Credits

Open-source credits

The open-source software and models that run in your browser on this site, with their licenses and where to get their source code.

whisper.cpp and ggml

whisper.cpp is the speech recognition engine that transcribes your file. We compile it to WebAssembly, with a small interface of our own, and it runs in your browser. ggml, the machine learning library inside it, comes from the same authors under the same license. Copyright (c) 2023-2026 The ggml authors.

Source: github.com/ggml-org/whisper.cpp, version 1.9.4. The multi-threaded build adds one line to it: in the voice detector's setup, the voice detector runs on the run's own number of threads (at most 4) instead of a fixed 4, so it never asks for more threads than the run has. The single-thread and WebGPU builds use version 1.9.4 unchanged. Licensed under the MIT License.

Whisper model weights

Whisper is the speech recognition model. OpenAI released its code and weights; this site uses two of its sizes, base and small, in the quantized ggml files that the whisper.cpp project publishes. Copyright (c) 2022 OpenAI.

Source: github.com/openai/whisper; the ggml files: huggingface.co/ggerganov/whisper.cpp. Licensed under the MIT License.

Silero VAD

Silero VAD is the voice detector that marks the stretches with speech, so the engine skips silence. Copyright (c) 2020-present Silero Team.

Source: github.com/snakers4/silero-vad, version 5.1.2, in the ggml file from huggingface.co/ggml-org/whisper-vad. Licensed under the MIT License.

JSON for Modern C++

This library, by Niels Lohmann, is compiled into the engine, where it reads the settings the page sends and writes the results back. Copyright (c) 2013-2022 Niels Lohmann.

Source: github.com/nlohmann/json, version 3.11.2, as included with whisper.cpp. Licensed under the MIT License.

Emscripten runtime

Emscripten is the compiler that turns the engine into WebAssembly. Parts of its runtime are inside every engine build: the JavaScript glue code and the C and C++ standard libraries.

emdawnwebgpu and webgpu-headers

These are inside the engine's WebGPU build only, which the page downloads only when your browser offers a usable WebGPU adapter. They connect the engine to your browser's WebGPU.

Mediabunny

Where the file can be read in slices, Mediabunny reads the file you choose a slice at a time inside your browser: it decodes compressed sound with your browser's own decoders, and uncompressed sound, as in a WAV file, itself. When the file cannot be read that way, the page decodes the whole file with your browser instead, without Mediabunny. Copyright (c) 2026-present, Vanilagy and contributors.

Source: github.com/Vanilagy/mediabunny, version 1.59.1, used without changes. Licensed under the Mozilla Public License 2.0; the source code of every file we use is in that repository at that version.