Web demo
In this section, we describe the web demo for forced alignment. It is a pure-browser app running on WebAssembly: pick an audio/video file (or record from the microphone), enter its transcript, and get the start/end time of every word.
Hint
It runs locally in your browser, without a backend. No audio is uploaded anywhere.
The source code is in react-examples/forced-alignment (the WebAssembly module is in wasm/forced-alignment).
Try it in your browser
One demo per model; click the URL of the model you want to use. The first visit downloads the model (3–37 MB) with a progress bar; later visits use the browser cache.
Model |
Language |
Entry point |
|
Chinese |
https://k2-fsa.github.io/sherpa/onnx/forced-alignment/demo/zh-aishell3-pinyin/ |
|
Chinese |
https://k2-fsa.github.io/sherpa/onnx/forced-alignment/demo/zh-biaobei-pinyin/ |
|
Chinese |
https://k2-fsa.github.io/sherpa/onnx/forced-alignment/demo/zh-large-pinyin/ |
|
English |
https://k2-fsa.github.io/sherpa/onnx/forced-alignment/demo/english-mfa-v2.2.1/ |
|
English |
https://k2-fsa.github.io/sherpa/onnx/forced-alignment/demo/english-mfa-v3.0.0/ |
|
English |
https://k2-fsa.github.io/sherpa/onnx/forced-alignment/demo/english-mfa-v3.1.0/ |
The same demos can be downloaded from the
react-fa
release as sherpa-onnx-react-fa-<model>.zip — unzip a file and serve the
directory with any static HTTP server to host it yourself. See
scripts/mfa/web.md
for a description of each file.
Build from source
Please see react-examples/forced-alignment/README.md for how to build the wasm module and the web app yourself.