Licenses & credits
This site runs entirely in your browser using the following open components.
Speech recognition models
A different model is used per language. All of them are redistributed unmodified apart from packaging for WebAssembly.
- Kroko ASR Community models — © Banafo. Used for English, Italian, German, Spanish, French, Dutch, Portuguese, Swedish and Turkish. Licensed under CC BY-SA 4.0. Source: huggingface.co/Banafo/Kroko-ASR
- Hebrew streaming Zipformer2 — © Yakov Kolani, mirrored by ivrit.ai. Licensed under CC BY 4.0. Source: ivrit-ai/zipformer2-66m-streaming-he.
- Korean streaming zipformer — Apache License 2.0. Trained on KsponSpeech with icefall and published as johnBamma/icefall-asr-ksponspeech-pruned-transducer-stateless7-streaming; converted to ONNX by k2-fsa.
- Russian streaming zipformer — Apache License 2.0. © Alpha Cephei (Vosk). Source: alphacep/vosk-model-small-streaming-ru
- Silero VAD — MIT License. github.com/snakers4/silero-vad
Sample audio
- The sample files on the English and Italian pages are excerpts from FLEURS (Google), licensed under CC BY 4.0. Source: huggingface.co/datasets/google/fleurs
Runtime
- sherpa-onnx — Apache License 2.0. github.com/k2-fsa/sherpa-onnx
- ONNX Runtime — MIT License. github.com/microsoft/onnxruntime
Media handling
- mp4box.js — BSD 3-Clause License. © Telecom ParisTech. Used to read the audio track out of MP4 / MOV / M4A files without loading the video. github.com/gpac/mp4box.js · licence text. Bundled unmodified (v0.5.4) and loaded only when you open an MP4-family file.
Interface
- Poppins — SIL Open Font License 1.1, served by Google Fonts.
Privacy
Audio and video files are processed locally and are never uploaded. Anonymous usage counters (page views, processing time, cache hits) are collected and contain no file names, audio, or transcripts. Google Analytics also counts visits and uses first-party analytics cookies; this site does not show a consent dialog. IP addresses are not stored in our first-party usage dataset. To tell automated traffic apart from people, the country and the name of the network operator of the connection (for example “Deutsche Telekom” or “Microsoft”) are recorded.
BudouX subtitle phrase models
Copyright 2021 Google LLC. Japanese, Simplified Chinese, Traditional Chinese and Thai phrase models and pure parser, version 0.10.3. Source · Apache License 2.0.