@ztimson/ai-utils 0.7.3 → 0.7.5

package/dist/asr.d.ts

@@ -1 +1 @@
-export declare function canDiarization(): Promise<boolean>;
+export declare function canDiarization(): Promise<string | null>;

package/dist/asr.js

@@ -1,10 +1,10 @@
-"use strict";Object.defineProperty(exports,Symbol.toStringTag,{value:"Module"});const S=require("@xenova/transformers"),u=require("worker_threads"),g=require("node:child_process"),d=require("node:fs"),y=require("node:path"),k=require("node:os"),h=require("wavefile");let f;async function w(){return new Promise(o=>{const r=g.spawn("python",["-c","import pyannote.audio"]);r.on("close",n=>o(n===0)),r.on("error",()=>o(!1))})}async function P(o,r,n){const s=`
+"use strict";Object.defineProperty(exports,Symbol.toStringTag,{value:"Module"});const k=require("@xenova/transformers"),l=require("worker_threads"),y=require("node:child_process"),m=require("node:fs"),h=require("node:path"),S=require("node:os"),g=require("wavefile");let d;async function w(){const s=r=>new Promise(t=>{const n=y.spawn(r,["-c","import pyannote.audio"]);n.on("close",o=>t(o===0)),n.on("error",()=>t(!1))});return await s("python3")?"python3":await s("python")?"python":null}async function P(s,r,t,n){const o=`
 import sys
 import json
 import os
 from pyannote.audio import Pipeline
 
-os.environ['TORCH_HOME'] = r"${r}"
+os.environ['TORCH_HOME'] = r"${t}"
 pipeline = Pipeline.from_pretrained("pyannote/speaker-diarization-3.1", token="${n}")
 output = pipeline(sys.argv[1])
 
@@ -13,6 +13,6 @@ for turn, speaker in output.speaker_diarization:
     segments.append({"start": turn.start, "end": turn.end, "speaker": speaker})
 
 print(json.dumps(segments))
-`;return new Promise((i,t)=>{let a="";const e=g.spawn("python",["-c",s,o]);e.stdout.on("data",p=>a+=p.toString()),e.stderr.on("data",p=>console.error(p.toString())),e.on("close",p=>{if(p===0)try{i(JSON.parse(a))}catch{t(new Error("Failed to parse diarization output"))}else t(new Error(`Python process exited with code ${p}`))}),e.on("error",t)})}function v(o,r){const n=new Map;let s=0;r.forEach(e=>{n.has(e.speaker)||n.set(e.speaker,++s)});const i=[];let t=-1,a="";return o.forEach(e=>{const p=e.timestamp[0],l=r.find(m=>p>=m.start&&p<=m.end),c=l?n.get(l.speaker):1;c!==t?(a&&i.push(`[Speaker ${t}]: ${a.trim()}`),t=c,a=e.text):a+=e.text}),a&&i.push(`[Speaker ${t}]: ${a.trim()}`),i.join(`
-`)}function _(o){let r,n;try{r=new h.WaveFile(d.readFileSync(o))}catch{n=y.join(d.mkdtempSync(y.join(k.tmpdir(),"audio-")),"converted.wav"),g.execSync(`ffmpeg -i "${o}" -ar 16000 -ac 1 -f wav "${n}"`,{stdio:"ignore"}),r=new h.WaveFile(d.readFileSync(n))}finally{r.toBitDepth("32f"),r.toSampleRate(16e3);const s=r.getSamples();if(Array.isArray(s)){const i=s[0],t=s[1],a=new Float32Array(i.length);for(let e=0;e<i.length;e++)a[e]=(i[e]+t[e])/2;return[n||o,a]}return[n||o,s]}}u.parentPort?.on("message",async({file:o,speaker:r,model:n,modelDir:s,token:i})=>{try{f||(f=await S.pipeline("automatic-speech-recognition",`Xenova/${n}`,{cache_dir:s,quantized:!0}));const[t,a]=_(o),e=r&&await w(),[p,l]=await Promise.all([f(a,{return_timestamps:r?"word":!1}),!r||!i||!e?Promise.resolve():P(t,s,i)]);o!=t&&d.rmSync(t,{recursive:!0,force:!0});const c=p.text?.trim()||null;if(!r)return u.parentPort?.postMessage({text:c});if(!i)return u.parentPort?.postMessage({text:c,error:"HuggingFace token required"});if(!e)return u.parentPort?.postMessage({text:c,error:"Speaker diarization unavailable"});const m=v(p.chunks||[],l||[]);u.parentPort?.postMessage({text:m})}catch(t){u.parentPort?.postMessage({error:t.stack||t.message})}});exports.canDiarization=w;
+`;return new Promise((a,i)=>{let e="";const c=y.spawn(s,["-c",o,r]);c.stdout.on("data",p=>e+=p.toString()),c.stderr.on("data",p=>console.error(p.toString())),c.on("close",p=>{if(p===0)try{a(JSON.parse(e))}catch{i(new Error("Failed to parse diarization output"))}else i(new Error(`Python process exited with code ${p}`))}),c.on("error",i)})}function v(s,r){const t=new Map;let n=0;r.forEach(e=>{t.has(e.speaker)||t.set(e.speaker,++n)});const o=[];let a=-1,i="";return s.forEach(e=>{const c=e.timestamp[0],p=r.find(f=>c>=f.start&&c<=f.end),u=p?t.get(p.speaker):1;u!==a?(i&&o.push(`[Speaker ${a}]: ${i.trim()}`),a=u,i=e.text):i+=e.text}),i&&o.push(`[Speaker ${a}]: ${i.trim()}`),o.join(`
+`)}function _(s){let r,t;try{r=new g.WaveFile(m.readFileSync(s))}catch{t=h.join(m.mkdtempSync(h.join(S.tmpdir(),"audio-")),"converted.wav"),y.execSync(`ffmpeg -i "${s}" -ar 16000 -ac 1 -f wav "${t}"`,{stdio:"ignore"}),r=new g.WaveFile(m.readFileSync(t))}finally{r.toBitDepth("32f"),r.toSampleRate(16e3);const n=r.getSamples();if(Array.isArray(n)){const o=n[0],a=n[1],i=new Float32Array(o.length);for(let e=0;e<o.length;e++)i[e]=(o[e]+a[e])/2;return[t||s,i]}return[t||s,n]}}l.parentPort?.on("message",async({file:s,speaker:r,model:t,modelDir:n,token:o})=>{try{d||(d=await k.pipeline("automatic-speech-recognition",`Xenova/${t}`,{cache_dir:n,quantized:!0}));const[a,i]=_(s),e=await w(),[c,p]=await Promise.all([d(i,{return_timestamps:r?"word":!1}),!r||!o||!e?Promise.resolve():P(e,a,n,o)]);s!=a&&m.rmSync(a,{recursive:!0,force:!0});const u=c.text?.trim()||null;if(!r)return l.parentPort?.postMessage({text:u});if(!o)return l.parentPort?.postMessage({text:u,error:"HuggingFace token required"});if(!e)return l.parentPort?.postMessage({text:u,error:"Speaker diarization unavailable"});const f=v(c.chunks||[],p||[]);l.parentPort?.postMessage({text:f})}catch(a){l.parentPort?.postMessage({error:a.stack||a.message})}});exports.canDiarization=w;
 //# sourceMappingURL=asr.js.map

package/dist/asr.js.map

@@ -1 +1 @@
-{"version":3,"file":"asr.js","sources":["../src/asr.ts"],"sourcesContent":["import { pipeline } from '@xenova/transformers';\nimport { parentPort } from 'worker_threads';\nimport { spawn } from 'node:child_process';\nimport { execSync } from 'node:child_process';\nimport { mkdtempSync, rmSync, readFileSync } from 'node:fs';\nimport { join } from 'node:path';\nimport { tmpdir } from 'node:os';\nimport wavefile from 'wavefile';\n\nlet whisperPipeline: any;\n\nexport async function canDiarization(): Promise<boolean> {\n\treturn new Promise((resolve) => {\n\t\tconst proc = spawn('python', ['-c', 'import pyannote.audio']);\n\t\tproc.on('close', (code: number) => resolve(code === 0));\n\t\tproc.on('error', () => resolve(false));\n\t});\n}\n\nasync function runDiarization(audioPath: string, dir: string, token: string): Promise<any[]> {\n\tconst script = `\nimport sys\nimport json\nimport os\nfrom pyannote.audio import Pipeline\n\nos.environ['TORCH_HOME'] = r\"${dir}\"\npipeline = Pipeline.from_pretrained(\"pyannote/speaker-diarization-3.1\", token=\"${token}\")\noutput = pipeline(sys.argv[1])\n\nsegments = []\nfor turn, speaker in output.speaker_diarization:\n    segments.append({\"start\": turn.start, \"end\": turn.end, \"speaker\": speaker})\n\nprint(json.dumps(segments))\n`;\n\n\treturn new Promise((resolve, reject) => {\n\t\tlet output = '';\n\t\tconst proc = spawn('python', ['-c', script, audioPath]);\n\t\tproc.stdout.on('data', (data: Buffer) => output += data.toString());\n\t\tproc.stderr.on('data', (data: Buffer) => console.error(data.toString()));\n\t\tproc.on('close', (code: number) => {\n\t\t\tif(code === 0) {\n\t\t\t\ttry {\n\t\t\t\t\tresolve(JSON.parse(output));\n\t\t\t\t} catch (err) {\n\t\t\t\t\treject(new Error('Failed to parse diarization output'));\n\t\t\t\t}\n\t\t\t} else {\n\t\t\t\treject(new Error(`Python process exited with code ${code}`));\n\t\t\t}\n\t\t});\n\t\tproc.on('error', reject);\n\t});\n}\n\nfunction combineSpeakerTranscript(chunks: any[], speakers: any[]): string {\n\tconst speakerMap = new Map();\n\tlet speakerCount = 0;\n\tspeakers.forEach((seg: any) => {\n\t\tif(!speakerMap.has(seg.speaker)) speakerMap.set(seg.speaker, ++speakerCount);\n\t});\n\n\tconst lines: string[] = [];\n\tlet currentSpeaker = -1;\n\tlet currentText = '';\n\tchunks.forEach((chunk: any) => {\n\t\tconst time = chunk.timestamp[0];\n\t\tconst speaker = speakers.find((s: any) => time >= s.start && time <= s.end);\n\t\tconst speakerNum = speaker ? speakerMap.get(speaker.speaker) : 1;\n\t\tif (speakerNum !== currentSpeaker) {\n\t\t\tif(currentText) lines.push(`[Speaker ${currentSpeaker}]: ${currentText.trim()}`);\n\t\t\tcurrentSpeaker = speakerNum;\n\t\t\tcurrentText = chunk.text;\n\t\t} else {\n\t\t\tcurrentText += chunk.text;\n\t\t}\n\t});\n\tif(currentText) lines.push(`[Speaker ${currentSpeaker}]: ${currentText.trim()}`);\n\treturn lines.join('\\n');\n}\n\nfunction prepareAudioBuffer(file: string): [string, Float32Array] {\n\tlet wav: any, tmp;\n\ttry {\n\t\twav = new wavefile.WaveFile(readFileSync(file));\n\t} catch(err) {\n\t\ttmp = join(mkdtempSync(join(tmpdir(), 'audio-')), 'converted.wav');\n\t\texecSync(`ffmpeg -i \"${file}\" -ar 16000 -ac 1 -f wav \"${tmp}\"`, { stdio: 'ignore' });\n\t\twav = new wavefile.WaveFile(readFileSync(tmp));\n\t} finally {\n\t\twav.toBitDepth('32f');\n\t\twav.toSampleRate(16000);\n\t\tconst samples = wav.getSamples();\n\t\tif(Array.isArray(samples)) {\n\t\t\tconst left = samples[0];\n\t\t\tconst right = samples[1];\n\t\t\tconst buffer = new Float32Array(left.length);\n\t\t\tfor (let i = 0; i < left.length; i++) buffer[i] = (left[i] + right[i]) / 2;\n\t\t\treturn [tmp || file, buffer];\n\t\t}\n\t\treturn [tmp || file, samples];\n\t}\n}\n\nparentPort?.on('message', async ({ file, speaker, model, modelDir, token }) => {\n\ttry {\n\t\tif(!whisperPipeline) whisperPipeline = await pipeline('automatic-speech-recognition', `Xenova/${model}`, {cache_dir: modelDir, quantized: true});\n\n\t\t// Prepare audio file\n\t\tconst [f, buffer] = prepareAudioBuffer(file);\n\n\t\t// Fetch transcript and speakers\n\t\tconst hasDiarization = speaker && await canDiarization();\n\t\tconst [transcript, speakers] = await Promise.all([\n\t\t\twhisperPipeline(buffer, {return_timestamps: speaker ? 'word' : false}),\n\t\t\t(!speaker || !token || !hasDiarization) ? Promise.resolve(): runDiarization(f, modelDir, token),\n\t\t]);\n\t\tif(file != f) rmSync(f, { recursive: true, force: true });\n\n\t\t// Return any results / errors if no more processing required\n\t\tconst text = transcript.text?.trim() || null;\n\t\tif(!speaker) return parentPort?.postMessage({ text });\n\t\tif(!token) return parentPort?.postMessage({ text, error: 'HuggingFace token required' });\n\t\tif(!hasDiarization) return parentPort?.postMessage({ text, error: 'Speaker diarization unavailable' });\n\n\t\t// Combine transcript and speakers\n\t\tconst combined = combineSpeakerTranscript(transcript.chunks || [], speakers || []);\n\t\tparentPort?.postMessage({ text: combined });\n\t} catch (err: any) {\n\t\tparentPort?.postMessage({ error: err.stack || err.message });\n\t}\n});\n"],"names":["whisperPipeline","canDiarization","resolve","proc","spawn","code","runDiarization","audioPath","dir","token","script","reject","output","data","combineSpeakerTranscript","chunks","speakers","speakerMap","speakerCount","seg","lines","currentSpeaker","currentText","chunk","time","speaker","s","speakerNum","prepareAudioBuffer","file","wav","tmp","wavefile","readFileSync","join","mkdtempSync","tmpdir","execSync","samples","left","right","buffer","i","parentPort","model","modelDir","pipeline","f","hasDiarization","transcript","rmSync","text","combined","err"],"mappings":"2QASA,IAAIA,EAEJ,eAAsBC,GAAmC,CACxD,OAAO,IAAI,QAASC,GAAY,CAC/B,MAAMC,EAAOC,EAAAA,MAAM,SAAU,CAAC,KAAM,uBAAuB,CAAC,EAC5DD,EAAK,GAAG,QAAUE,GAAiBH,EAAQG,IAAS,CAAC,CAAC,EACtDF,EAAK,GAAG,QAAS,IAAMD,EAAQ,EAAK,CAAC,CACtC,CAAC,CACF,CAEA,eAAeI,EAAeC,EAAmBC,EAAaC,EAA+B,CAC5F,MAAMC,EAAS;AAAA;AAAA;AAAA;AAAA;AAAA;AAAA,+BAMeF,CAAG;AAAA,iFAC+CC,CAAK;AAAA;AAAA;AAAA;AAAA;AAAA;AAAA;AAAA;AAAA,EAUrF,OAAO,IAAI,QAAQ,CAACP,EAASS,IAAW,CACvC,IAAIC,EAAS,GACb,MAAMT,EAAOC,EAAAA,MAAM,SAAU,CAAC,KAAMM,EAAQH,CAAS,CAAC,EACtDJ,EAAK,OAAO,GAAG,OAASU,GAAiBD,GAAUC,EAAK,UAAU,EAClEV,EAAK,OAAO,GAAG,OAASU,GAAiB,QAAQ,MAAMA,EAAK,SAAA,CAAU,CAAC,EACvEV,EAAK,GAAG,QAAUE,GAAiB,CAClC,GAAGA,IAAS,EACX,GAAI,CACHH,EAAQ,KAAK,MAAMU,CAAM,CAAC,CAC3B,MAAc,CACbD,EAAO,IAAI,MAAM,oCAAoC,CAAC,CACvD,MAEAA,EAAO,IAAI,MAAM,mCAAmCN,CAAI,EAAE,CAAC,CAE7D,CAAC,EACDF,EAAK,GAAG,QAASQ,CAAM,CACxB,CAAC,CACF,CAEA,SAASG,EAAyBC,EAAeC,EAAyB,CACzE,MAAMC,MAAiB,IACvB,IAAIC,EAAe,EACnBF,EAAS,QAASG,GAAa,CAC1BF,EAAW,IAAIE,EAAI,OAAO,GAAGF,EAAW,IAAIE,EAAI,QAAS,EAAED,CAAY,CAC5E,CAAC,EAED,MAAME,EAAkB,CAAA,EACxB,IAAIC,EAAiB,GACjBC,EAAc,GAClB,OAAAP,EAAO,QAASQ,GAAe,CAC9B,MAAMC,EAAOD,EAAM,UAAU,CAAC,EACxBE,EAAUT,EAAS,KAAMU,GAAWF,GAAQE,EAAE,OAASF,GAAQE,EAAE,GAAG,EACpEC,EAAaF,EAAUR,EAAW,IAAIQ,EAAQ,OAAO,EAAI,EAC3DE,IAAeN,GACfC,KAAmB,KAAK,YAAYD,CAAc,MAAMC,EAAY,KAAA,CAAM,EAAE,EAC/ED,EAAiBM,EACjBL,EAAcC,EAAM,MAEpBD,GAAeC,EAAM,IAEvB,CAAC,EACED,KAAmB,KAAK,YAAYD,CAAc,MAAMC,EAAY,KAAA,CAAM,EAAE,EACxEF,EAAM,KAAK;AAAA,CAAI,CACvB,CAEA,SAASQ,EAAmBC,EAAsC,CACjE,IAAIC,EAAUC,EACd,GAAI,CACHD,EAAM,IAAIE,EAAS,SAASC,EAAAA,aAAaJ,CAAI,CAAC,CAC/C,MAAa,CACZE,EAAMG,EAAAA,KAAKC,EAAAA,YAAYD,EAAAA,KAAKE,EAAAA,OAAA,EAAU,QAAQ,CAAC,EAAG,eAAe,EACjEC,WAAS,cAAcR,CAAI,6BAA6BE,CAAG,IAAK,CAAE,MAAO,SAAU,EACnFD,EAAM,IAAIE,EAAS,SAASC,EAAAA,aAAaF,CAAG,CAAC,CAC9C,QAAA,CACCD,EAAI,WAAW,KAAK,EACpBA,EAAI,aAAa,IAAK,EACtB,MAAMQ,EAAUR,EAAI,WAAA,EACpB,GAAG,MAAM,QAAQQ,CAAO,EAAG,CAC1B,MAAMC,EAAOD,EAAQ,CAAC,EAChBE,EAAQF,EAAQ,CAAC,EACjBG,EAAS,IAAI,aAAaF,EAAK,MAAM,EAC3C,QAASG,EAAI,EAAGA,EAAIH,EAAK,OAAQG,IAAKD,EAAOC,CAAC,GAAKH,EAAKG,CAAC,EAAIF,EAAME,CAAC,GAAK,EACzE,MAAO,CAACX,GAAOF,EAAMY,CAAM,CAC5B,CACA,MAAO,CAACV,GAAOF,EAAMS,CAAO,CAC7B,CACD,CAEAK,EAAAA,YAAY,GAAG,UAAW,MAAO,CAAE,KAAAd,EAAM,QAAAJ,EAAS,MAAAmB,EAAO,SAAAC,EAAU,MAAApC,KAAY,CAC9E,GAAI,CACCT,IAAiBA,EAAkB,MAAM8C,EAAAA,SAAS,+BAAgC,UAAUF,CAAK,GAAI,CAAC,UAAWC,EAAU,UAAW,GAAK,GAG/I,KAAM,CAACE,EAAGN,CAAM,EAAIb,EAAmBC,CAAI,EAGrCmB,EAAiBvB,GAAW,MAAMxB,EAAA,EAClC,CAACgD,EAAYjC,CAAQ,EAAI,MAAM,QAAQ,IAAI,CAChDhB,EAAgByC,EAAQ,CAAC,kBAAmBhB,EAAU,OAAS,GAAM,EACpE,CAACA,GAAW,CAAChB,GAAS,CAACuC,EAAkB,QAAQ,QAAA,EAAW1C,EAAeyC,EAAGF,EAAUpC,CAAK,CAAA,CAC9F,EACEoB,GAAQkB,GAAGG,EAAAA,OAAOH,EAAG,CAAE,UAAW,GAAM,MAAO,GAAM,EAGxD,MAAMI,EAAOF,EAAW,MAAM,KAAA,GAAU,KACxC,GAAG,CAACxB,EAAS,OAAOkB,EAAAA,YAAY,YAAY,CAAE,KAAAQ,EAAM,EACpD,GAAG,CAAC1C,EAAO,OAAOkC,cAAY,YAAY,CAAE,KAAAQ,EAAM,MAAO,6BAA8B,EACvF,GAAG,CAACH,EAAgB,OAAOL,cAAY,YAAY,CAAE,KAAAQ,EAAM,MAAO,kCAAmC,EAGrG,MAAMC,EAAWtC,EAAyBmC,EAAW,QAAU,CAAA,EAAIjC,GAAY,EAAE,EACjF2B,EAAAA,YAAY,YAAY,CAAE,KAAMS,CAAA,CAAU,CAC3C,OAASC,EAAU,CAClBV,EAAAA,YAAY,YAAY,CAAE,MAAOU,EAAI,OAASA,EAAI,QAAS,CAC5D,CACD,CAAC"}
+{"version":3,"file":"asr.js","sources":["../src/asr.ts"],"sourcesContent":["import { pipeline } from '@xenova/transformers';\nimport { parentPort } from 'worker_threads';\nimport { spawn } from 'node:child_process';\nimport { execSync } from 'node:child_process';\nimport { mkdtempSync, rmSync, readFileSync } from 'node:fs';\nimport { join } from 'node:path';\nimport { tmpdir } from 'node:os';\nimport wavefile from 'wavefile';\n\nlet whisperPipeline: any;\n\nexport async function canDiarization(): Promise<string | null> {\n\tconst checkPython = (cmd: string) => {\n\t\treturn new Promise<boolean>((resolve) => {\n\t\t\tconst proc = spawn(cmd, ['-c', 'import pyannote.audio']);\n\t\t\tproc.on('close', (code: number) => resolve(code === 0));\n\t\t\tproc.on('error', () => resolve(false));\n\t\t});\n\t};\n\tif(await checkPython('python3')) return 'python3';\n\tif(await checkPython('python')) return 'python';\n\treturn null;\n}\n\nasync function runDiarization(binary: string, audioPath: string, dir: string, token: string): Promise<any[]> {\n\tconst script = `\nimport sys\nimport json\nimport os\nfrom pyannote.audio import Pipeline\n\nos.environ['TORCH_HOME'] = r\"${dir}\"\npipeline = Pipeline.from_pretrained(\"pyannote/speaker-diarization-3.1\", token=\"${token}\")\noutput = pipeline(sys.argv[1])\n\nsegments = []\nfor turn, speaker in output.speaker_diarization:\n    segments.append({\"start\": turn.start, \"end\": turn.end, \"speaker\": speaker})\n\nprint(json.dumps(segments))\n`;\n\n\treturn new Promise((resolve, reject) => {\n\t\tlet output = '';\n\t\tconst proc = spawn(binary, ['-c', script, audioPath]);\n\t\tproc.stdout.on('data', (data: Buffer) => output += data.toString());\n\t\tproc.stderr.on('data', (data: Buffer) => console.error(data.toString()));\n\t\tproc.on('close', (code: number) => {\n\t\t\tif(code === 0) {\n\t\t\t\ttry {\n\t\t\t\t\tresolve(JSON.parse(output));\n\t\t\t\t} catch (err) {\n\t\t\t\t\treject(new Error('Failed to parse diarization output'));\n\t\t\t\t}\n\t\t\t} else {\n\t\t\t\treject(new Error(`Python process exited with code ${code}`));\n\t\t\t}\n\t\t});\n\t\tproc.on('error', reject);\n\t});\n}\n\nfunction combineSpeakerTranscript(chunks: any[], speakers: any[]): string {\n\tconst speakerMap = new Map();\n\tlet speakerCount = 0;\n\tspeakers.forEach((seg: any) => {\n\t\tif(!speakerMap.has(seg.speaker)) speakerMap.set(seg.speaker, ++speakerCount);\n\t});\n\n\tconst lines: string[] = [];\n\tlet currentSpeaker = -1;\n\tlet currentText = '';\n\tchunks.forEach((chunk: any) => {\n\t\tconst time = chunk.timestamp[0];\n\t\tconst speaker = speakers.find((s: any) => time >= s.start && time <= s.end);\n\t\tconst speakerNum = speaker ? speakerMap.get(speaker.speaker) : 1;\n\t\tif (speakerNum !== currentSpeaker) {\n\t\t\tif(currentText) lines.push(`[Speaker ${currentSpeaker}]: ${currentText.trim()}`);\n\t\t\tcurrentSpeaker = speakerNum;\n\t\t\tcurrentText = chunk.text;\n\t\t} else {\n\t\t\tcurrentText += chunk.text;\n\t\t}\n\t});\n\tif(currentText) lines.push(`[Speaker ${currentSpeaker}]: ${currentText.trim()}`);\n\treturn lines.join('\\n');\n}\n\nfunction prepareAudioBuffer(file: string): [string, Float32Array] {\n\tlet wav: any, tmp;\n\ttry {\n\t\twav = new wavefile.WaveFile(readFileSync(file));\n\t} catch(err) {\n\t\ttmp = join(mkdtempSync(join(tmpdir(), 'audio-')), 'converted.wav');\n\t\texecSync(`ffmpeg -i \"${file}\" -ar 16000 -ac 1 -f wav \"${tmp}\"`, { stdio: 'ignore' });\n\t\twav = new wavefile.WaveFile(readFileSync(tmp));\n\t} finally {\n\t\twav.toBitDepth('32f');\n\t\twav.toSampleRate(16000);\n\t\tconst samples = wav.getSamples();\n\t\tif(Array.isArray(samples)) {\n\t\t\tconst left = samples[0];\n\t\t\tconst right = samples[1];\n\t\t\tconst buffer = new Float32Array(left.length);\n\t\t\tfor (let i = 0; i < left.length; i++) buffer[i] = (left[i] + right[i]) / 2;\n\t\t\treturn [tmp || file, buffer];\n\t\t}\n\t\treturn [tmp || file, samples];\n\t}\n}\n\nparentPort?.on('message', async ({ file, speaker, model, modelDir, token }) => {\n\ttry {\n\t\tif(!whisperPipeline) whisperPipeline = await pipeline('automatic-speech-recognition', `Xenova/${model}`, {cache_dir: modelDir, quantized: true});\n\n\t\t// Prepare audio file\n\t\tconst [f, buffer] = prepareAudioBuffer(file);\n\n\t\t// Fetch transcript and speakers\n\t\tconst hasDiarization = await canDiarization();\n\t\tconst [transcript, speakers] = await Promise.all([\n\t\t\twhisperPipeline(buffer, {return_timestamps: speaker ? 'word' : false}),\n\t\t\t(!speaker || !token || !hasDiarization) ? Promise.resolve(): runDiarization(hasDiarization, f, modelDir, token),\n\t\t]);\n\t\tif(file != f) rmSync(f, { recursive: true, force: true });\n\n\t\t// Return any results / errors if no more processing required\n\t\tconst text = transcript.text?.trim() || null;\n\t\tif(!speaker) return parentPort?.postMessage({ text });\n\t\tif(!token) return parentPort?.postMessage({ text, error: 'HuggingFace token required' });\n\t\tif(!hasDiarization) return parentPort?.postMessage({ text, error: 'Speaker diarization unavailable' });\n\n\t\t// Combine transcript and speakers\n\t\tconst combined = combineSpeakerTranscript(transcript.chunks || [], speakers || []);\n\t\tparentPort?.postMessage({ text: combined });\n\t} catch (err: any) {\n\t\tparentPort?.postMessage({ error: err.stack || err.message });\n\t}\n});\n"],"names":["whisperPipeline","canDiarization","checkPython","cmd","resolve","proc","spawn","code","runDiarization","binary","audioPath","dir","token","script","reject","output","data","combineSpeakerTranscript","chunks","speakers","speakerMap","speakerCount","seg","lines","currentSpeaker","currentText","chunk","time","speaker","s","speakerNum","prepareAudioBuffer","file","wav","tmp","wavefile","readFileSync","join","mkdtempSync","tmpdir","execSync","samples","left","right","buffer","i","parentPort","model","modelDir","pipeline","f","hasDiarization","transcript","rmSync","text","combined","err"],"mappings":"2QASA,IAAIA,EAEJ,eAAsBC,GAAyC,CAC9D,MAAMC,EAAeC,GACb,IAAI,QAAkBC,GAAY,CACxC,MAAMC,EAAOC,EAAAA,MAAMH,EAAK,CAAC,KAAM,uBAAuB,CAAC,EACvDE,EAAK,GAAG,QAAUE,GAAiBH,EAAQG,IAAS,CAAC,CAAC,EACtDF,EAAK,GAAG,QAAS,IAAMD,EAAQ,EAAK,CAAC,CACtC,CAAC,EAEF,OAAG,MAAMF,EAAY,SAAS,EAAU,UACrC,MAAMA,EAAY,QAAQ,EAAU,SAChC,IACR,CAEA,eAAeM,EAAeC,EAAgBC,EAAmBC,EAAaC,EAA+B,CAC5G,MAAMC,EAAS;AAAA;AAAA;AAAA;AAAA;AAAA;AAAA,+BAMeF,CAAG;AAAA,iFAC+CC,CAAK;AAAA;AAAA;AAAA;AAAA;AAAA;AAAA;AAAA;AAAA,EAUrF,OAAO,IAAI,QAAQ,CAACR,EAASU,IAAW,CACvC,IAAIC,EAAS,GACb,MAAMV,EAAOC,EAAAA,MAAMG,EAAQ,CAAC,KAAMI,EAAQH,CAAS,CAAC,EACpDL,EAAK,OAAO,GAAG,OAASW,GAAiBD,GAAUC,EAAK,UAAU,EAClEX,EAAK,OAAO,GAAG,OAASW,GAAiB,QAAQ,MAAMA,EAAK,SAAA,CAAU,CAAC,EACvEX,EAAK,GAAG,QAAUE,GAAiB,CAClC,GAAGA,IAAS,EACX,GAAI,CACHH,EAAQ,KAAK,MAAMW,CAAM,CAAC,CAC3B,MAAc,CACbD,EAAO,IAAI,MAAM,oCAAoC,CAAC,CACvD,MAEAA,EAAO,IAAI,MAAM,mCAAmCP,CAAI,EAAE,CAAC,CAE7D,CAAC,EACDF,EAAK,GAAG,QAASS,CAAM,CACxB,CAAC,CACF,CAEA,SAASG,EAAyBC,EAAeC,EAAyB,CACzE,MAAMC,MAAiB,IACvB,IAAIC,EAAe,EACnBF,EAAS,QAASG,GAAa,CAC1BF,EAAW,IAAIE,EAAI,OAAO,GAAGF,EAAW,IAAIE,EAAI,QAAS,EAAED,CAAY,CAC5E,CAAC,EAED,MAAME,EAAkB,CAAA,EACxB,IAAIC,EAAiB,GACjBC,EAAc,GAClB,OAAAP,EAAO,QAASQ,GAAe,CAC9B,MAAMC,EAAOD,EAAM,UAAU,CAAC,EACxBE,EAAUT,EAAS,KAAMU,GAAWF,GAAQE,EAAE,OAASF,GAAQE,EAAE,GAAG,EACpEC,EAAaF,EAAUR,EAAW,IAAIQ,EAAQ,OAAO,EAAI,EAC3DE,IAAeN,GACfC,KAAmB,KAAK,YAAYD,CAAc,MAAMC,EAAY,KAAA,CAAM,EAAE,EAC/ED,EAAiBM,EACjBL,EAAcC,EAAM,MAEpBD,GAAeC,EAAM,IAEvB,CAAC,EACED,KAAmB,KAAK,YAAYD,CAAc,MAAMC,EAAY,KAAA,CAAM,EAAE,EACxEF,EAAM,KAAK;AAAA,CAAI,CACvB,CAEA,SAASQ,EAAmBC,EAAsC,CACjE,IAAIC,EAAUC,EACd,GAAI,CACHD,EAAM,IAAIE,EAAS,SAASC,EAAAA,aAAaJ,CAAI,CAAC,CAC/C,MAAa,CACZE,EAAMG,EAAAA,KAAKC,EAAAA,YAAYD,EAAAA,KAAKE,EAAAA,OAAA,EAAU,QAAQ,CAAC,EAAG,eAAe,EACjEC,WAAS,cAAcR,CAAI,6BAA6BE,CAAG,IAAK,CAAE,MAAO,SAAU,EACnFD,EAAM,IAAIE,EAAS,SAASC,EAAAA,aAAaF,CAAG,CAAC,CAC9C,QAAA,CACCD,EAAI,WAAW,KAAK,EACpBA,EAAI,aAAa,IAAK,EACtB,MAAMQ,EAAUR,EAAI,WAAA,EACpB,GAAG,MAAM,QAAQQ,CAAO,EAAG,CAC1B,MAAMC,EAAOD,EAAQ,CAAC,EAChBE,EAAQF,EAAQ,CAAC,EACjBG,EAAS,IAAI,aAAaF,EAAK,MAAM,EAC3C,QAASG,EAAI,EAAGA,EAAIH,EAAK,OAAQG,IAAKD,EAAOC,CAAC,GAAKH,EAAKG,CAAC,EAAIF,EAAME,CAAC,GAAK,EACzE,MAAO,CAACX,GAAOF,EAAMY,CAAM,CAC5B,CACA,MAAO,CAACV,GAAOF,EAAMS,CAAO,CAC7B,CACD,CAEAK,EAAAA,YAAY,GAAG,UAAW,MAAO,CAAE,KAAAd,EAAM,QAAAJ,EAAS,MAAAmB,EAAO,SAAAC,EAAU,MAAApC,KAAY,CAC9E,GAAI,CACCZ,IAAiBA,EAAkB,MAAMiD,EAAAA,SAAS,+BAAgC,UAAUF,CAAK,GAAI,CAAC,UAAWC,EAAU,UAAW,GAAK,GAG/I,KAAM,CAACE,EAAGN,CAAM,EAAIb,EAAmBC,CAAI,EAGrCmB,EAAiB,MAAMlD,EAAA,EACvB,CAACmD,EAAYjC,CAAQ,EAAI,MAAM,QAAQ,IAAI,CAChDnB,EAAgB4C,EAAQ,CAAC,kBAAmBhB,EAAU,OAAS,GAAM,EACpE,CAACA,GAAW,CAAChB,GAAS,CAACuC,EAAkB,QAAQ,QAAA,EAAW3C,EAAe2C,EAAgBD,EAAGF,EAAUpC,CAAK,CAAA,CAC9G,EACEoB,GAAQkB,GAAGG,EAAAA,OAAOH,EAAG,CAAE,UAAW,GAAM,MAAO,GAAM,EAGxD,MAAMI,EAAOF,EAAW,MAAM,KAAA,GAAU,KACxC,GAAG,CAACxB,EAAS,OAAOkB,EAAAA,YAAY,YAAY,CAAE,KAAAQ,EAAM,EACpD,GAAG,CAAC1C,EAAO,OAAOkC,cAAY,YAAY,CAAE,KAAAQ,EAAM,MAAO,6BAA8B,EACvF,GAAG,CAACH,EAAgB,OAAOL,cAAY,YAAY,CAAE,KAAAQ,EAAM,MAAO,kCAAmC,EAGrG,MAAMC,EAAWtC,EAAyBmC,EAAW,QAAU,CAAA,EAAIjC,GAAY,EAAE,EACjF2B,EAAAA,YAAY,YAAY,CAAE,KAAMS,CAAA,CAAU,CAC3C,OAASC,EAAU,CAClBV,EAAAA,YAAY,YAAY,CAAE,MAAOU,EAAI,OAASA,EAAI,QAAS,CAC5D,CACD,CAAC"}

package/dist/asr.mjs

@@ -2,24 +2,25 @@ import { pipeline as w } from "@xenova/transformers";
 import { parentPort as m } from "worker_threads";
 import { execSync as k, spawn as h } from "node:child_process";
 import { rmSync as S, readFileSync as d, mkdtempSync as v } from "node:fs";
-import { join as g } from "node:path";
-import { tmpdir as $ } from "node:os";
-import y from "wavefile";
+import { join as y } from "node:path";
+import { tmpdir as P } from "node:os";
+import g from "wavefile";
 let l;
-async function x() {
-  return new Promise((o) => {
-    const r = h("python", ["-c", "import pyannote.audio"]);
-    r.on("close", (n) => o(n === 0)), r.on("error", () => o(!1));
+async function $() {
+  const s = (r) => new Promise((t) => {
+    const n = h(r, ["-c", "import pyannote.audio"]);
+    n.on("close", (o) => t(o === 0)), n.on("error", () => t(!1));
   });
+  return await s("python3") ? "python3" : await s("python") ? "python" : null;
 }
-async function P(o, r, n) {
-  const s = `
+async function x(s, r, t, n) {
+  const o = `
 import sys
 import json
 import os
 from pyannote.audio import Pipeline
 
-os.environ['TORCH_HOME'] = r"${r}"
+os.environ['TORCH_HOME'] = r"${t}"
 pipeline = Pipeline.from_pretrained("pyannote/speaker-diarization-3.1", token="${n}")
 output = pipeline(sys.argv[1])
 
@@ -29,71 +30,71 @@ for turn, speaker in output.speaker_diarization:
 
 print(json.dumps(segments))
 `;
-  return new Promise((a, t) => {
-    let i = "";
-    const e = h("python", ["-c", s, o]);
-    e.stdout.on("data", (p) => i += p.toString()), e.stderr.on("data", (p) => console.error(p.toString())), e.on("close", (p) => {
+  return new Promise((i, a) => {
+    let e = "";
+    const c = h(s, ["-c", o, r]);
+    c.stdout.on("data", (p) => e += p.toString()), c.stderr.on("data", (p) => console.error(p.toString())), c.on("close", (p) => {
       if (p === 0)
         try {
-          a(JSON.parse(i));
+          i(JSON.parse(e));
         } catch {
-          t(new Error("Failed to parse diarization output"));
+          a(new Error("Failed to parse diarization output"));
         }
       else
-        t(new Error(`Python process exited with code ${p}`));
-    }), e.on("error", t);
+        a(new Error(`Python process exited with code ${p}`));
+    }), c.on("error", a);
   });
 }
-function z(o, r) {
-  const n = /* @__PURE__ */ new Map();
-  let s = 0;
+function z(s, r) {
+  const t = /* @__PURE__ */ new Map();
+  let n = 0;
   r.forEach((e) => {
-    n.has(e.speaker) || n.set(e.speaker, ++s);
+    t.has(e.speaker) || t.set(e.speaker, ++n);
   });
-  const a = [];
-  let t = -1, i = "";
-  return o.forEach((e) => {
-    const p = e.timestamp[0], u = r.find((f) => p >= f.start && p <= f.end), c = u ? n.get(u.speaker) : 1;
-    c !== t ? (i && a.push(`[Speaker ${t}]: ${i.trim()}`), t = c, i = e.text) : i += e.text;
-  }), i && a.push(`[Speaker ${t}]: ${i.trim()}`), a.join(`
+  const o = [];
+  let i = -1, a = "";
+  return s.forEach((e) => {
+    const c = e.timestamp[0], p = r.find((f) => c >= f.start && c <= f.end), u = p ? t.get(p.speaker) : 1;
+    u !== i ? (a && o.push(`[Speaker ${i}]: ${a.trim()}`), i = u, a = e.text) : a += e.text;
+  }), a && o.push(`[Speaker ${i}]: ${a.trim()}`), o.join(`
 `);
 }
-function M(o) {
-  let r, n;
+function M(s) {
+  let r, t;
   try {
-    r = new y.WaveFile(d(o));
+    r = new g.WaveFile(d(s));
   } catch {
-    n = g(v(g($(), "audio-")), "converted.wav"), k(`ffmpeg -i "${o}" -ar 16000 -ac 1 -f wav "${n}"`, { stdio: "ignore" }), r = new y.WaveFile(d(n));
+    t = y(v(y(P(), "audio-")), "converted.wav"), k(`ffmpeg -i "${s}" -ar 16000 -ac 1 -f wav "${t}"`, { stdio: "ignore" }), r = new g.WaveFile(d(t));
   } finally {
     r.toBitDepth("32f"), r.toSampleRate(16e3);
-    const s = r.getSamples();
-    if (Array.isArray(s)) {
-      const a = s[0], t = s[1], i = new Float32Array(a.length);
-      for (let e = 0; e < a.length; e++) i[e] = (a[e] + t[e]) / 2;
-      return [n || o, i];
+    const n = r.getSamples();
+    if (Array.isArray(n)) {
+      const o = n[0], i = n[1], a = new Float32Array(o.length);
+      for (let e = 0; e < o.length; e++) a[e] = (o[e] + i[e]) / 2;
+      return [t || s, a];
     }
-    return [n || o, s];
+    return [t || s, n];
   }
 }
-m?.on("message", async ({ file: o, speaker: r, model: n, modelDir: s, token: a }) => {
+m?.on("message", async ({ file: s, speaker: r, model: t, modelDir: n, token: o }) => {
   try {
-    l || (l = await w("automatic-speech-recognition", `Xenova/${n}`, { cache_dir: s, quantized: !0 }));
-    const [t, i] = M(o), e = r && await x(), [p, u] = await Promise.all([
-      l(i, { return_timestamps: r ? "word" : !1 }),
-      !r || !a || !e ? Promise.resolve() : P(t, s, a)
+    l || (l = await w("automatic-speech-recognition", `Xenova/${t}`, { cache_dir: n, quantized: !0 }));
+    const [i, a] = M(s), e = await $(), [c, p] = await Promise.all([
+      l(a, { return_timestamps: r ? "word" : !1 }),
+      !r || !o || !e ? Promise.resolve() : x(e, i, n, o)
     ]);
-    o != t && S(t, { recursive: !0, force: !0 });
-    const c = p.text?.trim() || null;
-    if (!r) return m?.postMessage({ text: c });
-    if (!a) return m?.postMessage({ text: c, error: "HuggingFace token required" });
-    if (!e) return m?.postMessage({ text: c, error: "Speaker diarization unavailable" });
-    const f = z(p.chunks || [], u || []);
+    s != i && S(i, { recursive: !0, force: !0 });
+    const u = c.text?.trim() || null;
+    if (!r) return m?.postMessage({ text: u });
+    if (!o) return m?.postMessage({ text: u, error: "HuggingFace token required" });
+    if (!e) return m?.postMessage({ text: u, error: "Speaker diarization unavailable" });
+    const f = z(c.chunks || [], p || []);
     m?.postMessage({ text: f });
-  } catch (t) {
-    m?.postMessage({ error: t.stack || t.message });
+  } catch (i) {
+    m?.postMessage({ error: i.stack || i.message });
   }
 });
 export {
-  x as canDiarization
+  $ as canDiarization
 };
 //# sourceMappingURL=asr.mjs.map

package/dist/asr.mjs.map

@@ -1 +1 @@
-{"version":3,"file":"asr.mjs","sources":["../src/asr.ts"],"sourcesContent":["import { pipeline } from '@xenova/transformers';\nimport { parentPort } from 'worker_threads';\nimport { spawn } from 'node:child_process';\nimport { execSync } from 'node:child_process';\nimport { mkdtempSync, rmSync, readFileSync } from 'node:fs';\nimport { join } from 'node:path';\nimport { tmpdir } from 'node:os';\nimport wavefile from 'wavefile';\n\nlet whisperPipeline: any;\n\nexport async function canDiarization(): Promise<boolean> {\n\treturn new Promise((resolve) => {\n\t\tconst proc = spawn('python', ['-c', 'import pyannote.audio']);\n\t\tproc.on('close', (code: number) => resolve(code === 0));\n\t\tproc.on('error', () => resolve(false));\n\t});\n}\n\nasync function runDiarization(audioPath: string, dir: string, token: string): Promise<any[]> {\n\tconst script = `\nimport sys\nimport json\nimport os\nfrom pyannote.audio import Pipeline\n\nos.environ['TORCH_HOME'] = r\"${dir}\"\npipeline = Pipeline.from_pretrained(\"pyannote/speaker-diarization-3.1\", token=\"${token}\")\noutput = pipeline(sys.argv[1])\n\nsegments = []\nfor turn, speaker in output.speaker_diarization:\n    segments.append({\"start\": turn.start, \"end\": turn.end, \"speaker\": speaker})\n\nprint(json.dumps(segments))\n`;\n\n\treturn new Promise((resolve, reject) => {\n\t\tlet output = '';\n\t\tconst proc = spawn('python', ['-c', script, audioPath]);\n\t\tproc.stdout.on('data', (data: Buffer) => output += data.toString());\n\t\tproc.stderr.on('data', (data: Buffer) => console.error(data.toString()));\n\t\tproc.on('close', (code: number) => {\n\t\t\tif(code === 0) {\n\t\t\t\ttry {\n\t\t\t\t\tresolve(JSON.parse(output));\n\t\t\t\t} catch (err) {\n\t\t\t\t\treject(new Error('Failed to parse diarization output'));\n\t\t\t\t}\n\t\t\t} else {\n\t\t\t\treject(new Error(`Python process exited with code ${code}`));\n\t\t\t}\n\t\t});\n\t\tproc.on('error', reject);\n\t});\n}\n\nfunction combineSpeakerTranscript(chunks: any[], speakers: any[]): string {\n\tconst speakerMap = new Map();\n\tlet speakerCount = 0;\n\tspeakers.forEach((seg: any) => {\n\t\tif(!speakerMap.has(seg.speaker)) speakerMap.set(seg.speaker, ++speakerCount);\n\t});\n\n\tconst lines: string[] = [];\n\tlet currentSpeaker = -1;\n\tlet currentText = '';\n\tchunks.forEach((chunk: any) => {\n\t\tconst time = chunk.timestamp[0];\n\t\tconst speaker = speakers.find((s: any) => time >= s.start && time <= s.end);\n\t\tconst speakerNum = speaker ? speakerMap.get(speaker.speaker) : 1;\n\t\tif (speakerNum !== currentSpeaker) {\n\t\t\tif(currentText) lines.push(`[Speaker ${currentSpeaker}]: ${currentText.trim()}`);\n\t\t\tcurrentSpeaker = speakerNum;\n\t\t\tcurrentText = chunk.text;\n\t\t} else {\n\t\t\tcurrentText += chunk.text;\n\t\t}\n\t});\n\tif(currentText) lines.push(`[Speaker ${currentSpeaker}]: ${currentText.trim()}`);\n\treturn lines.join('\\n');\n}\n\nfunction prepareAudioBuffer(file: string): [string, Float32Array] {\n\tlet wav: any, tmp;\n\ttry {\n\t\twav = new wavefile.WaveFile(readFileSync(file));\n\t} catch(err) {\n\t\ttmp = join(mkdtempSync(join(tmpdir(), 'audio-')), 'converted.wav');\n\t\texecSync(`ffmpeg -i \"${file}\" -ar 16000 -ac 1 -f wav \"${tmp}\"`, { stdio: 'ignore' });\n\t\twav = new wavefile.WaveFile(readFileSync(tmp));\n\t} finally {\n\t\twav.toBitDepth('32f');\n\t\twav.toSampleRate(16000);\n\t\tconst samples = wav.getSamples();\n\t\tif(Array.isArray(samples)) {\n\t\t\tconst left = samples[0];\n\t\t\tconst right = samples[1];\n\t\t\tconst buffer = new Float32Array(left.length);\n\t\t\tfor (let i = 0; i < left.length; i++) buffer[i] = (left[i] + right[i]) / 2;\n\t\t\treturn [tmp || file, buffer];\n\t\t}\n\t\treturn [tmp || file, samples];\n\t}\n}\n\nparentPort?.on('message', async ({ file, speaker, model, modelDir, token }) => {\n\ttry {\n\t\tif(!whisperPipeline) whisperPipeline = await pipeline('automatic-speech-recognition', `Xenova/${model}`, {cache_dir: modelDir, quantized: true});\n\n\t\t// Prepare audio file\n\t\tconst [f, buffer] = prepareAudioBuffer(file);\n\n\t\t// Fetch transcript and speakers\n\t\tconst hasDiarization = speaker && await canDiarization();\n\t\tconst [transcript, speakers] = await Promise.all([\n\t\t\twhisperPipeline(buffer, {return_timestamps: speaker ? 'word' : false}),\n\t\t\t(!speaker || !token || !hasDiarization) ? Promise.resolve(): runDiarization(f, modelDir, token),\n\t\t]);\n\t\tif(file != f) rmSync(f, { recursive: true, force: true });\n\n\t\t// Return any results / errors if no more processing required\n\t\tconst text = transcript.text?.trim() || null;\n\t\tif(!speaker) return parentPort?.postMessage({ text });\n\t\tif(!token) return parentPort?.postMessage({ text, error: 'HuggingFace token required' });\n\t\tif(!hasDiarization) return parentPort?.postMessage({ text, error: 'Speaker diarization unavailable' });\n\n\t\t// Combine transcript and speakers\n\t\tconst combined = combineSpeakerTranscript(transcript.chunks || [], speakers || []);\n\t\tparentPort?.postMessage({ text: combined });\n\t} catch (err: any) {\n\t\tparentPort?.postMessage({ error: err.stack || err.message });\n\t}\n});\n"],"names":["whisperPipeline","canDiarization","resolve","proc","spawn","code","runDiarization","audioPath","dir","token","script","reject","output","data","combineSpeakerTranscript","chunks","speakers","speakerMap","speakerCount","seg","lines","currentSpeaker","currentText","chunk","time","speaker","s","speakerNum","prepareAudioBuffer","file","wav","tmp","wavefile","readFileSync","join","mkdtempSync","tmpdir","execSync","samples","left","right","buffer","i","parentPort","model","modelDir","pipeline","f","hasDiarization","transcript","rmSync","text","combined","err"],"mappings":";;;;;;;AASA,IAAIA;AAEJ,eAAsBC,IAAmC;AACxD,SAAO,IAAI,QAAQ,CAACC,MAAY;AAC/B,UAAMC,IAAOC,EAAM,UAAU,CAAC,MAAM,uBAAuB,CAAC;AAC5D,IAAAD,EAAK,GAAG,SAAS,CAACE,MAAiBH,EAAQG,MAAS,CAAC,CAAC,GACtDF,EAAK,GAAG,SAAS,MAAMD,EAAQ,EAAK,CAAC;AAAA,EACtC,CAAC;AACF;AAEA,eAAeI,EAAeC,GAAmBC,GAAaC,GAA+B;AAC5F,QAAMC,IAAS;AAAA;AAAA;AAAA;AAAA;AAAA;AAAA,+BAMeF,CAAG;AAAA,iFAC+CC,CAAK;AAAA;AAAA;AAAA;AAAA;AAAA;AAAA;AAAA;AAAA;AAUrF,SAAO,IAAI,QAAQ,CAACP,GAASS,MAAW;AACvC,QAAIC,IAAS;AACb,UAAMT,IAAOC,EAAM,UAAU,CAAC,MAAMM,GAAQH,CAAS,CAAC;AACtD,IAAAJ,EAAK,OAAO,GAAG,QAAQ,CAACU,MAAiBD,KAAUC,EAAK,UAAU,GAClEV,EAAK,OAAO,GAAG,QAAQ,CAACU,MAAiB,QAAQ,MAAMA,EAAK,SAAA,CAAU,CAAC,GACvEV,EAAK,GAAG,SAAS,CAACE,MAAiB;AAClC,UAAGA,MAAS;AACX,YAAI;AACH,UAAAH,EAAQ,KAAK,MAAMU,CAAM,CAAC;AAAA,QAC3B,QAAc;AACb,UAAAD,EAAO,IAAI,MAAM,oCAAoC,CAAC;AAAA,QACvD;AAAA;AAEA,QAAAA,EAAO,IAAI,MAAM,mCAAmCN,CAAI,EAAE,CAAC;AAAA,IAE7D,CAAC,GACDF,EAAK,GAAG,SAASQ,CAAM;AAAA,EACxB,CAAC;AACF;AAEA,SAASG,EAAyBC,GAAeC,GAAyB;AACzE,QAAMC,wBAAiB,IAAA;AACvB,MAAIC,IAAe;AACnB,EAAAF,EAAS,QAAQ,CAACG,MAAa;AAC9B,IAAIF,EAAW,IAAIE,EAAI,OAAO,KAAGF,EAAW,IAAIE,EAAI,SAAS,EAAED,CAAY;AAAA,EAC5E,CAAC;AAED,QAAME,IAAkB,CAAA;AACxB,MAAIC,IAAiB,IACjBC,IAAc;AAClB,SAAAP,EAAO,QAAQ,CAACQ,MAAe;AAC9B,UAAMC,IAAOD,EAAM,UAAU,CAAC,GACxBE,IAAUT,EAAS,KAAK,CAACU,MAAWF,KAAQE,EAAE,SAASF,KAAQE,EAAE,GAAG,GACpEC,IAAaF,IAAUR,EAAW,IAAIQ,EAAQ,OAAO,IAAI;AAC/D,IAAIE,MAAeN,KACfC,OAAmB,KAAK,YAAYD,CAAc,MAAMC,EAAY,KAAA,CAAM,EAAE,GAC/ED,IAAiBM,GACjBL,IAAcC,EAAM,QAEpBD,KAAeC,EAAM;AAAA,EAEvB,CAAC,GACED,OAAmB,KAAK,YAAYD,CAAc,MAAMC,EAAY,KAAA,CAAM,EAAE,GACxEF,EAAM,KAAK;AAAA,CAAI;AACvB;AAEA,SAASQ,EAAmBC,GAAsC;AACjE,MAAIC,GAAUC;AACd,MAAI;AACH,IAAAD,IAAM,IAAIE,EAAS,SAASC,EAAaJ,CAAI,CAAC;AAAA,EAC/C,QAAa;AACZ,IAAAE,IAAMG,EAAKC,EAAYD,EAAKE,EAAA,GAAU,QAAQ,CAAC,GAAG,eAAe,GACjEC,EAAS,cAAcR,CAAI,6BAA6BE,CAAG,KAAK,EAAE,OAAO,UAAU,GACnFD,IAAM,IAAIE,EAAS,SAASC,EAAaF,CAAG,CAAC;AAAA,EAC9C,UAAA;AACC,IAAAD,EAAI,WAAW,KAAK,GACpBA,EAAI,aAAa,IAAK;AACtB,UAAMQ,IAAUR,EAAI,WAAA;AACpB,QAAG,MAAM,QAAQQ,CAAO,GAAG;AAC1B,YAAMC,IAAOD,EAAQ,CAAC,GAChBE,IAAQF,EAAQ,CAAC,GACjBG,IAAS,IAAI,aAAaF,EAAK,MAAM;AAC3C,eAASG,IAAI,GAAGA,IAAIH,EAAK,QAAQG,IAAK,CAAAD,EAAOC,CAAC,KAAKH,EAAKG,CAAC,IAAIF,EAAME,CAAC,KAAK;AACzE,aAAO,CAACX,KAAOF,GAAMY,CAAM;AAAA,IAC5B;AACA,WAAO,CAACV,KAAOF,GAAMS,CAAO;AAAA,EAC7B;AACD;AAEAK,GAAY,GAAG,WAAW,OAAO,EAAE,MAAAd,GAAM,SAAAJ,GAAS,OAAAmB,GAAO,UAAAC,GAAU,OAAApC,QAAY;AAC9E,MAAI;AACH,IAAIT,MAAiBA,IAAkB,MAAM8C,EAAS,gCAAgC,UAAUF,CAAK,IAAI,EAAC,WAAWC,GAAU,WAAW,IAAK;AAG/I,UAAM,CAACE,GAAGN,CAAM,IAAIb,EAAmBC,CAAI,GAGrCmB,IAAiBvB,KAAW,MAAMxB,EAAA,GAClC,CAACgD,GAAYjC,CAAQ,IAAI,MAAM,QAAQ,IAAI;AAAA,MAChDhB,EAAgByC,GAAQ,EAAC,mBAAmBhB,IAAU,SAAS,IAAM;AAAA,MACpE,CAACA,KAAW,CAAChB,KAAS,CAACuC,IAAkB,QAAQ,QAAA,IAAW1C,EAAeyC,GAAGF,GAAUpC,CAAK;AAAA,IAAA,CAC9F;AACD,IAAGoB,KAAQkB,KAAGG,EAAOH,GAAG,EAAE,WAAW,IAAM,OAAO,IAAM;AAGxD,UAAMI,IAAOF,EAAW,MAAM,KAAA,KAAU;AACxC,QAAG,CAACxB,EAAS,QAAOkB,GAAY,YAAY,EAAE,MAAAQ,GAAM;AACpD,QAAG,CAAC1C,EAAO,QAAOkC,GAAY,YAAY,EAAE,MAAAQ,GAAM,OAAO,8BAA8B;AACvF,QAAG,CAACH,EAAgB,QAAOL,GAAY,YAAY,EAAE,MAAAQ,GAAM,OAAO,mCAAmC;AAGrG,UAAMC,IAAWtC,EAAyBmC,EAAW,UAAU,CAAA,GAAIjC,KAAY,EAAE;AACjF,IAAA2B,GAAY,YAAY,EAAE,MAAMS,EAAA,CAAU;AAAA,EAC3C,SAASC,GAAU;AAClB,IAAAV,GAAY,YAAY,EAAE,OAAOU,EAAI,SAASA,EAAI,SAAS;AAAA,EAC5D;AACD,CAAC;"}
+{"version":3,"file":"asr.mjs","sources":["../src/asr.ts"],"sourcesContent":["import { pipeline } from '@xenova/transformers';\nimport { parentPort } from 'worker_threads';\nimport { spawn } from 'node:child_process';\nimport { execSync } from 'node:child_process';\nimport { mkdtempSync, rmSync, readFileSync } from 'node:fs';\nimport { join } from 'node:path';\nimport { tmpdir } from 'node:os';\nimport wavefile from 'wavefile';\n\nlet whisperPipeline: any;\n\nexport async function canDiarization(): Promise<string | null> {\n\tconst checkPython = (cmd: string) => {\n\t\treturn new Promise<boolean>((resolve) => {\n\t\t\tconst proc = spawn(cmd, ['-c', 'import pyannote.audio']);\n\t\t\tproc.on('close', (code: number) => resolve(code === 0));\n\t\t\tproc.on('error', () => resolve(false));\n\t\t});\n\t};\n\tif(await checkPython('python3')) return 'python3';\n\tif(await checkPython('python')) return 'python';\n\treturn null;\n}\n\nasync function runDiarization(binary: string, audioPath: string, dir: string, token: string): Promise<any[]> {\n\tconst script = `\nimport sys\nimport json\nimport os\nfrom pyannote.audio import Pipeline\n\nos.environ['TORCH_HOME'] = r\"${dir}\"\npipeline = Pipeline.from_pretrained(\"pyannote/speaker-diarization-3.1\", token=\"${token}\")\noutput = pipeline(sys.argv[1])\n\nsegments = []\nfor turn, speaker in output.speaker_diarization:\n    segments.append({\"start\": turn.start, \"end\": turn.end, \"speaker\": speaker})\n\nprint(json.dumps(segments))\n`;\n\n\treturn new Promise((resolve, reject) => {\n\t\tlet output = '';\n\t\tconst proc = spawn(binary, ['-c', script, audioPath]);\n\t\tproc.stdout.on('data', (data: Buffer) => output += data.toString());\n\t\tproc.stderr.on('data', (data: Buffer) => console.error(data.toString()));\n\t\tproc.on('close', (code: number) => {\n\t\t\tif(code === 0) {\n\t\t\t\ttry {\n\t\t\t\t\tresolve(JSON.parse(output));\n\t\t\t\t} catch (err) {\n\t\t\t\t\treject(new Error('Failed to parse diarization output'));\n\t\t\t\t}\n\t\t\t} else {\n\t\t\t\treject(new Error(`Python process exited with code ${code}`));\n\t\t\t}\n\t\t});\n\t\tproc.on('error', reject);\n\t});\n}\n\nfunction combineSpeakerTranscript(chunks: any[], speakers: any[]): string {\n\tconst speakerMap = new Map();\n\tlet speakerCount = 0;\n\tspeakers.forEach((seg: any) => {\n\t\tif(!speakerMap.has(seg.speaker)) speakerMap.set(seg.speaker, ++speakerCount);\n\t});\n\n\tconst lines: string[] = [];\n\tlet currentSpeaker = -1;\n\tlet currentText = '';\n\tchunks.forEach((chunk: any) => {\n\t\tconst time = chunk.timestamp[0];\n\t\tconst speaker = speakers.find((s: any) => time >= s.start && time <= s.end);\n\t\tconst speakerNum = speaker ? speakerMap.get(speaker.speaker) : 1;\n\t\tif (speakerNum !== currentSpeaker) {\n\t\t\tif(currentText) lines.push(`[Speaker ${currentSpeaker}]: ${currentText.trim()}`);\n\t\t\tcurrentSpeaker = speakerNum;\n\t\t\tcurrentText = chunk.text;\n\t\t} else {\n\t\t\tcurrentText += chunk.text;\n\t\t}\n\t});\n\tif(currentText) lines.push(`[Speaker ${currentSpeaker}]: ${currentText.trim()}`);\n\treturn lines.join('\\n');\n}\n\nfunction prepareAudioBuffer(file: string): [string, Float32Array] {\n\tlet wav: any, tmp;\n\ttry {\n\t\twav = new wavefile.WaveFile(readFileSync(file));\n\t} catch(err) {\n\t\ttmp = join(mkdtempSync(join(tmpdir(), 'audio-')), 'converted.wav');\n\t\texecSync(`ffmpeg -i \"${file}\" -ar 16000 -ac 1 -f wav \"${tmp}\"`, { stdio: 'ignore' });\n\t\twav = new wavefile.WaveFile(readFileSync(tmp));\n\t} finally {\n\t\twav.toBitDepth('32f');\n\t\twav.toSampleRate(16000);\n\t\tconst samples = wav.getSamples();\n\t\tif(Array.isArray(samples)) {\n\t\t\tconst left = samples[0];\n\t\t\tconst right = samples[1];\n\t\t\tconst buffer = new Float32Array(left.length);\n\t\t\tfor (let i = 0; i < left.length; i++) buffer[i] = (left[i] + right[i]) / 2;\n\t\t\treturn [tmp || file, buffer];\n\t\t}\n\t\treturn [tmp || file, samples];\n\t}\n}\n\nparentPort?.on('message', async ({ file, speaker, model, modelDir, token }) => {\n\ttry {\n\t\tif(!whisperPipeline) whisperPipeline = await pipeline('automatic-speech-recognition', `Xenova/${model}`, {cache_dir: modelDir, quantized: true});\n\n\t\t// Prepare audio file\n\t\tconst [f, buffer] = prepareAudioBuffer(file);\n\n\t\t// Fetch transcript and speakers\n\t\tconst hasDiarization = await canDiarization();\n\t\tconst [transcript, speakers] = await Promise.all([\n\t\t\twhisperPipeline(buffer, {return_timestamps: speaker ? 'word' : false}),\n\t\t\t(!speaker || !token || !hasDiarization) ? Promise.resolve(): runDiarization(hasDiarization, f, modelDir, token),\n\t\t]);\n\t\tif(file != f) rmSync(f, { recursive: true, force: true });\n\n\t\t// Return any results / errors if no more processing required\n\t\tconst text = transcript.text?.trim() || null;\n\t\tif(!speaker) return parentPort?.postMessage({ text });\n\t\tif(!token) return parentPort?.postMessage({ text, error: 'HuggingFace token required' });\n\t\tif(!hasDiarization) return parentPort?.postMessage({ text, error: 'Speaker diarization unavailable' });\n\n\t\t// Combine transcript and speakers\n\t\tconst combined = combineSpeakerTranscript(transcript.chunks || [], speakers || []);\n\t\tparentPort?.postMessage({ text: combined });\n\t} catch (err: any) {\n\t\tparentPort?.postMessage({ error: err.stack || err.message });\n\t}\n});\n"],"names":["whisperPipeline","canDiarization","checkPython","cmd","resolve","proc","spawn","code","runDiarization","binary","audioPath","dir","token","script","reject","output","data","combineSpeakerTranscript","chunks","speakers","speakerMap","speakerCount","seg","lines","currentSpeaker","currentText","chunk","time","speaker","s","speakerNum","prepareAudioBuffer","file","wav","tmp","wavefile","readFileSync","join","mkdtempSync","tmpdir","execSync","samples","left","right","buffer","i","parentPort","model","modelDir","pipeline","f","hasDiarization","transcript","rmSync","text","combined","err"],"mappings":";;;;;;;AASA,IAAIA;AAEJ,eAAsBC,IAAyC;AAC9D,QAAMC,IAAc,CAACC,MACb,IAAI,QAAiB,CAACC,MAAY;AACxC,UAAMC,IAAOC,EAAMH,GAAK,CAAC,MAAM,uBAAuB,CAAC;AACvD,IAAAE,EAAK,GAAG,SAAS,CAACE,MAAiBH,EAAQG,MAAS,CAAC,CAAC,GACtDF,EAAK,GAAG,SAAS,MAAMD,EAAQ,EAAK,CAAC;AAAA,EACtC,CAAC;AAEF,SAAG,MAAMF,EAAY,SAAS,IAAU,YACrC,MAAMA,EAAY,QAAQ,IAAU,WAChC;AACR;AAEA,eAAeM,EAAeC,GAAgBC,GAAmBC,GAAaC,GAA+B;AAC5G,QAAMC,IAAS;AAAA;AAAA;AAAA;AAAA;AAAA;AAAA,+BAMeF,CAAG;AAAA,iFAC+CC,CAAK;AAAA;AAAA;AAAA;AAAA;AAAA;AAAA;AAAA;AAAA;AAUrF,SAAO,IAAI,QAAQ,CAACR,GAASU,MAAW;AACvC,QAAIC,IAAS;AACb,UAAMV,IAAOC,EAAMG,GAAQ,CAAC,MAAMI,GAAQH,CAAS,CAAC;AACpD,IAAAL,EAAK,OAAO,GAAG,QAAQ,CAACW,MAAiBD,KAAUC,EAAK,UAAU,GAClEX,EAAK,OAAO,GAAG,QAAQ,CAACW,MAAiB,QAAQ,MAAMA,EAAK,SAAA,CAAU,CAAC,GACvEX,EAAK,GAAG,SAAS,CAACE,MAAiB;AAClC,UAAGA,MAAS;AACX,YAAI;AACH,UAAAH,EAAQ,KAAK,MAAMW,CAAM,CAAC;AAAA,QAC3B,QAAc;AACb,UAAAD,EAAO,IAAI,MAAM,oCAAoC,CAAC;AAAA,QACvD;AAAA;AAEA,QAAAA,EAAO,IAAI,MAAM,mCAAmCP,CAAI,EAAE,CAAC;AAAA,IAE7D,CAAC,GACDF,EAAK,GAAG,SAASS,CAAM;AAAA,EACxB,CAAC;AACF;AAEA,SAASG,EAAyBC,GAAeC,GAAyB;AACzE,QAAMC,wBAAiB,IAAA;AACvB,MAAIC,IAAe;AACnB,EAAAF,EAAS,QAAQ,CAACG,MAAa;AAC9B,IAAIF,EAAW,IAAIE,EAAI,OAAO,KAAGF,EAAW,IAAIE,EAAI,SAAS,EAAED,CAAY;AAAA,EAC5E,CAAC;AAED,QAAME,IAAkB,CAAA;AACxB,MAAIC,IAAiB,IACjBC,IAAc;AAClB,SAAAP,EAAO,QAAQ,CAACQ,MAAe;AAC9B,UAAMC,IAAOD,EAAM,UAAU,CAAC,GACxBE,IAAUT,EAAS,KAAK,CAACU,MAAWF,KAAQE,EAAE,SAASF,KAAQE,EAAE,GAAG,GACpEC,IAAaF,IAAUR,EAAW,IAAIQ,EAAQ,OAAO,IAAI;AAC/D,IAAIE,MAAeN,KACfC,OAAmB,KAAK,YAAYD,CAAc,MAAMC,EAAY,KAAA,CAAM,EAAE,GAC/ED,IAAiBM,GACjBL,IAAcC,EAAM,QAEpBD,KAAeC,EAAM;AAAA,EAEvB,CAAC,GACED,OAAmB,KAAK,YAAYD,CAAc,MAAMC,EAAY,KAAA,CAAM,EAAE,GACxEF,EAAM,KAAK;AAAA,CAAI;AACvB;AAEA,SAASQ,EAAmBC,GAAsC;AACjE,MAAIC,GAAUC;AACd,MAAI;AACH,IAAAD,IAAM,IAAIE,EAAS,SAASC,EAAaJ,CAAI,CAAC;AAAA,EAC/C,QAAa;AACZ,IAAAE,IAAMG,EAAKC,EAAYD,EAAKE,EAAA,GAAU,QAAQ,CAAC,GAAG,eAAe,GACjEC,EAAS,cAAcR,CAAI,6BAA6BE,CAAG,KAAK,EAAE,OAAO,UAAU,GACnFD,IAAM,IAAIE,EAAS,SAASC,EAAaF,CAAG,CAAC;AAAA,EAC9C,UAAA;AACC,IAAAD,EAAI,WAAW,KAAK,GACpBA,EAAI,aAAa,IAAK;AACtB,UAAMQ,IAAUR,EAAI,WAAA;AACpB,QAAG,MAAM,QAAQQ,CAAO,GAAG;AAC1B,YAAMC,IAAOD,EAAQ,CAAC,GAChBE,IAAQF,EAAQ,CAAC,GACjBG,IAAS,IAAI,aAAaF,EAAK,MAAM;AAC3C,eAASG,IAAI,GAAGA,IAAIH,EAAK,QAAQG,IAAK,CAAAD,EAAOC,CAAC,KAAKH,EAAKG,CAAC,IAAIF,EAAME,CAAC,KAAK;AACzE,aAAO,CAACX,KAAOF,GAAMY,CAAM;AAAA,IAC5B;AACA,WAAO,CAACV,KAAOF,GAAMS,CAAO;AAAA,EAC7B;AACD;AAEAK,GAAY,GAAG,WAAW,OAAO,EAAE,MAAAd,GAAM,SAAAJ,GAAS,OAAAmB,GAAO,UAAAC,GAAU,OAAApC,QAAY;AAC9E,MAAI;AACH,IAAIZ,MAAiBA,IAAkB,MAAMiD,EAAS,gCAAgC,UAAUF,CAAK,IAAI,EAAC,WAAWC,GAAU,WAAW,IAAK;AAG/I,UAAM,CAACE,GAAGN,CAAM,IAAIb,EAAmBC,CAAI,GAGrCmB,IAAiB,MAAMlD,EAAA,GACvB,CAACmD,GAAYjC,CAAQ,IAAI,MAAM,QAAQ,IAAI;AAAA,MAChDnB,EAAgB4C,GAAQ,EAAC,mBAAmBhB,IAAU,SAAS,IAAM;AAAA,MACpE,CAACA,KAAW,CAAChB,KAAS,CAACuC,IAAkB,QAAQ,QAAA,IAAW3C,EAAe2C,GAAgBD,GAAGF,GAAUpC,CAAK;AAAA,IAAA,CAC9G;AACD,IAAGoB,KAAQkB,KAAGG,EAAOH,GAAG,EAAE,WAAW,IAAM,OAAO,IAAM;AAGxD,UAAMI,IAAOF,EAAW,MAAM,KAAA,KAAU;AACxC,QAAG,CAACxB,EAAS,QAAOkB,GAAY,YAAY,EAAE,MAAAQ,GAAM;AACpD,QAAG,CAAC1C,EAAO,QAAOkC,GAAY,YAAY,EAAE,MAAAQ,GAAM,OAAO,8BAA8B;AACvF,QAAG,CAACH,EAAgB,QAAOL,GAAY,YAAY,EAAE,MAAAQ,GAAM,OAAO,mCAAmC;AAGrG,UAAMC,IAAWtC,EAAyBmC,EAAW,UAAU,CAAA,GAAIjC,KAAY,EAAE;AACjF,IAAA2B,GAAY,YAAY,EAAE,MAAMS,EAAA,CAAU;AAAA,EAC3C,SAASC,GAAU;AAClB,IAAAV,GAAY,YAAY,EAAE,OAAOU,EAAI,SAASA,EAAI,SAAS;AAAA,EAC5D;AACD,CAAC;"}

package/dist/audio.d.ts

@@ -1,5 +1,4 @@
 import { AbortablePromise, Ai } from './ai.ts';
-import { canDiarization } from './asr.ts';
 export declare class Audio {
     private ai;
     constructor(ai: Ai);
@@ -7,5 +6,5 @@ export declare class Audio {
         model?: string;
         speaker?: boolean | 'id';
     }): AbortablePromise<string | null>;
-    canDiarization: typeof canDiarization;
+    canDiarization: () => Promise<boolean>;
 }

package/dist/index.js

@@ -1,31 +1,31 @@
-"use strict";Object.defineProperty(exports,Symbol.toStringTag,{value:"Module"});const R=require("node:os"),f=require("@ztimson/utils"),$=require("@anthropic-ai/sdk"),v=require("openai"),x=require("worker_threads"),S=require("url"),w=require("path"),j=require("./asr.js"),L=require("tesseract.js");require("./embedder.js");const N=require("cheerio"),T=require("@ztimson/node-utils");var y=typeof document<"u"?document.currentScript:null;function q(h){const r=Object.create(null,{[Symbol.toStringTag]:{value:"Module"}});if(h){for(const e in h)if(e!=="default"){const t=Object.getOwnPropertyDescriptor(h,e);Object.defineProperty(r,e,t.get?t:{enumerable:!0,get:()=>h[e]})}}return r.default=h,Object.freeze(r)}const C=q(R),D=q(N);class k{}class P extends k{constructor(r,e,t){super(),this.ai=r,this.apiToken=e,this.model=t,this.client=new $.Anthropic({apiKey:e})}client;toStandard(r){const e=Date.now(),t=[];for(let c of r)if(typeof c.content=="string")t.push({timestamp:e,...c});else{const s=c.content?.filter(n=>n.type=="text").map(n=>n.text).join(`
+"use strict";Object.defineProperty(exports,Symbol.toStringTag,{value:"Module"});const R=require("node:os"),f=require("@ztimson/utils"),$=require("@anthropic-ai/sdk"),v=require("openai"),x=require("worker_threads"),S=require("url"),w=require("path"),j=require("./asr.js"),L=require("tesseract.js");require("./embedder.js");const N=require("cheerio"),T=require("@ztimson/node-utils");var g=typeof document<"u"?document.currentScript:null;function q(h){const r=Object.create(null,{[Symbol.toStringTag]:{value:"Module"}});if(h){for(const e in h)if(e!=="default"){const t=Object.getOwnPropertyDescriptor(h,e);Object.defineProperty(r,e,t.get?t:{enumerable:!0,get:()=>h[e]})}}return r.default=h,Object.freeze(r)}const C=q(R),D=q(N);class k{}class P extends k{constructor(r,e,t){super(),this.ai=r,this.apiToken=e,this.model=t,this.client=new $.Anthropic({apiKey:e})}client;toStandard(r){const e=Date.now(),t=[];for(let c of r)if(typeof c.content=="string")t.push({timestamp:e,...c});else{const n=c.content?.filter(s=>s.type=="text").map(s=>s.text).join(`
 
-`);s&&t.push({timestamp:e,role:c.role,content:s}),c.content.forEach(n=>{if(n.type=="tool_use")t.push({timestamp:e,role:"tool",id:n.id,name:n.name,args:n.input,content:void 0});else if(n.type=="tool_result"){const l=t.findLast(o=>o.id==n.tool_use_id);l&&(l[n.is_error?"error":"content"]=n.content)}})}return t}fromStandard(r){for(let e=0;e<r.length;e++)if(r[e].role=="tool"){const t=r[e];r.splice(e,1,{role:"assistant",content:[{type:"tool_use",id:t.id,name:t.name,input:t.args}]},{role:"user",content:[{type:"tool_result",tool_use_id:t.id,is_error:!!t.error,content:t.error||t.content}]}),e++}return r.map(({timestamp:e,...t})=>t)}ask(r,e={}){const t=new AbortController;return Object.assign(new Promise(async c=>{let s=this.fromStandard([...e.history||[],{role:"user",content:r,timestamp:Date.now()}]);const n=e.tools||this.ai.options.llm?.tools||[],l={model:e.model||this.model,max_tokens:e.max_tokens||this.ai.options.llm?.max_tokens||4096,system:e.system||this.ai.options.llm?.system||"",temperature:e.temperature||this.ai.options.llm?.temperature||.7,tools:n.map(m=>({name:m.name,description:m.description,input_schema:{type:"object",properties:m.args?f.objectMap(m.args,(i,d)=>({...d,required:void 0})):{},required:m.args?Object.entries(m.args).filter(i=>i[1].required).map(i=>i[0]):[]},fn:void 0})),messages:s,stream:!!e.stream};let o,a=!0;do{if(o=await this.client.messages.create(l).catch(i=>{throw i.message+=`
+`);n&&t.push({timestamp:e,role:c.role,content:n}),c.content.forEach(s=>{if(s.type=="tool_use")t.push({timestamp:e,role:"tool",id:s.id,name:s.name,args:s.input,content:void 0});else if(s.type=="tool_result"){const l=t.findLast(o=>o.id==s.tool_use_id);l&&(l[s.is_error?"error":"content"]=s.content)}})}return t}fromStandard(r){for(let e=0;e<r.length;e++)if(r[e].role=="tool"){const t=r[e];r.splice(e,1,{role:"assistant",content:[{type:"tool_use",id:t.id,name:t.name,input:t.args}]},{role:"user",content:[{type:"tool_result",tool_use_id:t.id,is_error:!!t.error,content:t.error||t.content}]}),e++}return r.map(({timestamp:e,...t})=>t)}ask(r,e={}){const t=new AbortController;return Object.assign(new Promise(async c=>{let n=this.fromStandard([...e.history||[],{role:"user",content:r,timestamp:Date.now()}]);const s=e.tools||this.ai.options.llm?.tools||[],l={model:e.model||this.model,max_tokens:e.max_tokens||this.ai.options.llm?.max_tokens||4096,system:e.system||this.ai.options.llm?.system||"",temperature:e.temperature||this.ai.options.llm?.temperature||.7,tools:s.map(m=>({name:m.name,description:m.description,input_schema:{type:"object",properties:m.args?f.objectMap(m.args,(i,d)=>({...d,required:void 0})):{},required:m.args?Object.entries(m.args).filter(i=>i[1].required).map(i=>i[0]):[]},fn:void 0})),messages:n,stream:!!e.stream};let o,a=!0;do{if(o=await this.client.messages.create(l).catch(i=>{throw i.message+=`
 
 Messages:
-${JSON.stringify(s,null,2)}`,i}),e.stream){a?a=!1:e.stream({text:`
+${JSON.stringify(n,null,2)}`,i}),e.stream){a?a=!1:e.stream({text:`
 
-`}),o.content=[];for await(const i of o){if(t.signal.aborted)break;if(i.type==="content_block_start")i.content_block.type==="text"?o.content.push({type:"text",text:""}):i.content_block.type==="tool_use"&&o.content.push({type:"tool_use",id:i.content_block.id,name:i.content_block.name,input:""});else if(i.type==="content_block_delta")if(i.delta.type==="text_delta"){const d=i.delta.text;o.content.at(-1).text+=d,e.stream({text:d})}else i.delta.type==="input_json_delta"&&(o.content.at(-1).input+=i.delta.partial_json);else if(i.type==="content_block_stop"){const d=o.content.at(-1);d.input!=null&&(d.input=d.input?f.JSONAttemptParse(d.input,{}):{})}else if(i.type==="message_stop")break}}const m=o.content.filter(i=>i.type==="tool_use");if(m.length&&!t.signal.aborted){s.push({role:"assistant",content:o.content});const i=await Promise.all(m.map(async d=>{const p=n.find(f.findByProp("name",d.name));if(e.stream&&e.stream({tool:d.name}),!p)return{tool_use_id:d.id,is_error:!0,content:"Tool not found"};try{const u=await p.fn(d.input,e?.stream,this.ai);return{type:"tool_result",tool_use_id:d.id,content:f.JSONSanitize(u)}}catch(u){return{type:"tool_result",tool_use_id:d.id,is_error:!0,content:u?.message||u?.toString()||"Unknown"}}}));s.push({role:"user",content:i}),l.messages=s}}while(!t.signal.aborted&&o.content.some(m=>m.type==="tool_use"));s.push({role:"assistant",content:o.content.filter(m=>m.type=="text").map(m=>m.text).join(`
+`}),o.content=[];for await(const i of o){if(t.signal.aborted)break;if(i.type==="content_block_start")i.content_block.type==="text"?o.content.push({type:"text",text:""}):i.content_block.type==="tool_use"&&o.content.push({type:"tool_use",id:i.content_block.id,name:i.content_block.name,input:""});else if(i.type==="content_block_delta")if(i.delta.type==="text_delta"){const d=i.delta.text;o.content.at(-1).text+=d,e.stream({text:d})}else i.delta.type==="input_json_delta"&&(o.content.at(-1).input+=i.delta.partial_json);else if(i.type==="content_block_stop"){const d=o.content.at(-1);d.input!=null&&(d.input=d.input?f.JSONAttemptParse(d.input,{}):{})}else if(i.type==="message_stop")break}}const m=o.content.filter(i=>i.type==="tool_use");if(m.length&&!t.signal.aborted){n.push({role:"assistant",content:o.content});const i=await Promise.all(m.map(async d=>{const p=s.find(f.findByProp("name",d.name));if(e.stream&&e.stream({tool:d.name}),!p)return{tool_use_id:d.id,is_error:!0,content:"Tool not found"};try{const u=await p.fn(d.input,e?.stream,this.ai);return{type:"tool_result",tool_use_id:d.id,content:f.JSONSanitize(u)}}catch(u){return{type:"tool_result",tool_use_id:d.id,is_error:!0,content:u?.message||u?.toString()||"Unknown"}}}));n.push({role:"user",content:i}),l.messages=n}}while(!t.signal.aborted&&o.content.some(m=>m.type==="tool_use"));n.push({role:"assistant",content:o.content.filter(m=>m.type=="text").map(m=>m.text).join(`
 
-`)}),s=this.toStandard(s),e.stream&&e.stream({done:!0}),e.history&&e.history.splice(0,e.history.length,...s),c(s.at(-1)?.content)}),{abort:()=>t.abort()})}}class _ extends k{constructor(r,e,t,c){super(),this.ai=r,this.host=e,this.token=t,this.model=c,this.client=new v.OpenAI(f.clean({baseURL:e,apiKey:t}))}client;toStandard(r){for(let e=0;e<r.length;e++){const t=r[e];if(t.role==="assistant"&&t.tool_calls){const c=t.tool_calls.map(s=>({role:"tool",id:s.id,name:s.function.name,args:f.JSONAttemptParse(s.function.arguments,{}),timestamp:t.timestamp}));r.splice(e,1,...c),e+=c.length-1}else if(t.role==="tool"&&t.content){const c=r.find(s=>t.tool_call_id==s.id);c&&(t.content.includes('"error":')?c.error=t.content:c.content=t.content),r.splice(e,1),e--}r[e]?.timestamp||(r[e].timestamp=Date.now())}return r}fromStandard(r){return r.reduce((e,t)=>{if(t.role==="tool")e.push({role:"assistant",content:null,tool_calls:[{id:t.id,type:"function",function:{name:t.name,arguments:JSON.stringify(t.args)}}],refusal:null,annotations:[]},{role:"tool",tool_call_id:t.id,content:t.error||t.content});else{const{timestamp:c,...s}=t;e.push(s)}return e},[])}ask(r,e={}){const t=new AbortController;return Object.assign(new Promise(async(c,s)=>{e.system&&e.history?.[0]?.role!="system"&&e.history?.splice(0,0,{role:"system",content:e.system,timestamp:Date.now()});let n=this.fromStandard([...e.history||[],{role:"user",content:r,timestamp:Date.now()}]);const l=e.tools||this.ai.options.llm?.tools||[],o={model:e.model||this.model,messages:n,stream:!!e.stream,max_tokens:e.max_tokens||this.ai.options.llm?.max_tokens||4096,temperature:e.temperature||this.ai.options.llm?.temperature||.7,tools:l.map(i=>({type:"function",function:{name:i.name,description:i.description,parameters:{type:"object",properties:i.args?f.objectMap(i.args,(d,p)=>({...p,required:void 0})):{},required:i.args?Object.entries(i.args).filter(d=>d[1].required).map(d=>d[0]):[]}}}))};let a,m=!0;do{if(a=await this.client.chat.completions.create(o).catch(d=>{throw d.message+=`
+`)}),n=this.toStandard(n),e.stream&&e.stream({done:!0}),e.history&&e.history.splice(0,e.history.length,...n),c(n.at(-1)?.content)}),{abort:()=>t.abort()})}}class _ extends k{constructor(r,e,t,c){super(),this.ai=r,this.host=e,this.token=t,this.model=c,this.client=new v.OpenAI(f.clean({baseURL:e,apiKey:t}))}client;toStandard(r){for(let e=0;e<r.length;e++){const t=r[e];if(t.role==="assistant"&&t.tool_calls){const c=t.tool_calls.map(n=>({role:"tool",id:n.id,name:n.function.name,args:f.JSONAttemptParse(n.function.arguments,{}),timestamp:t.timestamp}));r.splice(e,1,...c),e+=c.length-1}else if(t.role==="tool"&&t.content){const c=r.find(n=>t.tool_call_id==n.id);c&&(t.content.includes('"error":')?c.error=t.content:c.content=t.content),r.splice(e,1),e--}r[e]?.timestamp||(r[e].timestamp=Date.now())}return r}fromStandard(r){return r.reduce((e,t)=>{if(t.role==="tool")e.push({role:"assistant",content:null,tool_calls:[{id:t.id,type:"function",function:{name:t.name,arguments:JSON.stringify(t.args)}}],refusal:null,annotations:[]},{role:"tool",tool_call_id:t.id,content:t.error||t.content});else{const{timestamp:c,...n}=t;e.push(n)}return e},[])}ask(r,e={}){const t=new AbortController;return Object.assign(new Promise(async(c,n)=>{e.system&&e.history?.[0]?.role!="system"&&e.history?.splice(0,0,{role:"system",content:e.system,timestamp:Date.now()});let s=this.fromStandard([...e.history||[],{role:"user",content:r,timestamp:Date.now()}]);const l=e.tools||this.ai.options.llm?.tools||[],o={model:e.model||this.model,messages:s,stream:!!e.stream,max_tokens:e.max_tokens||this.ai.options.llm?.max_tokens||4096,temperature:e.temperature||this.ai.options.llm?.temperature||.7,tools:l.map(i=>({type:"function",function:{name:i.name,description:i.description,parameters:{type:"object",properties:i.args?f.objectMap(i.args,(d,p)=>({...p,required:void 0})):{},required:i.args?Object.entries(i.args).filter(d=>d[1].required).map(d=>d[0]):[]}}}))};let a,m=!0;do{if(a=await this.client.chat.completions.create(o).catch(d=>{throw d.message+=`
 
 Messages:
-${JSON.stringify(n,null,2)}`,d}),e.stream){m?m=!1:e.stream({text:`
+${JSON.stringify(s,null,2)}`,d}),e.stream){m?m=!1:e.stream({text:`
 
-`}),a.choices=[{message:{content:"",tool_calls:[]}}];for await(const d of a){if(t.signal.aborted)break;d.choices[0].delta.content&&(a.choices[0].message.content+=d.choices[0].delta.content,e.stream({text:d.choices[0].delta.content})),d.choices[0].delta.tool_calls&&(a.choices[0].message.tool_calls=d.choices[0].delta.tool_calls)}}const i=a.choices[0].message.tool_calls||[];if(i.length&&!t.signal.aborted){n.push(a.choices[0].message);const d=await Promise.all(i.map(async p=>{const u=l?.find(f.findByProp("name",p.function.name));if(e.stream&&e.stream({tool:p.function.name}),!u)return{role:"tool",tool_call_id:p.id,content:'{"error": "Tool not found"}'};try{const g=f.JSONAttemptParse(p.function.arguments,{}),b=await u.fn(g,e.stream,this.ai);return{role:"tool",tool_call_id:p.id,content:f.JSONSanitize(b)}}catch(g){return{role:"tool",tool_call_id:p.id,content:f.JSONSanitize({error:g?.message||g?.toString()||"Unknown"})}}}));n.push(...d),o.messages=n}}while(!t.signal.aborted&&a.choices?.[0]?.message?.tool_calls?.length);n.push({role:"assistant",content:a.choices[0].message.content||""}),n=this.toStandard(n),e.stream&&e.stream({done:!0}),e.history&&e.history.splice(0,e.history.length,...n),c(n.at(-1)?.content)}),{abort:()=>t.abort()})}}class J{constructor(r){this.ai=r,r.options.llm?.models&&Object.entries(r.options.llm.models).forEach(([e,t])=>{this.defaultModel||(this.defaultModel=e),t.proto=="anthropic"?this.models[e]=new P(this.ai,t.token,e):t.proto=="ollama"?this.models[e]=new _(this.ai,t.host,"not-needed",e):t.proto=="openai"&&(this.models[e]=new _(this.ai,t.host||null,t.token,e))})}defaultModel;models={};ask(r,e={}){const t=e.model||this.defaultModel;if(!this.models[t])throw new Error(`Model does not exist: ${t}`);let c=()=>{};return Object.assign(new Promise(async s=>{if(e.history||(e.history=[]),e.memory){e.system=(e.system||"")+`
+`}),a.choices=[{message:{content:"",tool_calls:[]}}];for await(const d of a){if(t.signal.aborted)break;d.choices[0].delta.content&&(a.choices[0].message.content+=d.choices[0].delta.content,e.stream({text:d.choices[0].delta.content})),d.choices[0].delta.tool_calls&&(a.choices[0].message.tool_calls=d.choices[0].delta.tool_calls)}}const i=a.choices[0].message.tool_calls||[];if(i.length&&!t.signal.aborted){s.push(a.choices[0].message);const d=await Promise.all(i.map(async p=>{const u=l?.find(f.findByProp("name",p.function.name));if(e.stream&&e.stream({tool:p.function.name}),!u)return{role:"tool",tool_call_id:p.id,content:'{"error": "Tool not found"}'};try{const y=f.JSONAttemptParse(p.function.arguments,{}),b=await u.fn(y,e.stream,this.ai);return{role:"tool",tool_call_id:p.id,content:f.JSONSanitize(b)}}catch(y){return{role:"tool",tool_call_id:p.id,content:f.JSONSanitize({error:y?.message||y?.toString()||"Unknown"})}}}));s.push(...d),o.messages=s}}while(!t.signal.aborted&&a.choices?.[0]?.message?.tool_calls?.length);s.push({role:"assistant",content:a.choices[0].message.content||""}),s=this.toStandard(s),e.stream&&e.stream({done:!0}),e.history&&e.history.splice(0,e.history.length,...s),c(s.at(-1)?.content)}),{abort:()=>t.abort()})}}class J{constructor(r){this.ai=r,r.options.llm?.models&&Object.entries(r.options.llm.models).forEach(([e,t])=>{this.defaultModel||(this.defaultModel=e),t.proto=="anthropic"?this.models[e]=new P(this.ai,t.token,e):t.proto=="ollama"?this.models[e]=new _(this.ai,t.host,"not-needed",e):t.proto=="openai"&&(this.models[e]=new _(this.ai,t.host||null,t.token,e))})}defaultModel;models={};ask(r,e={}){const t=e.model||this.defaultModel;if(!this.models[t])throw new Error(`Model does not exist: ${t}`);let c=()=>{};return Object.assign(new Promise(async n=>{if(e.history||(e.history=[]),e.memory){e.system=(e.system||"")+`
 You have passive persistent memory never make any mention of your memory capabilities and what you can/cannot remember
-`;const l=async(a,m,i=50)=>{const[d,p]=await Promise.all([m?this.embedding(m):Promise.resolve(null),a?this.embedding(a):Promise.resolve(null)]);return(e.memory||[]).map(u=>({...u,score:d?this.cosineSimilarity(u.embeddings[0],d[0].embedding):1})).filter(u=>u.score>=.8).map(u=>({...u,score:p?this.cosineSimilarity(u.embeddings[1],p[0].embedding):u.score})).filter(u=>u.score>=.2).toSorted((u,g)=>u.score-g.score).slice(0,i)},o=await l(r);o.length&&e.history.push({role:"assistant",content:`Things I remembered:
+`;const l=async(a,m,i=50)=>{const[d,p]=await Promise.all([m?this.embedding(m):Promise.resolve(null),a?this.embedding(a):Promise.resolve(null)]);return(e.memory||[]).map(u=>({...u,score:d?this.cosineSimilarity(u.embeddings[0],d[0].embedding):1})).filter(u=>u.score>=.8).map(u=>({...u,score:p?this.cosineSimilarity(u.embeddings[1],p[0].embedding):u.score})).filter(u=>u.score>=.2).toSorted((u,y)=>u.score-y.score).slice(0,i)},o=await l(r);o.length&&e.history.push({role:"assistant",content:`Things I remembered:
 `+o.map(a=>`${a.owner}: ${a.fact}`).join(`
-`)}),e.tools=[...e.tools||[],{name:"read_memory",description:"Check your long-term memory for more information",args:{subject:{type:"string",description:"Find information by a subject topic, can be used with or without query argument"},query:{type:"string",description:"Search memory based on a query, can be used with or without subject argument"},limit:{type:"number",description:"Result limit, default 5"}},fn:a=>{if(!a.subject&&!a.query)throw new Error("Either a subject or query argument is required");return l(a.query,a.subject,a.limit||5)}}]}const n=await this.models[t].ask(r,e);if(e.memory){const l=e.history?.findIndex(o=>o.role=="assistant"&&o.content.startsWith("Things I remembered:"));l!=null&&l>=0&&e.history?.splice(l,1)}if(e.compress||e.memory){let l=null;if(e.compress)l=await this.ai.language.compressHistory(e.history,e.compress.max,e.compress.min,e),e.history.splice(0,e.history.length,...l.history);else{const o=e.history?.findLastIndex(a=>a.role=="user")??-1;l=await this.ai.language.compressHistory(o!=-1?e.history.slice(o):e.history,0,0,e)}if(e.memory){const o=e.memory.filter(a=>!l.memory.some(m=>this.cosineSimilarity(a.embeddings[1],m.embeddings[1])>.8)).concat(l.memory);e.memory.splice(0,e.memory.length,...o)}}return s(n)}),{abort:c})}async compressHistory(r,e,t,c){if(this.estimateTokens(r)<e)return{history:r,memory:[]};let s=0,n=0;for(let u of r.toReversed())if(n+=this.estimateTokens(u.content),n<t)s++;else break;if(r.length<=s)return{history:r,memory:[]};const l=r[0].role=="system"?r[0]:null,o=s==0?[]:r.slice(-s),a=(s==0?r:r.slice(0,-s)).filter(u=>u.role==="assistant"||u.role==="user"),m=await this.json(a.map(u=>`${u.role}: ${u.content}`).join(`
+`)}),e.tools=[...e.tools||[],{name:"read_memory",description:"Check your long-term memory for more information",args:{subject:{type:"string",description:"Find information by a subject topic, can be used with or without query argument"},query:{type:"string",description:"Search memory based on a query, can be used with or without subject argument"},limit:{type:"number",description:"Result limit, default 5"}},fn:a=>{if(!a.subject&&!a.query)throw new Error("Either a subject or query argument is required");return l(a.query,a.subject,a.limit||5)}}]}const s=await this.models[t].ask(r,e);if(e.memory){const l=e.history?.findIndex(o=>o.role=="assistant"&&o.content.startsWith("Things I remembered:"));l!=null&&l>=0&&e.history?.splice(l,1)}if(e.compress||e.memory){let l=null;if(e.compress)l=await this.ai.language.compressHistory(e.history,e.compress.max,e.compress.min,e),e.history.splice(0,e.history.length,...l.history);else{const o=e.history?.findLastIndex(a=>a.role=="user")??-1;l=await this.ai.language.compressHistory(o!=-1?e.history.slice(o):e.history,0,0,e)}if(e.memory){const o=e.memory.filter(a=>!l.memory.some(m=>this.cosineSimilarity(a.embeddings[1],m.embeddings[1])>.8)).concat(l.memory);e.memory.splice(0,e.memory.length,...o)}}return n(s)}),{abort:c})}async compressHistory(r,e,t,c){if(this.estimateTokens(r)<e)return{history:r,memory:[]};let n=0,s=0;for(let u of r.toReversed())if(s+=this.estimateTokens(u.content),s<t)n++;else break;if(r.length<=n)return{history:r,memory:[]};const l=r[0].role=="system"?r[0]:null,o=n==0?[]:r.slice(-n),a=(n==0?r:r.slice(0,-n)).filter(u=>u.role==="assistant"||u.role==="user"),m=await this.json(a.map(u=>`${u.role}: ${u.content}`).join(`
 
-`),"{summary: string, facts: [[subject, fact]]}",{system:"Create the smallest summary possible, no more than 500 tokens. Create a list of NEW facts (split by subject [pro]noun and fact) about what you learned from this conversation that you didn't already know or get from a tool call or system prompt. Focus only on new information about people, topics, or facts. Avoid generating facts about the AI.",model:c?.model,temperature:c?.temperature||.3}),i=new Date,d=await Promise.all((m?.facts||[])?.map(async([u,g])=>{const b=await Promise.all([this.embedding(u),this.embedding(`${u}: ${g}`)]);return{owner:u,fact:g,embeddings:[b[0][0].embedding,b[1][0].embedding],timestamp:i}})),p=[{role:"assistant",content:`Conversation Summary: ${m?.summary}`,timestamp:Date.now()},...o];return l&&p.splice(0,0,l),{history:p,memory:d}}cosineSimilarity(r,e){if(r.length!==e.length)throw new Error("Vectors must be same length");let t=0,c=0,s=0;for(let l=0;l<r.length;l++)t+=r[l]*e[l],c+=r[l]*r[l],s+=e[l]*e[l];const n=Math.sqrt(c)*Math.sqrt(s);return n===0?0:t/n}chunk(r,e=500,t=50){const c=(o,a="")=>o?Object.entries(o).flatMap(([m,i])=>{const d=a?`${a}${isNaN(+m)?`.${m}`:`[${m}]`}`:m;return typeof i=="object"&&!Array.isArray(i)?c(i,d):`${d}: ${Array.isArray(i)?i.join(", "):i}`}):[],n=(typeof r=="object"?c(r):r.split(`
+`),"{summary: string, facts: [[subject, fact]]}",{system:"Create the smallest summary possible, no more than 500 tokens. Create a list of NEW facts (split by subject [pro]noun and fact) about what you learned from this conversation that you didn't already know or get from a tool call or system prompt. Focus only on new information about people, topics, or facts. Avoid generating facts about the AI.",model:c?.model,temperature:c?.temperature||.3}),i=new Date,d=await Promise.all((m?.facts||[])?.map(async([u,y])=>{const b=await Promise.all([this.embedding(u),this.embedding(`${u}: ${y}`)]);return{owner:u,fact:y,embeddings:[b[0][0].embedding,b[1][0].embedding],timestamp:i}})),p=[{role:"assistant",content:`Conversation Summary: ${m?.summary}`,timestamp:Date.now()},...o];return l&&p.splice(0,0,l),{history:p,memory:d}}cosineSimilarity(r,e){if(r.length!==e.length)throw new Error("Vectors must be same length");let t=0,c=0,n=0;for(let l=0;l<r.length;l++)t+=r[l]*e[l],c+=r[l]*r[l],n+=e[l]*e[l];const s=Math.sqrt(c)*Math.sqrt(n);return s===0?0:t/s}chunk(r,e=500,t=50){const c=(o,a="")=>o?Object.entries(o).flatMap(([m,i])=>{const d=a?`${a}${isNaN(+m)?`.${m}`:`[${m}]`}`:m;return typeof i=="object"&&!Array.isArray(i)?c(i,d):`${d}: ${Array.isArray(i)?i.join(", "):i}`}):[],s=(typeof r=="object"?c(r):r.split(`
 `)).flatMap(o=>[...o.split(/\s+/).filter(Boolean),`
-`]),l=[];for(let o=0;o<n.length;){let a="",m=o;for(;m<n.length;){const d=a+(a?" ":"")+n[m];if(this.estimateTokens(d.replace(/\s*\n\s*/g,`
+`]),l=[];for(let o=0;o<s.length;){let a="",m=o;for(;m<s.length;){const d=a+(a?" ":"")+s[m];if(this.estimateTokens(d.replace(/\s*\n\s*/g,`
 `))>e&&a)break;a=d,m++}const i=a.replace(/\s*\n\s*/g,`
-`).trim();i&&l.push(i),o=Math.max(m-t,m===o?o+1:m)}return l}async embedding(r,e={}){let{maxTokens:t=500,overlapTokens:c=50}=e;const s=o=>new Promise((a,m)=>{const i=new x.Worker(w.join(w.dirname(S.fileURLToPath(typeof document>"u"?require("url").pathToFileURL(__filename).href:y&&y.tagName.toUpperCase()==="SCRIPT"&&y.src||new URL("index.js",document.baseURI).href)),"embedder.js")),d=({embedding:u})=>{i.terminate(),a(u)},p=u=>{i.terminate(),m(u)};i.on("message",d),i.on("error",p),i.on("exit",u=>{u!==0&&m(new Error(`Worker exited with code ${u}`))}),i.postMessage({text:o,model:this.ai.options?.embedder||"bge-small-en-v1.5",modelDir:this.ai.options.path})}),n=this.chunk(r,t,c),l=[];for(let o=0;o<n.length;o++){const a=n[o],m=await s(a);l.push({index:o,embedding:m,text:a,tokens:this.estimateTokens(a)})}return l}estimateTokens(r){const e=JSON.stringify(r);return Math.ceil(e.length/4*1.2)}fuzzyMatch(r,...e){if(e.length<2)throw new Error("Requires at least 2 strings to compare");const t=(n,l=10)=>n.toLowerCase().split("").map((o,a)=>o.charCodeAt(0)*(a+1)%l/l).slice(0,l),c=t(r),s=e.map(n=>t(n)).map(n=>this.cosineSimilarity(c,n));return{avg:s.reduce((n,l)=>n+l,0)/s.length,max:Math.max(...s),similarities:s}}async json(r,e,t){let c=await this.ask(r,{...t,system:(t?.system?`${t.system}
+`).trim();i&&l.push(i),o=Math.max(m-t,m===o?o+1:m)}return l}async embedding(r,e={}){let{maxTokens:t=500,overlapTokens:c=50}=e;const n=o=>new Promise((a,m)=>{const i=new x.Worker(w.join(w.dirname(S.fileURLToPath(typeof document>"u"?require("url").pathToFileURL(__filename).href:g&&g.tagName.toUpperCase()==="SCRIPT"&&g.src||new URL("index.js",document.baseURI).href)),"embedder.js")),d=({embedding:u})=>{i.terminate(),a(u)},p=u=>{i.terminate(),m(u)};i.on("message",d),i.on("error",p),i.on("exit",u=>{u!==0&&m(new Error(`Worker exited with code ${u}`))}),i.postMessage({text:o,model:this.ai.options?.embedder||"bge-small-en-v1.5",modelDir:this.ai.options.path})}),s=this.chunk(r,t,c),l=[];for(let o=0;o<s.length;o++){const a=s[o],m=await n(a);l.push({index:o,embedding:m,text:a,tokens:this.estimateTokens(a)})}return l}estimateTokens(r){const e=JSON.stringify(r);return Math.ceil(e.length/4*1.2)}fuzzyMatch(r,...e){if(e.length<2)throw new Error("Requires at least 2 strings to compare");const t=(s,l=10)=>s.toLowerCase().split("").map((o,a)=>o.charCodeAt(0)*(a+1)%l/l).slice(0,l),c=t(r),n=e.map(s=>t(s)).map(s=>this.cosineSimilarity(c,s));return{avg:n.reduce((s,l)=>s+l,0)/n.length,max:Math.max(...n),similarities:n}}async json(r,e,t){let c=await this.ask(r,{...t,system:(t?.system?`${t.system}
 `:"")+`Only respond using a JSON code block matching this schema:
 \`\`\`json
 ${e}
-\`\`\``});if(!c)return{};const s=/```(?:.+)?\s*([\s\S]*?)```/.exec(c),n=s?s[1].trim():c;return f.JSONAttemptParse(n,{})}summarize(r,e,t){return this.ask(r,{system:`Generate a brief summary <= ${e} tokens. Output nothing else`,temperature:.3,...t})}}class O{constructor(r){this.ai=r}asr(r,e={}){const{model:t=this.ai.options.asr||"whisper-base",speaker:c=!1}=e;let s=!1;const n=()=>{s=!0};let l=new Promise((o,a)=>{const m=new x.Worker(w.join(w.dirname(S.fileURLToPath(typeof document>"u"?require("url").pathToFileURL(__filename).href:y&&y.tagName.toUpperCase()==="SCRIPT"&&y.src||new URL("index.js",document.baseURI).href)),"asr.js")),i=({text:p,warning:u,error:g})=>{m.terminate(),!s&&(g?a(new Error(g)):(u&&console.warn(u),o(p)))},d=p=>{m.terminate(),s||a(p)};m.on("message",i),m.on("error",d),m.on("exit",p=>{p!==0&&!s&&a(new Error(`Worker exited with code ${p}`))}),m.postMessage({file:r,model:t,speaker:c,modelDir:this.ai.options.path,token:this.ai.options.hfToken})});if(e.speaker=="id"){if(!this.ai.language.defaultModel)throw new Error("Configure an LLM for advanced ASR speaker detection");l=l.then(async o=>{if(!o)return o;let a=this.ai.language.chunk(o,500,0);a.length>4&&(a=[...a.slice(0,3),a.at(-1)]);const m=await this.ai.language.json(a.join(`
-`),'{1: "Detected Name"}',{system:"Use this following transcript to identify speakers. Only identify speakers you are sure about",temperature:.1});return Object.entries(m).forEach(([i,d])=>{o=o.replaceAll(`[Speaker ${i}]`,`[${d}]`)}),o})}return Object.assign(l,{abort:n})}canDiarization=j.canDiarization}class M{constructor(r){this.ai=r}ocr(r){let e;const t=new Promise(async c=>{e=await L.createWorker(this.ai.options.ocr||"eng",2,{cachePath:this.ai.options.path});const{data:s}=await e.recognize(r);await e.terminate(),c(s.text.trim()||null)});return Object.assign(t,{abort:()=>e?.terminate()})}}class W{constructor(r){this.options=r,r.path||(r.path=C.tmpdir()),process.env.TRANSFORMERS_CACHE=r.path,this.audio=new O(this),this.language=new J(this),this.vision=new M(this)}audio;language;vision}const A={name:"cli",description:"Use the command line interface, returns any output",args:{command:{type:"string",description:"Command to run",required:!0}},fn:h=>T.$`${h.command}`},z={name:"get_datetime",description:"Get current UTC date / time",args:{},fn:async()=>new Date().toUTCString()},I={name:"exec",description:"Run code/scripts",args:{language:{type:"string",description:"Execution language",enum:["cli","node","python"],required:!0},code:{type:"string",description:"Code to execute",required:!0}},fn:async(h,r,e)=>{try{switch(h.type){case"bash":return await A.fn({command:h.code},r,e);case"node":return await U.fn({code:h.code},r,e);case"python":return await E.fn({code:h.code},r,e)}}catch(t){return{error:t?.message||t.toString()}}}},F={name:"fetch",description:"Make HTTP request to URL",args:{url:{type:"string",description:"URL to fetch",required:!0},method:{type:"string",description:"HTTP method to use",enum:["GET","POST","PUT","DELETE"],default:"GET"},headers:{type:"object",description:"HTTP headers to send",default:{}},body:{type:"object",description:"HTTP body to send"}},fn:h=>new f.Http({url:h.url,headers:h.headers}).request({method:h.method||"GET",body:h.body})},U={name:"exec_javascript",description:"Execute commonjs javascript",args:{code:{type:"string",description:"CommonJS javascript",required:!0}},fn:async h=>{const r=f.consoleInterceptor(null),e=await f.fn({console:r},h.code,!0).catch(t=>r.output.error.push(t));return{...r.output,return:e,stdout:void 0,stderr:void 0}}},E={name:"exec_javascript",description:"Execute commonjs javascript",args:{code:{type:"string",description:"CommonJS javascript",required:!0}},fn:async h=>({result:T.$Sync`python -c "${h.code}"`})},H={name:"read_webpage",description:"Extract clean, structured content from a webpage. Use after web_search to read specific URLs",args:{url:{type:"string",description:"URL to extract content from",required:!0},focus:{type:"string",description:'Optional: What aspect to focus on (e.g., "pricing", "features", "contact info")'}},fn:async h=>{const r=await fetch(h.url,{headers:{"User-Agent":"Mozilla/5.0 (Windows NT 10.0; Win64; x64)"}}).then(n=>n.text()).catch(n=>{throw new Error(`Failed to fetch: ${n.message}`)}),e=D.load(r);e('script, style, nav, footer, header, aside, iframe, noscript, [role="navigation"], [role="banner"], .ad, .ads, .cookie, .popup').remove();const t={title:e('meta[property="og:title"]').attr("content")||e("title").text()||"",description:e('meta[name="description"]').attr("content")||e('meta[property="og:description"]').attr("content")||""};let c="";const s=["article","main",'[role="main"]',".content",".post",".entry","body"];for(const n of s){const l=e(n).first();if(l.length&&l.text().trim().length>200){c=l.text();break}}return c||(c=e("body").text()),c=c.replace(/\s+/g," ").trim().slice(0,8e3),{url:h.url,title:t.title.trim(),description:t.description.trim(),content:c,focus:h.focus}}},B={name:"web_search",description:"Use duckduckgo (anonymous) to find find relevant online resources. Returns a list of URLs that works great with the `read_webpage` tool",args:{query:{type:"string",description:"Search string",required:!0},length:{type:"string",description:"Number of results to return",default:5}},fn:async h=>{const r=await fetch(`https://html.duckduckgo.com/html/?q=${encodeURIComponent(h.query)}`,{headers:{"User-Agent":"Mozilla/5.0 (Windows NT 10.0; Win64; x64)","Accept-Language":"en-US,en;q=0.9"}}).then(s=>s.text());let e,t=/<a .*?href="(.+?)".+?<\/a>/g;const c=new f.ASet;for(;(e=t.exec(r))!==null;){let s=/uddg=(.+)&amp?/.exec(decodeURIComponent(e[1]))?.[1];if(s&&(s=decodeURIComponent(s)),s&&c.add(s),c.size>=(h.length||5))break}return c}};exports.canDiarization=j.canDiarization;exports.Ai=W;exports.Anthropic=P;exports.Audio=O;exports.CliTool=A;exports.DateTimeTool=z;exports.ExecTool=I;exports.FetchTool=F;exports.JSTool=U;exports.LLMProvider=k;exports.OpenAi=_;exports.PythonTool=E;exports.ReadWebpageTool=H;exports.Vision=M;exports.WebSearchTool=B;
+\`\`\``});if(!c)return{};const n=/```(?:.+)?\s*([\s\S]*?)```/.exec(c),s=n?n[1].trim():c;return f.JSONAttemptParse(s,{})}summarize(r,e,t){return this.ask(r,{system:`Generate a brief summary <= ${e} tokens. Output nothing else`,temperature:.3,...t})}}class O{constructor(r){this.ai=r}asr(r,e={}){const{model:t=this.ai.options.asr||"whisper-base",speaker:c=!1}=e;let n=!1;const s=()=>{n=!0};let l=new Promise((o,a)=>{const m=new x.Worker(w.join(w.dirname(S.fileURLToPath(typeof document>"u"?require("url").pathToFileURL(__filename).href:g&&g.tagName.toUpperCase()==="SCRIPT"&&g.src||new URL("index.js",document.baseURI).href)),"asr.js")),i=({text:p,warning:u,error:y})=>{m.terminate(),!n&&(y?a(new Error(y)):(u&&console.warn(u),o(p)))},d=p=>{m.terminate(),n||a(p)};m.on("message",i),m.on("error",d),m.on("exit",p=>{p!==0&&!n&&a(new Error(`Worker exited with code ${p}`))}),m.postMessage({file:r,model:t,speaker:c,modelDir:this.ai.options.path,token:this.ai.options.hfToken})});if(e.speaker=="id"){if(!this.ai.language.defaultModel)throw new Error("Configure an LLM for advanced ASR speaker detection");l=l.then(async o=>{if(!o)return o;let a=this.ai.language.chunk(o,500,0);a.length>4&&(a=[...a.slice(0,3),a.at(-1)]);const m=await this.ai.language.json(a.join(`
+`),'{1: "Detected Name", 2: "Second Name"}',{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:.1});return Object.entries(m).forEach(([i,d])=>{o=o.replaceAll(`[Speaker ${i}]`,`[${d}]`)}),o})}return Object.assign(l,{abort:s})}canDiarization=()=>j.canDiarization().then(r=>!!r)}class M{constructor(r){this.ai=r}ocr(r){let e;const t=new Promise(async c=>{e=await L.createWorker(this.ai.options.ocr||"eng",2,{cachePath:this.ai.options.path});const{data:n}=await e.recognize(r);await e.terminate(),c(n.text.trim()||null)});return Object.assign(t,{abort:()=>e?.terminate()})}}class W{constructor(r){this.options=r,r.path||(r.path=C.tmpdir()),process.env.TRANSFORMERS_CACHE=r.path,this.audio=new O(this),this.language=new J(this),this.vision=new M(this)}audio;language;vision}const A={name:"cli",description:"Use the command line interface, returns any output",args:{command:{type:"string",description:"Command to run",required:!0}},fn:h=>T.$`${h.command}`},z={name:"get_datetime",description:"Get current UTC date / time",args:{},fn:async()=>new Date().toUTCString()},I={name:"exec",description:"Run code/scripts",args:{language:{type:"string",description:"Execution language",enum:["cli","node","python"],required:!0},code:{type:"string",description:"Code to execute",required:!0}},fn:async(h,r,e)=>{try{switch(h.type){case"bash":return await A.fn({command:h.code},r,e);case"node":return await U.fn({code:h.code},r,e);case"python":return await E.fn({code:h.code},r,e)}}catch(t){return{error:t?.message||t.toString()}}}},F={name:"fetch",description:"Make HTTP request to URL",args:{url:{type:"string",description:"URL to fetch",required:!0},method:{type:"string",description:"HTTP method to use",enum:["GET","POST","PUT","DELETE"],default:"GET"},headers:{type:"object",description:"HTTP headers to send",default:{}},body:{type:"object",description:"HTTP body to send"}},fn:h=>new f.Http({url:h.url,headers:h.headers}).request({method:h.method||"GET",body:h.body})},U={name:"exec_javascript",description:"Execute commonjs javascript",args:{code:{type:"string",description:"CommonJS javascript",required:!0}},fn:async h=>{const r=f.consoleInterceptor(null),e=await f.fn({console:r},h.code,!0).catch(t=>r.output.error.push(t));return{...r.output,return:e,stdout:void 0,stderr:void 0}}},E={name:"exec_javascript",description:"Execute commonjs javascript",args:{code:{type:"string",description:"CommonJS javascript",required:!0}},fn:async h=>({result:T.$Sync`python -c "${h.code}"`})},H={name:"read_webpage",description:"Extract clean, structured content from a webpage. Use after web_search to read specific URLs",args:{url:{type:"string",description:"URL to extract content from",required:!0},focus:{type:"string",description:'Optional: What aspect to focus on (e.g., "pricing", "features", "contact info")'}},fn:async h=>{const r=await fetch(h.url,{headers:{"User-Agent":"Mozilla/5.0 (Windows NT 10.0; Win64; x64)"}}).then(s=>s.text()).catch(s=>{throw new Error(`Failed to fetch: ${s.message}`)}),e=D.load(r);e('script, style, nav, footer, header, aside, iframe, noscript, [role="navigation"], [role="banner"], .ad, .ads, .cookie, .popup').remove();const t={title:e('meta[property="og:title"]').attr("content")||e("title").text()||"",description:e('meta[name="description"]').attr("content")||e('meta[property="og:description"]').attr("content")||""};let c="";const n=["article","main",'[role="main"]',".content",".post",".entry","body"];for(const s of n){const l=e(s).first();if(l.length&&l.text().trim().length>200){c=l.text();break}}return c||(c=e("body").text()),c=c.replace(/\s+/g," ").trim().slice(0,8e3),{url:h.url,title:t.title.trim(),description:t.description.trim(),content:c,focus:h.focus}}},B={name:"web_search",description:"Use duckduckgo (anonymous) to find find relevant online resources. Returns a list of URLs that works great with the `read_webpage` tool",args:{query:{type:"string",description:"Search string",required:!0},length:{type:"string",description:"Number of results to return",default:5}},fn:async h=>{const r=await fetch(`https://html.duckduckgo.com/html/?q=${encodeURIComponent(h.query)}`,{headers:{"User-Agent":"Mozilla/5.0 (Windows NT 10.0; Win64; x64)","Accept-Language":"en-US,en;q=0.9"}}).then(n=>n.text());let e,t=/<a .*?href="(.+?)".+?<\/a>/g;const c=new f.ASet;for(;(e=t.exec(r))!==null;){let n=/uddg=(.+)&amp?/.exec(decodeURIComponent(e[1]))?.[1];if(n&&(n=decodeURIComponent(n)),n&&c.add(n),c.size>=(h.length||5))break}return c}};exports.canDiarization=j.canDiarization;exports.Ai=W;exports.Anthropic=P;exports.Audio=O;exports.CliTool=A;exports.DateTimeTool=z;exports.ExecTool=I;exports.FetchTool=F;exports.JSTool=U;exports.LLMProvider=k;exports.OpenAi=_;exports.PythonTool=E;exports.ReadWebpageTool=H;exports.Vision=M;exports.WebSearchTool=B;
 //# sourceMappingURL=index.js.map