These are the open-source software and AI models FreeSoundTools uses. The full license texts are in the files linked below. Checked: October 25, 2026.
AI models downloaded on this device
Model files that AI tools download once are stored in this browser. See the size of each model here and delete the ones you don't need. The list is read only on this device.
| Name (version) | Used for | License | Copyright notice and full text |
|---|---|---|---|
| OpenAI Whisper | Speech recognition model | MIT | Copyright (c) 2022 OpenAI · Full text · https://github.com/openai/whisper |
| Xenova/whisper-tiny·base·small (ONNX conversions) | Speech recognition model files | Apache-2.0 (as stated on the model card) | https://huggingface.co/Xenova/whisper-small (original distributor: Hugging Face · the unchanged files are downloaded from the Dagochim family model storage, models.dagotools.com, with Hugging Face as a fallback) |
| Supertone Supertonic 3 | High-quality AI voice (text to speech) model files | BigScience Open RAIL-M (use restrictions apply) | https://huggingface.co/Supertone/supertonic-3 (original distributor: Hugging Face · the unchanged files are downloaded from the Dagochim family model storage, models.dagotools.com, with Hugging Face as a fallback) · License text · The use restrictions are reproduced in section 3 of the Terms. Inference code follows Supertone's official example (MIT). |
| onnx-community/Qwen3-0.6B-ONNX | AI English conversation model (WebGPU) | Apache-2.0 | Qwen3 (c) Alibaba Cloud · https://huggingface.co/onnx-community/Qwen3-0.6B-ONNX (original distributor: Hugging Face · the unchanged files are downloaded from the Dagochim family model storage, models.dagotools.com, with Hugging Face as a fallback) |
| onnx-community/Qwen2.5-0.5B-Instruct | AI English conversation model (without WebGPU) | Apache-2.0 | Qwen2.5 (c) Alibaba Cloud · https://huggingface.co/onnx-community/Qwen2.5-0.5B-Instruct (original distributor: Hugging Face · the unchanged files are downloaded from the Dagochim family model storage, models.dagotools.com, with Hugging Face as a fallback) |
| pyannote segmentation-3.0 (onnx-community/pyannote-segmentation-3.0) | Speaker separation in meeting notes | MIT | Copyright (c) 2023 CNRS (pyannote) · https://huggingface.co/onnx-community/pyannote-segmentation-3.0 (original distributor: Hugging Face · the unchanged files are downloaded from the Dagochim family model storage, models.dagotools.com, with Hugging Face as a fallback) |
| Transformers.js 4.3.0 | Running models | Apache-2.0 | Full text · https://github.com/huggingface/transformers.js · Self-hosted (only the runtime file path changed to this site) |
| Transformers.js 3.7.2 | Running the high-quality English conversation model | Apache-2.0 | Full text · Self-hosted (only the runtime file path changed to this site) |
| ONNX Runtime Web 1.31 (for Transformers.js 4.3.0) and 1.22.0 (for the AI voice) | Model runtime engine | MIT | Copyright (c) Microsoft Corporation · Full text · Runtime (wasm) files are also served from this site |
| RNNoise (@shiguredo/rnnoise-wasm 2025.1.5) | Voice noise reduction | Apache-2.0 (wrapper), RNNoise is BSD-style | Copyright 2021-2025 Takeru Ohta, Shiguredo Inc. / RNNoise: Copyright (c) 2007-2017, 2024 Jean-Marc Valin, (c) 2023 Amazon, (c) 2017 Mozilla, (c) 2005-2017 Xiph.Org Foundation, (c) 2003-2004 Mark Borgerding · Full text |
| FreeSoundTools MP3 encoder (our own, 1.0.0) | Saving MP3 files | Our own code, all rights reserved | Source · Huffman table data from go-mp3 (Apache-2.0, https://github.com/hajimehoshi/go-mp3) |
| fflate 0.8.2 | ZIP downloads, Word (DOCX) export | MIT | Copyright (c) 2023 Arjun Barrett · Full text · https://github.com/101arrowz/fflate · served unmodified from this site |
| mp4-muxer 5.1.3 | Packaging M4A/M4R (AAC) files | MIT | Copyright (c) 2023 Vanilagy · Full text · https://github.com/Vanilagy/mp4-muxer · AAC encoding uses only the codec already in your device and browser (WebCodecs) |
| Basic Pitch (Spotify) 1.0.1 | Audio→MIDI "Polyphonic (AI)" | Apache-2.0 | © Spotify AB · model files unmodified, served from this site · runtime: TensorFlow.js 3.21.0 (Apache-2.0, © Google LLC) · License · NOTICE |
| mp4box.js 0.5.2 | Copies the MP4/MOV video track without re-encoding (Noise removal, Studio sound) | BSD-3-Clause | © GPAC / Telecom Paris · License |
| Pretendard Variable 1.3.9 | Site font | SIL Open Font License 1.1 | Copyright (c) 2021, Kil Hyung-jin, with Reserved Font Name Pretendard · Full text · https://github.com/orioncactus/pretendard · official files served unmodified from this site |
| Inter Variable 4.1 (Latin Extended subset) | Fills accented Latin letters missing from the site font, loaded only on the Spanish and Portuguese pages | SIL Open Font License 1.1 | Copyright (c) 2016 The Inter Project Authors · Full text · https://github.com/rsms/inter · subset of the official font, served from this site |
| Noto Sans JP (variable, Unicode subsets) | Shows Japanese characters (kana and kanji), loaded only on the Japanese pages (in preparation) | SIL Open Font License 1.1 | Copyright 2014-2021 Adobe (http://www.adobe.com/), with Reserved Font Name 'Source' · Full text · https://github.com/notofonts/noto-cjk · official Google Fonts subsets served unmodified from this site (Latin subsets left out) |
| pdf-lib 1.17.1 | Creates transcript PDFs | MIT | Copyright (c) 2019 Andrew Dillon · Full text · https://github.com/Hopding/pdf-lib · official files served unmodified from this site |
| @pdf-lib/fontkit | Embeds fonts in transcript PDFs | MIT | Full text · https://github.com/Hopding/fontkit · official files served unmodified from this site |
| NanumGothic Regular | Korean font embedded whole in transcript PDFs | SIL Open Font License 1.1 | Copyright (c) 2010, NHN Corporation, with Reserved Font Name Nanum · Full text · official font file served unmodified from this site |
| GTCRN (gtcrn_simple.onnx) | Speech enhancement: "High quality AI" in Noise reduction and Studio sound | MIT | Copyright (c) 2024 Rong Xiaobin · full text · https://github.com/Xiaobin-Rong/gtcrn · ONNX from the sherpa-onnx speech-enhancement release, served unmodified from this site |
| Whisper large-v3-turbo (onnx-community ONNX) | "Highest accuracy" speech recognition (WebGPU only) | MIT | Copyright (c) 2022 OpenAI · https://huggingface.co/onnx-community/whisper-large-v3-turbo · downloaded unmodified from models.dagotools.com, falling back to Hugging Face |
The speech recognition models (Whisper) are used by Audio to Text, Subtitle Generator and AI Meeting Notes. Supertonic 3 is used by the High-quality AI voice in Text to Speech. Voice Noise Remover uses RNNoise and GTCRN, and MP3 saving uses our own encoder.
Supertonic 3 use restrictions (passed on to downstream users)
The use-based restrictions of Supertonic 3 (BigScience Open RAIL-M) carry over to everyone who uses the model and to derivatives. To use the "High-quality AI voice" you must follow the restrictions copied into section 3 of our Terms (unlawful use, exploiting minors, disinformation, impersonation and deepfakes, publishing AI output without disclosing it, medical advice, discrimination, and so on), and if you redistribute the model files you must pass on the same restrictions and the full license text. Full license: copy on models.dagotools.com
Sound processing we wrote ourselves
These features run on our own code (all rights reserved), with no third-party library or trained model weights.
- FLAC saving (lossless encoder) and the M4A/M4R glue code (
assets/encoders.js) - Loudness measurement and normalizing (ITU-R BS.1770-4 style LUFS, true-peak estimate, limiter)
- One-click studio sound (high-pass, hum removal, de-esser, EQ, compressor, leveler)
- Vocal remover (stereo center mask + harmonic/percussive DSP)
- Chord finder (chroma, chord templates, Viterbi smoothing, key detection)
- Voice changer (real-time pitch, robot and echo effects)
- Silence and filler cutter (energy-based speech detection)
- Word (DOCX) export (
assets/docx-lite.js)
When we add a new AI model, we list its name, license and source in the table above.
MP3 saving (our own encoder)
This site saves MP3 files with an encoder we wrote from the public MPEG audio specification (ISO/IEC 11172-3 and 13818-3, Layer III), assets/ttdu/sori-audio.js, our own code with all rights reserved. FLAC saving and loudness (LUFS) measurement are our own code in the same file. No GPL/LGPL code such as LAME or lamejs is used, and the MP3 patents expired in 2017. The Huffman code values defined by the standard were taken from the table data published by go-mp3 (Apache-2.0). Short blocks are used on sharp attacks (e.g. percussion) to reduce pre-echo.
Fonts
The site font, Pretendard, is served from this site using the unmodified official files (see the table above). We don't connect to external font servers.
Trademarks and affiliation
OpenAI, Whisper, Microsoft, Hugging Face, Google, Cloudflare, Xiph.Org and other names are trademarks or names of their respective owners and are used only to identify the source. FreeSoundTools is not affiliated with or endorsed by them.
Operator information
Operator: 행복유통
Representative: 박지훈
Business registration no. 280-30-01413
Address: 경상남도 사천시 사남면 병둔2길 70 (Republic of Korea)
Phone: +82-10-2101-6758
Email: [email protected]
All tools on this site are free and nothing is sold, so no mail-order business registration number applies.