Textify model catalog

NVIDIA Nemotron 3.5 ASR model

Nemotron model assets are optional external downloads and are not included in the application bundle or this repository snapshot.

• Original model: https://huggingface.co/nvidia/nemotron-3.5-asr-streaming-0.6b
• Attribution: Licensed by NVIDIA Corporation under the NVIDIA Open Model License.
• MLX conversion: https://huggingface.co/mlx-community/nemotron-3.5-asr-streaming-0.6b at e550040c0478027ed679b2b6b0d055502c103663
• GGUF conversion: https://huggingface.co/handy-computer/nemotron-3.5-asr-streaming-0.6b-gguf at 6d44e540bc31b0de1dbe174a3cea87f53a7f22fb
• The exact pinned MLX source card declares the NVIDIA Open Model License; the exact pinned GGUF source card declares OpenMDW-1.1. The signed catalog records the license attached to each selected source.
• Copied model license texts: THIRD_PARTY_LICENSES/NVIDIA_Open_Model_License.txt and THIRD_PARTY_LICENSES/OpenMDW-1.1.txt
• The GGUF repository publishes F16, Q8_0, and Q5_K_M, not BF16.
• Textify uses batch transcription only and does not expose upstream streaming partials.
• Every selected artifact, byte size, SHA-256, and source revision is pinned by the signed Textify model catalog.

Cohere Transcribe 03-2026 model

Cohere Transcribe model assets are optional external downloads and are not included in the application bundle or this repository snapshot.

• Original model: https://huggingface.co/CohereLabs/cohere-transcribe-03-2026
• 8-bit MLX conversion: https://huggingface.co/beshkenadze/cohere-transcribe-03-2026-mlx-8bit
• Exact MLX revision: d1f843476f84846e6fe7aa58a6033f17882f0ec9
• Original model and conversion license: Apache-2.0
• Every runtime artifact byte size and checksum is pinned by the signed Textify model catalog

Paraformer-large Chinese model

Paraformer model assets are optional external downloads and are not included in the application bundle or this repository snapshot.

• Original model: https://modelscope.cn/models/iic/speech_paraformer-large_asr_nat-zh-cn-16k-common-vocab8404-pytorch
• Core ML conversion: https://huggingface.co/FluidInference/paraformer-large-zh-coreml
• Upstream model license: Apache-2.0
• Runtime artifact revision pinned by the signed Textify model catalog

sherpa-onnx and ONNX Runtime

Textify bundles the arm64 C runtime from sherpa-onnx 1.13.2 and its matching ONNX Runtime 1.24.4 dependency to run eligible ONNX speech models fully offline.

• sherpa-onnx repository: https://github.com/k2-fsa/sherpa-onnx
• sherpa-onnx commit: 13d0ae6c539d2809d32f5eaa3ef1db0c459d0b24
• sherpa-onnx license: Apache-2.0
• ONNX Runtime repository: https://github.com/microsoft/onnxruntime
• ONNX Runtime version: 1.24.4
• ONNX Runtime license: MIT
• Copied license texts: THIRD_PARTY_LICENSES/sherpa-onnx.txt, THIRD_PARTY_LICENSES/ONNX_Runtime.txt, and the complete THIRD_PARTY_LICENSES/ONNX_Runtime_ThirdPartyNotices.txt
• Exact release-asset and selected-file hashes: Vendor/sherpa-onnx/v1.13.2/UPSTREAM.md

Both libraries are copied to Contents/Frameworks and signed with the application. Textify verifies the sherpa-onnx runtime version before loading a model.

transcribe.cpp

Textify bundles a pinned arm64 build of transcribe.cpp 0.1.3 for eligible GGUF speech models. It is loaded locally and isolated from Textify’s embedded whisper.cpp symbols.

• Upstream repository: https://github.com/handy-computer/transcribe.cpp
• Upstream commit: 5a5a49664a8ea1f0e5b3be1dfc544730d1b62561
• Vendored ggml commit: 707321c4cf6d21cb4bc831aa8b687dbf01a521ce
• License: MIT
• Copied license texts: THIRD_PARTY_LICENSES/transcribe.cpp.txt, THIRD_PARTY_LICENSES/transcribe.cpp-ggml.txt, and THIRD_PARTY_LICENSES/transcribe.cpp-miniz.txt
• Local language-prompt compatibility patch and exact build provenance: Vendor/transcribe.cpp/v0.1.3/UPSTREAM.md

The runtime is built for arm64 macOS 14 with Metal and Accelerate. Textify requires an actual Metal model backend and rejects CPU fallback for catalog entries using this engine.

IBM Granite Speech models

Granite Speech model assets are optional external downloads and are not included in the application bundle or this repository snapshot.

• Original autoregressive model: https://huggingface.co/ibm-granite/granite-speech-4.1-2b
• Original non-autoregressive model: https://huggingface.co/ibm-granite/granite-speech-4.1-2b-nar
• Q5_K_M GGUF conversions: https://huggingface.co/handy-computer/granite-speech-4.1-2b-gguf at 58e7710fd7039ded5a185668eef5f71ca5d9d919 and https://huggingface.co/handy-computer/granite-speech-4.1-2b-nar-gguf at ca53e8273416eb7e888f19bcebbcb9b6ab3edc17
• Original models and converted checkpoints: Apache-2.0
• transcribe.cpp runtime and conversion tooling: MIT
• Every selected artifact, revision, byte size, and checksum is pinned by the signed Textify model catalog

Mistral Voxtral Mini model

Voxtral model assets are optional external downloads and are not included in the application bundle or this repository snapshot.

• Original model: https://huggingface.co/mistralai/Voxtral-Mini-4B-Realtime-2602
• Q4_K_M GGUF conversion: https://huggingface.co/handy-computer/Voxtral-Mini-4B-Realtime-2602-gguf at b3e1c979e3775cbd0a49a65878a0ec7f06789ed7
• Original model and converted checkpoint: Apache-2.0
• transcribe.cpp runtime and conversion tooling: MIT
• Every selected artifact, revision, byte size, and checksum is pinned by the signed Textify model catalog

OpenMOSS MOSS Transcribe-Diarize model

MOSS Transcribe-Diarize model assets are optional external downloads and are not included in the application bundle or this repository snapshot.

• Original model: https://huggingface.co/OpenMOSS-Team/MOSS-Transcribe-Diarize
• Q5_K_M GGUF conversion: https://huggingface.co/handy-computer/MOSS-Transcribe-Diarize-gguf at 6fdfa33aed776bbb0ac11a1a9835634fe6d75dd7
• Original model and converted checkpoint: Apache-2.0
• transcribe.cpp runtime and conversion tooling: MIT
• Every selected artifact, revision, byte size, and checksum is pinned by the signed Textify model catalog

Fun-ASR MLT-Nano model

Fun-ASR MLT-Nano model assets are optional external downloads and are not included in the application bundle or this repository snapshot.

• Original model: https://huggingface.co/FunAudioLLM/Fun-ASR-MLT-Nano-2512
• Original model revision: 3dd802bfb8a05dadc440c9d9fc84ecfc1e15c0ff
• GGUF conversion: https://huggingface.co/handy-computer/Fun-ASR-MLT-Nano-2512-gguf
• GGUF revision: 0b8f9c7bc545a219658aeb1dd4eeaa55d1cf89f3
• Original model license: Apache-2.0
• Textify candidate artifact: Q8_0, with exact bytes pinned by the signed model catalog when promoted

Textify does not claim the checkpoint’s full published language list. Only language routes that pass Textify’s own Apple Silicon quality, latency, and robustness gates may appear in the catalog.

ReazonSpeech K2 V2 model

ReazonSpeech model assets are optional external downloads and are not included in the application bundle or this repository snapshot.

• Model: https://huggingface.co/reazon-research/reazonspeech-k2-v2
• Exact ONNX revision: 291488c8151be24d7da4bf7af26e533fad96e407
• Model license: Apache-2.0
• Every runtime artifact byte size and checksum is pinned by the signed Textify model catalog

SenseVoiceSmall model

SenseVoiceSmall model assets are optional external downloads and are not included in the application bundle or this repository snapshot.

• Original model: https://huggingface.co/FunAudioLLM/SenseVoiceSmall
• Int8 ONNX conversion: https://huggingface.co/csukuangfj/sherpa-onnx-sense-voice-zh-en-ja-ko-yue-2024-07-17
• Exact ONNX revision: 2365baeacb507f821a0c8120fcee3d484dba7a07
• Model license: FunASR Model Open Source License Agreement 1.1
• Required attribution: FunASR and SenseVoice; the SenseVoiceSmall model name is retained in Textify’s catalog and user interface
• Copied model license text: THIRD_PARTY_LICENSES/FunASR_Model_License_1.1.txt
• Every runtime artifact byte size and checksum is pinned by the signed Textify model catalog

Confucius4-R2T2 and audio.cpp

• Original model: https://huggingface.co/netease-youdao/Confucius4-R2T2
• Q8_0 and F16 conversions: https://huggingface.co/davidxifeng/Confucius4-R2T2-gguf, revision a8e6b385d7df7eae9519363e07034a209004797a.
• Runtime: https://github.com/0xShug0/audio.cpp, commit 9ba884179826c3b33dd305185b5f94c79175a03d.
• Model terms, controlling Chinese license, and conversion notice: THIRD_PARTY_LICENSES/Confucius4-R2T2.txt.
• Runtime and compiled dependency licenses: THIRD_PARTY_LICENSES/audio.cpp.txt (audio.cpp, ggml, cJSON, libyaml, SentencePiece and its included dependencies).

Any modifications made to the original model in this Derivative Work are not endorsed, warranted, or guaranteed by the original right-holder of the original model, and the original right-holder disclaims all liability related to this Derivative Work.

This site is open source. Improve this page.