segfault fix

package.json

@@ -1,6 +1,6 @@
 {
 	"name": "@ztimson/ai-utils",
-	"version": "0.7.4",
+	"version": "0.7.6",
 	"description": "AI Utility library",
 	"author": "Zak Timson",
 	"license": "MIT",

src/asr.ts

@@ -7,8 +7,6 @@ import { join } from 'node:path';
 import { tmpdir } from 'node:os';
 import wavefile from 'wavefile';
 
-let whisperPipeline: any;
-
 export async function canDiarization(): Promise<string | null> {
 	const checkPython = (cmd: string) => {
 		return new Promise<boolean>((resolve) => {
@@ -110,30 +108,27 @@ function prepareAudioBuffer(file: string): [string, Float32Array] {
 }
 
 parentPort?.on('message', async ({ file, speaker, model, modelDir, token }) => {
+	let tempFile = null;
 	try {
-		if(!whisperPipeline) whisperPipeline = await pipeline('automatic-speech-recognition', `Xenova/${model}`, {cache_dir: modelDir, quantized: true});
+		const asr: any = await pipeline('automatic-speech-recognition', `Xenova/${model}`, {cache_dir: modelDir, quantized: true});
-
-		// Prepare audio file
 		const [f, buffer] = prepareAudioBuffer(file);
-
+		tempFile = f !== file ? f : null;
-		// Fetch transcript and speakers
 		const hasDiarization = await canDiarization();
 		const [transcript, speakers] = await Promise.all([
-			whisperPipeline(buffer, {return_timestamps: speaker ? 'word' : false}),
+			asr(buffer, {return_timestamps: speaker ? 'word' : false}),
 			(!speaker || !token || !hasDiarization) ? Promise.resolve(): runDiarization(hasDiarization, f, modelDir, token),
 		]);
-		if(file != f) rmSync(f, { recursive: true, force: true });
 
-		// Return any results / errors if no more processing required
 		const text = transcript.text?.trim() || null;
 		if(!speaker) return parentPort?.postMessage({ text });
 		if(!token) return parentPort?.postMessage({ text, error: 'HuggingFace token required' });
 		if(!hasDiarization) return parentPort?.postMessage({ text, error: 'Speaker diarization unavailable' });
 
-		// Combine transcript and speakers
 		const combined = combineSpeakerTranscript(transcript.chunks || [], speakers || []);
 		parentPort?.postMessage({ text: combined });
 	} catch (err: any) {
 		parentPort?.postMessage({ error: err.stack || err.message });
+	} finally {
+		if(tempFile) rmSync(tempFile, { recursive: true, force: true });
 	}
 });

src/audio.ts

@@ -15,7 +15,7 @@ export class Audio {
 		let p = new Promise<string | null>((resolve, reject) => {
 			const worker = new Worker(join(dirname(fileURLToPath(import.meta.url)), 'asr.js'));
 			const handleMessage = ({ text, warning, error }: any) => {
-				worker.terminate();
+				setTimeout(() => worker.terminate(), 1000);
 				if(aborted) return;
 				if(error) reject(new Error(error));
 				else {
@@ -24,7 +24,7 @@ export class Audio {
 				}
 			};
 			const handleError = (err: Error) => {
-				worker.terminate();
+				setTimeout(() => worker.terminate(), 1000);
 				if(!aborted) reject(err);
 			};
 			worker.on('message', handleMessage);
@@ -42,8 +42,8 @@ export class Audio {
 				if(!transcript) return transcript;
 				let chunks = this.ai.language.chunk(transcript, 500, 0);
 				if(chunks.length > 4) chunks = [...chunks.slice(0, 3), <string>chunks.at(-1)];
-				const names = await this.ai.language.json(chunks.join('\n'), '{1: "Detected Name"}', {
+				const names = await this.ai.language.json(chunks.join('\n'), '{1: "Detected Name", 2: "Second Name"}', {
-					system: 'Use this following transcript to identify speakers. Only identify speakers you are sure about',
+					system: 'Use the following transcript to identify speakers. Only identify speakers you are positive about, dont mention speakers you are unsure about in your response',
 					temperature: 0.1,
 				});
 				Object.entries(names).forEach(([speaker, name]) => {