Piper Phonemizer Wasm

[iAmIlluminati]

trafficstars

I am trying to run the onnx model directly with onnx-web-runtime and I was able to. But I am finding it difficult to compile the piper-phonemizer for wasm. Can you share the script for it or a compiled Wasm version which generates phonemeId from text.

[csukuangfj]

I am using a fork of piper-phonemizer and espeak-ng, please see

• https://github.com/k2-fsa/sherpa-onnx/blob/master/cmake/espeak-ng-for-piper.cmake
• https://github.com/k2-fsa/sherpa-onnx/blob/master/cmake/piper-phonemize.cmake

Please find the scripts for building wasm at

• https://github.com/k2-fsa/sherpa-onnx/blob/master/build-wasm-simd-tts.sh

By the way, you can use sherpa-onnx for wasm with piper models directly.

Please see our doc at https://k2-fsa.github.io/sherpa/onnx/tts/wasm/index.html

We have done everything for you about using piper with wasm. You don't need to write your own. It supports all models from piper.

[iAmIlluminati]

Thanks Fangjun. Will check those build scripts again.

I started sherpa-onnx builds and it was surprisingly easy to swap the models and compile to new bundles with diff models. But my application requires a very low generation time, so I was just batching the tokenIds and generating the audio parallel with a window approach, as I am concerned about playing and not exporting the audio.

With the phonemizer bundled in sherpa onnx I would need to batch based on words to attempt something like this, but I guess the pauses between will be more artificial in word based approach.