Loading Moonshine js locally

[ArEnSc]

Hey can you allow a way to load the wasm requirements?
Also to load the model locally without the cdn
Effectively I am trying to get a Electron app put together.
import { env } from '@huggingface/transformers'
very similar to this
env.backends.onnx.wasm.wasmPaths = {
'ort-wasm-simd-threaded.mjs': ${base}ort-wasm-simd-threaded.mjs,
'ort-wasm-simd-threaded.wasm': ${base}ort-wasm-simd-threaded.wasm,
'ort-wasm-simd-threaded.jsep.mjs': ${base}ort-wasm-simd-threaded.jsep.mjs,
'ort-wasm-simd-threaded.jsep.wasm': ${base}ort-wasm-simd-threaded.jsep.wasm
}

@evmaki

[evmaki]

You can change the asset path to a local directory if you'd prefer not to use the CDN:

import * as Moonshine from "@usefulsensors/moonshine-js";

Moonshine.Settings.BASE_ASSET_PATH = "my/local/directory"

However, I'm noticing that I need to make some changes in order for this to also apply to the VAD model assets. I should be able to do that within the next 24 hours or so. I'll keep you posted.

[ArEnSc]

Another thing I noticed was sometimes the model takes in input but doesn't produce a transcript, usually on the second try it works.

[evmaki]

OK, I've published a new version on npm that will allow you to set each of the asset paths. You can do it like so:

import * as Moonshine from "@usefulsensors/moonshine-js"

Moonshine.Settings.BASE_ASSET_PATH = {
    ONNX_RUNTIME: "/assets/",
    MOONSHINE: "/assets/",
    SILERO_VAD: "/assets/"
}

Assuming you've installed the package from npm, i.e., npm i @usefulsensors/moonshine-js, you can run this shell script in your project directory to copy all of the files into your local assets directory:

ASSET_DIR="assets"
mkdir $ASSET_DIR

# copy moonshine models
cp -r node_modules/@usefulsensors/moonshine-js/dist/model $ASSET_DIR

# copy onnx runtime
cp node_modules/onnxruntime-web/dist/ort-wasm-simd-threaded.wasm $ASSET_DIR
cp node_modules/onnxruntime-web/dist/ort-wasm-simd-threaded.mjs $ASSET_DIR
cp node_modules/onnxruntime-web/dist/ort-wasm-simd-threaded.jsep.wasm $ASSET_DIR
cp node_modules/onnxruntime-web/dist/ort-wasm-simd-threaded.jsep.mjs $ASSET_DIR

# copy vad model and worklet
cp node_modules/@ricky0123/vad-web/dist/silero_vad_v5.onnx $ASSET_DIR
cp node_modules/@ricky0123/vad-web/dist/vad.worklet.bundle.min.js $ASSET_DIR

RE: the second question, this could be a variety of things. It might be that the buffer doesn't have enough audio in it to produce a transcript on the first pass, or something else. If you have a repeatable example with an audio file, I'd be happy to dig a little deeper.

Let me know if you have any other issues or requests as you build your Electron app. I haven't used it in that context so I'm interested to see how it works!

[ArEnSc]

Thanks Evan I will give it a shot!

[xinliu9451]

OK, I've published a new version on npm that will allow you to set each of the asset paths. You can do it like so:好的，我在 npm 上发布了一个新版本，允许你设置每个资源的路径。你可以这样做：

import * as Moonshine from "@usefulsensors/moonshine-js"

Moonshine.Settings.BASE_ASSET_PATH = {
ONNX_RUNTIME: "/assets/",
MOONSHINE: "/assets/",
SILERO_VAD: "/assets/"
}
Assuming you've installed the package from npm, i.e., npm i @usefulsensors/moonshine-js, you can run this shell script in your project directory to copy all of the files into your local assets directory:假设您已经从 npm 安装了软件包，即 npm i @usefulsensors/moonshine-js ，您可以在项目目录中运行此 shell 脚本，将所有文件复制到本地资产目录中：

ASSET_DIR="assets"
mkdir $ASSET_DIR

copy moonshine models

cp -r node_modules/@usefulsensors/moonshine-js/dist/model $ASSET_DIR

copy onnx runtime

cp node_modules/onnxruntime-web/dist/ort-wasm-simd-threaded.wasm $ASSET_DIR
cp node_modules/onnxruntime-web/dist/ort-wasm-simd-threaded.mjs $ASSET_DIR
cp node_modules/onnxruntime-web/dist/ort-wasm-simd-threaded.jsep.wasm $ASSET_DIR
cp node_modules/onnxruntime-web/dist/ort-wasm-simd-threaded.jsep.mjs $ASSET_DIR

copy vad model and worklet

cp node_modules/@ricky0123/vad-web/dist/silero_vad_v5.onnx $ASSET_DIR
cp node_modules/@ricky0123/vad-web/dist/vad.worklet.bundle.min.js $ASSET_DIR
RE: the second question, this could be a variety of things. It might be that the buffer doesn't have enough audio in it to produce a transcript on the first pass, or something else. If you have a repeatable example with an audio file, I'd be happy to dig a little deeper.回复：第二个问题，这可能有很多原因。可能是缓冲区中没有足够的音频，无法在第一次读取时生成转录文本，也可能是其他原因。如果您有可重复的音频文件示例，我很乐意进一步探讨。

Let me know if you have any other issues or requests as you build your Electron app. I haven't used it in that context so I'm interested to see how it works!如果您在构建 Electron 应用时遇到任何其他问题或需求，请告诉我。我还没有在那种情况下使用过它，所以我很想看看它的效果！

Regarding the second issue, I have an audio transcription that doesn't output results, and this problem occurs very frequently in my scenario. Do you have any methods to resolve it?

audio.zip

[ArEnSc]

Basically my experience so I moved on @xinliu9451 I am pretty sure this is a bug of some kind