OpenAI's Whisper in the browser, run by OnnxStream (WASM)

This is a demo of the WASM build of OnnxStream: Whisper runs fully in the browser without any backend server. The OnnxStream JS API is very similar to the C++ API: you can specify a model as a series of operations in text format, you can add input tensors, read output tensors etc. The Whisper ONNX models from the sherpa-onnx project were converted to the text format compatible with OnnxStream using onnx2txt. Feel free to view the source of this page: it is about 1200 lines of HTML and JS code that demonstrate how to use the API (specifically the "initModel" and "runModel" functions). Test with JFK.wav or Record from the Microphone or Use MP3 or WAV file