Added hugging face token

package.json

@@ -1,6 +1,6 @@
 {
 	"name": "@ztimson/ai-utils",
-	"version": "0.6.6",
+	"version": "0.6.10",
 	"description": "AI Utility library",
 	"author": "Zak Timson",
 	"license": "MIT",

src/ai.ts

@@ -8,6 +8,8 @@ export type AbortablePromise<T> = Promise<T> & {
 };
 
 export type AiOptions = {
+	/** Token to pull models from hugging face */
+	hfToken?: string;
 	/** Path to models */
 	path?: string;
 	/** ASR model: whisper-tiny, whisper-base */

src/asr.ts

@@ -14,14 +14,15 @@ export async function canDiarization(): Promise<boolean> {
 	});
 }
 
-async function runDiarization(audioPath: string, torchHome: string): Promise<any[]> {
+async function runDiarization(audioPath: string, dir: string, token: string): Promise<any[]> {
 	const script = `
 import sys
 import json
 import os
 from pyannote.audio import Pipeline
 
-os.environ['TORCH_HOME'] = "${torchHome}"
+os.environ['TORCH_HOME'] = "${dir}"
+os.environ['HF_TOKEN'] = "${token}"
 pipeline = Pipeline.from_pretrained("pyannote/speaker-diarization-3.1")
 diarization = pipeline(sys.argv[1])
 
@@ -82,12 +83,13 @@ function combineSpeakerTranscript(chunks: any[], speakers: any[]): string {
 	return lines.join('\n');
 }
 
-parentPort?.on('message', async ({ path, model, speaker, torchHome }) => {
+parentPort?.on('message', async ({ file, speaker, model, modelDir, token }) => {
 	try {
-		if(!whisperPipeline) whisperPipeline = await pipeline('automatic-speech-recognition', `Xenova/${model}`, {cache_dir: torchHome, quantized: true});
+		console.log('worker', file);
+		if(!whisperPipeline) whisperPipeline = await pipeline('automatic-speech-recognition', `Xenova/${model}`, {cache_dir: modelDir, quantized: true});
 
 		// Prepare audio file (convert to mono channel wave)
-		const wav = new wavefile.WaveFile(fs.readFileSync(path));
+		const wav = new wavefile.WaveFile(fs.readFileSync(file));
 		wav.toBitDepth('32f');
 		wav.toSampleRate(16000);
 		const samples = wav.getSamples();
@@ -110,12 +112,12 @@ parentPort?.on('message', async ({ path, model, speaker, torchHome }) => {
 
 		// Speaker Diarization
 		const hasDiarization = await canDiarization();
-		if(!hasDiarization) {
+		if(!token || !hasDiarization) {
-			parentPort?.postMessage({ text: transcriptResult.text?.trim() || null, warning: 'Speaker diarization unavailable' });
+			parentPort?.postMessage({ text: transcriptResult.text?.trim() || null, error: 'Speaker diarization unavailable' });
 			return;
 		}
 
-		const speakers = await runDiarization(path, torchHome);
+		const speakers = await runDiarization(file, modelDir, token);
 		const combined = combineSpeakerTranscript(transcriptResult.chunks || [], speakers);
 		parentPort?.postMessage({ text: combined });
 	} catch (err) {

src/audio.ts

@@ -1,18 +1,19 @@
+import {fileURLToPath} from 'url';
 import {Worker} from 'worker_threads';
-import path from 'node:path';
 import {AbortablePromise, Ai} from './ai.ts';
 import {canDiarization} from './asr.ts';
+import {dirname, join} from 'path';
 
 export class Audio {
 	constructor(private ai: Ai) {}
 
-	asr(filepath: string, options: { model?: string; speaker?: boolean } = {}): AbortablePromise<string | null> {
+	asr(file: string, options: { model?: string; speaker?: boolean } = {}): AbortablePromise<string | null> {
 		const { model = this.ai.options.asr || 'whisper-base', speaker = false } = options;
 		let aborted = false;
 		const abort = () => { aborted = true; };
 
 		const p = new Promise<string | null>((resolve, reject) => {
-			const worker = new Worker(path.join(import.meta.dirname, 'asr.js'));
+			const worker = new Worker(join(dirname(fileURLToPath(import.meta.url)), 'asr.js'));
 			const handleMessage = ({ text, warning, error }: any) => {
 				worker.terminate();
 				if(aborted) return;
@@ -31,7 +32,7 @@ export class Audio {
 			worker.on('exit', (code) => {
 				if(code !== 0 && !aborted) reject(new Error(`Worker exited with code ${code}`));
 			});
-			worker.postMessage({path: filepath, model, speaker, torchHome: this.ai.options.path,});
+			worker.postMessage({file, model, speaker, modelDir: this.ai.options.path, token: this.ai.options.hfToken});
 		});
 		return Object.assign(p, { abort });
 	}

src/embedder.ts

@@ -3,12 +3,9 @@ import { parentPort } from 'worker_threads';
 
 let embedder: any;
 
-parentPort?.on('message', async ({ id, text, model, path }) => {
+parentPort?.on('message', async ({text, model, modelDir }) => {
-	if(!embedder) embedder = await pipeline('feature-extraction', 'Xenova/' + model, {
+	if(!embedder) embedder = await pipeline('feature-extraction', 'Xenova/' + model, {quantized: true, cache_dir: modelDir});
-		quantized: true,
-		cache_dir: path,
-	});
 	const output = await embedder(text, { pooling: 'mean', normalize: true });
 	const embedding = Array.from(output.data);
-	parentPort?.postMessage({ id, embedding });
+	parentPort?.postMessage({embedding});
 });

src/llm.ts

@@ -271,7 +271,7 @@ class LLM {
 				worker.on('exit', (code) => {
 					if(code !== 0) reject(new Error(`Worker exited with code ${code}`));
 				});
-				worker.postMessage({text, model: this.ai.options?.embedder || 'bge-small-en-v1.5', path: this.ai.options.path});
+				worker.postMessage({text, model: this.ai.options?.embedder || 'bge-small-en-v1.5', modelDir: this.ai.options.path});
 			});
 		};
 		const chunks = this.chunk(target, maxTokens, overlapTokens);