Speech Engine SPA — v9.2 ONNX Load Fix

Fixes the “JS execution provider is not supported in this build” error by using browser-safe ONNX Runtime execution providers. This build loads ONNX models via local file upload and shows model compatibility details.
ONNX Runtime: checking... WASM: checking... WebGPU: checking...

Execution Provider Fix

The browser build of ONNX Runtime Web supports execution providers such as wasm and optionally webgpu. It does not support Node-style cpu or internal JS execution providers.

This loader now tries:
1. executionProviders: ["wasm"]
2. executionProviders: [{ name: "wasm" }]
3. executionProviders: ["webgpu"] if WebGPU exists
4. executionProviders: [{ name: "webgpu" }] if WebGPU exists

It no longer requests cpu or js.

Model Loader

Silero VAD Model (.onnx)
Not loaded
No model metadata yet.
Acoustic CTC Model (.onnx)
Not loaded
No model metadata yet.
Speaker Embedding Model (.onnx)
Not loaded
No model metadata yet.
p-MLM Rescorer Model (.onnx)
Not loaded
No model metadata yet.

Compatibility Notes

A successful ONNX session load does not always mean the model is directly usable by the speech app.

For example, official Silero VAD ONNX models are often stateful and may require inputs like:
input, sr, h, c

The current simple bridge works best with models that have:
- one float32 audio input
- one float32 probability/output tensor

After loading a model, inspect the input/output names below the file picker.

Diagnostics Log

0 events