Invalid Data Location on audio models

[FelippeChemello]

System Info

@huggingface/[email protected]

[email protected]

Environment/Platform

• Website/web-app
• Browser extension
• Server-side (e.g., Node.js, Deno, Bun)
• Desktop app (e.g., Electron)
• Other (e.g., VSCode extension)

Description

I have tried both the Xenova/whisper-tiny and onnx-community/moonshine-tiny-ONNX models, and they throw the following error when the pipeline is called: Invalid data location: undefined for input.

Reproduction

import { pipeline } from "@huggingface/transformers"

console.log(`Loading models...`)

const model_id = "Xenova/whisper-tiny"
const transcriber = await pipeline("automatic-speech-recognition", model_id)

console.log(`Models loaded!`)

const transcribe = async (buffer: Float32Array) => {
    const output = await transcriber(buffer)

    console.log(output)
}

self.onmessage = async ({ data }) => {
    const { buffer } = data

    await transcribe(buffer)
}

const audioURL = URL.createObjectURL(audioFile)
read_audio(audioURL, 16000).then(audioData => {
    worker.current?.postMessage({ buffer: audioData })
})

[xenova]

Related: microsoft/onnxruntime#20431

Can you confirm that the audioData you are passing is valid?