ztimson/ai-utils
1
0
Fork 0
Code Issues Pull Requests Actions Packages Projects Releases Wiki Activity

Updataes?
All checks were successful
Publish Library / Build NPM Project (push) Successful in 26s
Details
Publish Library / Tag Version (push) Successful in 5s
Details

Browse Source
This commit is contained in:
Zakary Timson
2026-02-12 20:14:00 -05:00
parent ad1ee48763
commit e4399e1b7b
4 changed files with 12 additions and 15 deletions
Download Patch File Download Diff File Expand all files Collapse all files

2
package.json
View File

@@ -1,6 +1,6 @@
{ {
"name": "@ztimson/ai-utils", "name": "@ztimson/ai-utils",
"version": "0.6.6", "version": "0.6.7",
"description": "AI Utility library", "description": "AI Utility library",
"author": "Zak Timson", "author": "Zak Timson",
"license": "MIT", "license": "MIT",

10
src/asr.ts
View File

@@ -82,12 +82,12 @@ function combineSpeakerTranscript(chunks: any[], speakers: any[]): string {
return lines.join('\n'); return lines.join('\n');
} }
parentPort?.on('message', async ({ path, model, speaker, torchHome }) => { parentPort?.on('message', async ({ file, speaker, model, modelDir }) => {
try { try {
if(!whisperPipeline) whisperPipeline = await pipeline('automatic-speech-recognition', `Xenova/${model}`, {cache_dir: torchHome, quantized: true}); if(!whisperPipeline) whisperPipeline = await pipeline('automatic-speech-recognition', `Xenova/${model}`, {cache_dir: modelDir, quantized: true});
// Prepare audio file (convert to mono channel wave) // 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.toBitDepth('32f');
wav.toSampleRate(16000); wav.toSampleRate(16000);
const samples = wav.getSamples(); const samples = wav.getSamples();
@@ -111,11 +111,11 @@ parentPort?.on('message', async ({ path, model, speaker, torchHome }) => {
// Speaker Diarization // Speaker Diarization
const hasDiarization = await canDiarization(); const hasDiarization = await canDiarization();
if(!hasDiarization) { if(!hasDiarization) {
parentPort?.postMessage({ text: transcriptResult.text?.trim() || null, warning: 'Speaker diarization unavailable' }); parentPort?.postMessage({ text: transcriptResult.text?.trim() || null, error: 'Speaker diarization unavailable' });
return; return;
} }
const speakers = await runDiarization(path, torchHome); const speakers = await runDiarization(file, modelDir);
const combined = combineSpeakerTranscript(transcriptResult.chunks || [], speakers); const combined = combineSpeakerTranscript(transcriptResult.chunks || [], speakers);
parentPort?.postMessage({ text: combined }); parentPort?.postMessage({ text: combined });
} catch (err) { } catch (err) {

8
src/audio.ts
View File

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

7
src/embedder.ts
View File

@@ -3,11 +3,8 @@ import { parentPort } from 'worker_threads';
let embedder: any; let embedder: any;
parentPort?.on('message', async ({ id, text, model, path }) => { parentPort?.on('message', async ({ id, 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 output = await embedder(text, { pooling: 'mean', normalize: true });
const embedding = Array.from(output.data); const embedding = Array.from(output.data);
parentPort?.postMessage({ id, embedding }); parentPort?.postMessage({ id, embedding });