[{"data":1,"prerenderedAt":-1},["ShallowReactive",2],{"skill:01M6F9G52Q3M9J4QVPZDA3331C":3,"skill-same:01M6F9G52Q3M9J4QVPZDA3331C":73},{"skillNo":4,"name":5,"nameZh":6,"description":7,"descriptionZh":6,"repositoryFullName":8,"ownerLogin":9,"ownerAvatarUrl":10,"ownerAccountNo":11,"ownerDisplayName":12,"repositoryNo":13,"starsCount":14,"licenseSpdxId":15,"licenseFileUrl":6,"sourceHtmlUrl":16,"categories":17,"keywords":18,"publishedAt":41,"linkOnly":42,"markdownContent":43,"markdownBody":44,"frontMatter":45,"refName":50,"commitSha":51,"skillDirectory":52,"forksCount":53,"repositoryHtmlUrl":54,"repositoryHomepageUrl":6,"ownerHtmlUrl":55,"ownerBlogUrl":56,"ownerDescription":57,"updatedAt":58,"install":59,"resources":72},"01M6F9G52Q3M9J4QVPZDA3331C","twinmind-performance-tuning",null,"Optimize TwinMind transcription accuracy and processing speed.\nUse when improving transcription quality, reducing latency,\nor tuning model parameters for specific use cases.\nTrigger with phrases like \"twinmind performance\", \"improve transcription accuracy\",\n\"faster twinmind\", \"optimize twinmind\", \"transcription quality\".","Dicklesworthstone\u002Fpi_agent_rust","Dicklesworthstone","https:\u002F\u002Favatars.githubusercontent.com\u002Fu\u002F35050222?v=4","01M6CG26AR942CSVMN1M5E3DSE","Jeff Emanuel","01M6CG26EPP71PSHW1VDS71121",1992,"MIT","https:\u002F\u002Fgithub.com\u002FDicklesworthstone\u002Fpi_agent_rust\u002Fblob\u002Fmain\u002Ftests\u002Fext_conformance\u002Fartifacts\u002Fplugins-community\u002Fplugins\u002Fsaas-packs\u002Ftwinmind-pack\u002Fskills\u002Ftwinmind-performance-tuning\u002FSKILL.md",[],[19,20,21,22,23,24,25,26,27,28,29,30,31,32,33,34,35,36,37,38,39,40],"tuning","performance","twinmind performance tuning","performance tuning","when","accuracy","for","with","processing","trigger","io","latency","speed","optimize","transcription","use","edit","write","read","quality","model","improve","2026-11-10 10:12:30",false,"---\nname: twinmind-performance-tuning\ndescription: |\n  Optimize TwinMind transcription accuracy and processing speed.\n  Use when improving transcription quality, reducing latency,\n  or tuning model parameters for specific use cases.\n  Trigger with phrases like \"twinmind performance\", \"improve transcription accuracy\",\n  \"faster twinmind\", \"optimize twinmind\", \"transcription quality\".\nallowed-tools: Read, Write, Edit\nversion: 1.0.0\nlicense: MIT\nauthor: Jeremy Longshore \u003Cjeremy@intentsolutions.io>\n---\n\n# TwinMind Performance Tuning\n\n## Overview\nOptimize TwinMind for better transcription accuracy, faster processing, and improved user experience.\n\n## Prerequisites\n- TwinMind Pro\u002FEnterprise account\n- Understanding of audio processing concepts\n- Access to quality metrics and logs\n\n## Instructions\n\n### Step 1: Understand Performance Metrics\n\n```typescript\n\u002F\u002F src\u002Ftwinmind\u002Fmetrics\u002Fperformance.ts\nexport interface TranscriptionMetrics {\n  \u002F\u002F Accuracy metrics\n  wordErrorRate: number;         \u002F\u002F WER - lower is better (Ear-3: ~5.26%)\n  diarizationErrorRate: number;  \u002F\u002F DER - speaker labeling accuracy (~3.8%)\n  confidenceScore: number;       \u002F\u002F Average confidence (0-1)\n\n  \u002F\u002F Timing metrics\n  processingTime: number;        \u002F\u002F Time to complete (ms)\n  realtimeFactor: number;        \u002F\u002F Processing time \u002F audio duration\n  firstWordLatency: number;      \u002F\u002F Time to first result (streaming)\n\n  \u002F\u002F Quality metrics\n  speakerCount: number;          \u002F\u002F Detected speakers\n  languageDetected: string;      \u002F\u002F Detected language\n  noiseLevel: string;            \u002F\u002F low\u002Fmedium\u002Fhigh\n}\n\nexport async function getTranscriptionMetrics(transcriptId: string): Promise\u003CTranscriptionMetrics> {\n  const client = getTwinMindClient();\n  const response = await client.get(`\u002Ftranscripts\u002F${transcriptId}\u002Fmetrics`);\n  return response.data;\n}\n\n\u002F\u002F Log and analyze metrics\nexport function analyzePerformance(metrics: TranscriptionMetrics): string[] {\n  const recommendations: string[] = [];\n\n  if (metrics.wordErrorRate > 0.10) {\n    recommendations.push('High WER - consider audio quality improvements');\n  }\n\n  if (metrics.diarizationErrorRate > 0.05) {\n    recommendations.push('Speaker labeling issues - ensure clear audio separation');\n  }\n\n  if (metrics.realtimeFactor > 0.5) {\n    recommendations.push('Slow processing - consider model optimization');\n  }\n\n  if (metrics.noiseLevel === 'high') {\n    recommendations.push('High background noise - recommend noise reduction');\n  }\n\n  return recommendations;\n}\n```\n\n### Step 2: Audio Quality Optimization\n\n```typescript\n\u002F\u002F src\u002Ftwinmind\u002Fpreprocessing\u002Faudio.ts\nimport ffmpeg from 'fluent-ffmpeg';\n\ninterface AudioPreprocessOptions {\n  targetSampleRate?: number;   \u002F\u002F Default: 16000 Hz\n  channels?: number;           \u002F\u002F Default: 1 (mono)\n  noiseReduction?: boolean;    \u002F\u002F Enable noise reduction\n  normalization?: boolean;     \u002F\u002F Normalize volume\n  format?: string;             \u002F\u002F Output format (mp3, wav)\n}\n\nconst defaultOptions: AudioPreprocessOptions = {\n  targetSampleRate: 16000,\n  channels: 1,\n  noiseReduction: true,\n  normalization: true,\n  format: 'mp3',\n};\n\nexport async function preprocessAudio(\n  inputPath: string,\n  outputPath: string,\n  options: AudioPreprocessOptions = {}\n): Promise\u003Cvoid> {\n  const opts = { ...defaultOptions, ...options };\n\n  return new Promise((resolve, reject) => {\n    let command = ffmpeg(inputPath)\n      .audioFrequency(opts.targetSampleRate!)\n      .audioChannels(opts.channels!);\n\n    \u002F\u002F Noise reduction using highpass and lowpass filters\n    if (opts.noiseReduction) {\n      command = command.audioFilters([\n        'highpass=f=200',     \u002F\u002F Remove low frequency noise\n        'lowpass=f=3000',     \u002F\u002F Remove high frequency noise\n        'afftdn=nf=-25',      \u002F\u002F FFT denoiser\n      ]);\n    }\n\n    \u002F\u002F Volume normalization\n    if (opts.normalization) {\n      command = command.audioFilters([\n        'loudnorm=I=-16:TP=-1.5:LRA=11', \u002F\u002F EBU R128 loudness normalization\n      ]);\n    }\n\n    command\n      .toFormat(opts.format!)\n      .on('end', () => resolve())\n      .on('error', reject)\n      .save(outputPath);\n  });\n}\n\n\u002F\u002F Audio quality assessment\nexport async function assessAudioQuality(filePath: string): Promise\u003C{\n  quality: 'excellent' | 'good' | 'fair' | 'poor';\n  issues: string[];\n  recommendations: string[];\n}> {\n  const issues: string[] = [];\n  const recommendations: string[] = [];\n\n  \u002F\u002F Get audio metadata\n  const metadata = await getAudioMetadata(filePath);\n\n  \u002F\u002F Check sample rate\n  if (metadata.sampleRate \u003C 16000) {\n    issues.push(`Low sample rate: ${metadata.sampleRate} Hz`);\n    recommendations.push('Use 16 kHz or higher sample rate');\n  }\n\n  \u002F\u002F Check bit depth\n  if (metadata.bitDepth && metadata.bitDepth \u003C 16) {\n    issues.push(`Low bit depth: ${metadata.bitDepth} bits`);\n    recommendations.push('Use 16-bit or higher audio');\n  }\n\n  \u002F\u002F Check for clipping\n  if (metadata.peakLevel && metadata.peakLevel > -1) {\n    issues.push('Audio clipping detected');\n    recommendations.push('Reduce recording volume to prevent distortion');\n  }\n\n  \u002F\u002F Check noise floor\n  if (metadata.noiseFloor && metadata.noiseFloor > -40) {\n    issues.push(`High noise floor: ${metadata.noiseFloor} dB`);\n    recommendations.push('Use noise reduction or improve recording environment');\n  }\n\n  const quality = issues.length === 0 ? 'excellent' :\n                  issues.length \u003C= 1 ? 'good' :\n                  issues.length \u003C= 2 ? 'fair' : 'poor';\n\n  return { quality, issues, recommendations };\n}\n```\n\n### Step 3: Model Selection and Configuration\n\n```typescript\n\u002F\u002F src\u002Ftwinmind\u002Fmodels\u002Fconfig.ts\nexport interface ModelConfig {\n  model: 'ear-3' | 'ear-2' | 'ear-3-custom';\n  language?: string;           \u002F\u002F 'auto' or ISO code (en, es, fr, etc.)\n  diarization: boolean;        \u002F\u002F Speaker labeling\n  punctuation: boolean;        \u002F\u002F Auto-punctuation\n  profanityFilter: boolean;    \u002F\u002F Filter explicit content\n  vocabulary?: string[];       \u002F\u002F Custom vocabulary boost\n  speakerHints?: string[];     \u002F\u002F Expected speaker names\n}\n\n\u002F\u002F Optimized configs for different scenarios\nexport const modelConfigs: Record\u003Cstring, ModelConfig> = {\n  \u002F\u002F Standard meeting transcription\n  meeting: {\n    model: 'ear-3',\n    language: 'auto',\n    diarization: true,\n    punctuation: true,\n    profanityFilter: false,\n  },\n\n  \u002F\u002F Technical presentation with jargon\n  technical: {\n    model: 'ear-3',\n    language: 'en',\n    diarization: true,\n    punctuation: true,\n    profanityFilter: false,\n    vocabulary: [\n      'API', 'SDK', 'microservice', 'Kubernetes', 'Docker',\n      'CI\u002FCD', 'serverless', 'GraphQL', 'PostgreSQL'\n    ],\n  },\n\n  \u002F\u002F Call center \u002F customer support\n  callCenter: {\n    model: 'ear-3',\n    language: 'auto',\n    diarization: true,  \u002F\u002F Important for customer vs agent\n    punctuation: true,\n    profanityFilter: true,\n  },\n\n  \u002F\u002F Medical \u002F healthcare\n  medical: {\n    model: 'ear-3-custom',  \u002F\u002F Enterprise custom model\n    language: 'en',\n    diarization: true,\n    punctuation: true,\n    profanityFilter: false,\n    vocabulary: [\n      \u002F\u002F Medical terminology\n      'diagnosis', 'prognosis', 'contraindication',\n      'hematology', 'cardiology', 'oncology',\n    ],\n  },\n\n  \u002F\u002F Lecture \u002F educational\n  lecture: {\n    model: 'ear-3',\n    language: 'auto',\n    diarization: false,  \u002F\u002F Usually single speaker\n    punctuation: true,\n    profanityFilter: false,\n  },\n\n  \u002F\u002F Podcast \u002F interview\n  podcast: {\n    model: 'ear-3',\n    language: 'auto',\n    diarization: true,\n    punctuation: true,\n    profanityFilter: false,\n  },\n};\n\nexport function getOptimalConfig(useCase: string): ModelConfig {\n  return modelConfigs[useCase] || modelConfigs.meeting;\n}\n```\n\n### Step 4: Streaming Optimization\n\n```typescript\n\u002F\u002F src\u002Ftwinmind\u002Fstreaming\u002Foptimized.ts\nexport interface StreamingConfig {\n  chunkDurationMs: number;     \u002F\u002F Audio chunk size\n  overlapMs: number;           \u002F\u002F Overlap for continuity\n  maxBufferMs: number;         \u002F\u002F Maximum buffer before processing\n  interimResults: boolean;     \u002F\u002F Return partial results\n  endpointDetection: boolean;  \u002F\u002F Auto-detect speech endpoints\n}\n\nconst defaultStreamingConfig: StreamingConfig = {\n  chunkDurationMs: 100,        \u002F\u002F 100ms chunks for low latency\n  overlapMs: 50,               \u002F\u002F 50ms overlap\n  maxBufferMs: 5000,           \u002F\u002F 5 second max buffer\n  interimResults: true,\n  endpointDetection: true,\n};\n\nexport class OptimizedStreamingClient {\n  private config: StreamingConfig;\n  private buffer: Float32Array[] = [];\n  private lastInterimResult = '';\n\n  constructor(config: Partial\u003CStreamingConfig> = {}) {\n    this.config = { ...defaultStreamingConfig, ...config };\n  }\n\n  async processChunk(audioChunk: Float32Array): Promise\u003C{\n    interim?: string;\n    final?: string;\n    confidence: number;\n  }> {\n    this.buffer.push(audioChunk);\n\n    \u002F\u002F Check if we have enough data\n    const totalMs = this.buffer.length * (this.config.chunkDurationMs);\n\n    if (totalMs >= this.config.chunkDurationMs * 3) {\n      \u002F\u002F Process accumulated buffer\n      const result = await this.sendToApi(this.concatenateBuffer());\n\n      if (result.isFinal) {\n        this.buffer = [];  \u002F\u002F Clear buffer on final result\n        return {\n          final: result.text,\n          confidence: result.confidence,\n        };\n      } else {\n        \u002F\u002F Interim result - keep some buffer for context\n        this.trimBuffer();\n        return {\n          interim: result.text,\n          confidence: result.confidence,\n        };\n      }\n    }\n\n    return { confidence: 0 };\n  }\n\n  private concatenateBuffer(): Float32Array {\n    const totalLength = this.buffer.reduce((sum, arr) => sum + arr.length, 0);\n    const result = new Float32Array(totalLength);\n    let offset = 0;\n    for (const arr of this.buffer) {\n      result.set(arr, offset);\n      offset += arr.length;\n    }\n    return result;\n  }\n\n  private trimBuffer(): void {\n    \u002F\u002F Keep last 2 chunks for context overlap\n    if (this.buffer.length > 2) {\n      this.buffer = this.buffer.slice(-2);\n    }\n  }\n\n  private async sendToApi(audio: Float32Array): Promise\u003C{\n    text: string;\n    confidence: number;\n    isFinal: boolean;\n  }> {\n    \u002F\u002F Implementation to send to TwinMind streaming API\n    const client = getTwinMindClient();\n    const response = await client.post('\u002Fstream\u002Fprocess', {\n      audio: Buffer.from(audio.buffer).toString('base64'),\n      interim_results: this.config.interimResults,\n    });\n    return response.data;\n  }\n}\n```\n\n### Step 5: Caching and Deduplication\n\n```typescript\n\u002F\u002F src\u002Ftwinmind\u002Foptimization\u002Fcache.ts\nimport crypto from 'crypto';\n\ninterface CachedTranscript {\n  hash: string;\n  transcriptId: string;\n  createdAt: Date;\n  expiresAt: Date;\n}\n\nclass TranscriptCache {\n  private cache = new Map\u003Cstring, CachedTranscript>();\n  private ttlMs = 24 * 60 * 60 * 1000; \u002F\u002F 24 hours\n\n  \u002F\u002F Generate hash of audio content for deduplication\n  async hashAudio(audioUrl: string): Promise\u003Cstring> {\n    const response = await fetch(audioUrl);\n    const buffer = await response.arrayBuffer();\n\n    return crypto\n      .createHash('sha256')\n      .update(Buffer.from(buffer))\n      .digest('hex');\n  }\n\n  get(hash: string): string | null {\n    const cached = this.cache.get(hash);\n\n    if (!cached) return null;\n\n    if (new Date() > cached.expiresAt) {\n      this.cache.delete(hash);\n      return null;\n    }\n\n    return cached.transcriptId;\n  }\n\n  set(hash: string, transcriptId: string): void {\n    this.cache.set(hash, {\n      hash,\n      transcriptId,\n      createdAt: new Date(),\n      expiresAt: new Date(Date.now() + this.ttlMs),\n    });\n  }\n\n  \u002F\u002F Transcribe with deduplication\n  async transcribeWithCache(audioUrl: string): Promise\u003Cstring> {\n    const hash = await this.hashAudio(audioUrl);\n\n    \u002F\u002F Check cache\n    const cachedId = this.get(hash);\n    if (cachedId) {\n      console.log(`Cache hit for audio: ${hash.substring(0, 8)}...`);\n      return cachedId;\n    }\n\n    \u002F\u002F Process new audio\n    console.log(`Cache miss for audio: ${hash.substring(0, 8)}...`);\n    const client = getTwinMindClient();\n    const result = await client.transcribe(audioUrl);\n\n    \u002F\u002F Store in cache\n    this.set(hash, result.id);\n\n    return result.id;\n  }\n}\n\nexport const transcriptCache = new TranscriptCache();\n```\n\n## Output\n- Performance metrics tracking\n- Audio preprocessing pipeline\n- Model configuration for use cases\n- Streaming optimization\n- Caching and deduplication\n\n## Performance Benchmarks\n\n| Metric | Target | Ear-3 Actual |\n|--------|--------|--------------|\n| Word Error Rate | \u003C 10% | ~5.26% |\n| Diarization Error Rate | \u003C 5% | ~3.8% |\n| Real-time Factor | \u003C 0.5x | ~0.3x |\n| First Word Latency | \u003C 500ms | ~300ms |\n| Languages Supported | 100+ | 140+ |\n\n## Error Handling\n\n| Issue | Cause | Solution |\n|-------|-------|----------|\n| High WER | Poor audio quality | Apply preprocessing |\n| Slow processing | Large file | Use streaming API |\n| Wrong language | Auto-detect failed | Specify language explicitly |\n| Missing speakers | Low audio separation | Improve microphone setup |\n\n## Resources\n- [TwinMind Ear-3 Model](https:\u002F\u002Ftwinmind.com\u002Fear-3)\n- [Audio Best Practices](https:\u002F\u002Ftwinmind.com\u002Fdocs\u002Faudio-quality)\n- [Streaming API](https:\u002F\u002Ftwinmind.com\u002Fdocs\u002Fstreaming)\n\n## Next Steps\nFor cost optimization, see `twinmind-cost-tuning`.\n","# TwinMind Performance Tuning\n\n## Overview\nOptimize TwinMind for better transcription accuracy, faster processing, and improved user experience.\n\n## Prerequisites\n- TwinMind Pro\u002FEnterprise account\n- Understanding of audio processing concepts\n- Access to quality metrics and logs\n\n## Instructions\n\n### Step 1: Understand Performance Metrics\n\n```typescript\n\u002F\u002F src\u002Ftwinmind\u002Fmetrics\u002Fperformance.ts\nexport interface TranscriptionMetrics {\n  \u002F\u002F Accuracy metrics\n  wordErrorRate: number;         \u002F\u002F WER - lower is better (Ear-3: ~5.26%)\n  diarizationErrorRate: number;  \u002F\u002F DER - speaker labeling accuracy (~3.8%)\n  confidenceScore: number;       \u002F\u002F Average confidence (0-1)\n\n  \u002F\u002F Timing metrics\n  processingTime: number;        \u002F\u002F Time to complete (ms)\n  realtimeFactor: number;        \u002F\u002F Processing time \u002F audio duration\n  firstWordLatency: number;      \u002F\u002F Time to first result (streaming)\n\n  \u002F\u002F Quality metrics\n  speakerCount: number;          \u002F\u002F Detected speakers\n  languageDetected: string;      \u002F\u002F Detected language\n  noiseLevel: string;            \u002F\u002F low\u002Fmedium\u002Fhigh\n}\n\nexport async function getTranscriptionMetrics(transcriptId: string): Promise\u003CTranscriptionMetrics> {\n  const client = getTwinMindClient();\n  const response = await client.get(`\u002Ftranscripts\u002F${transcriptId}\u002Fmetrics`);\n  return response.data;\n}\n\n\u002F\u002F Log and analyze metrics\nexport function analyzePerformance(metrics: TranscriptionMetrics): string[] {\n  const recommendations: string[] = [];\n\n  if (metrics.wordErrorRate > 0.10) {\n    recommendations.push('High WER - consider audio quality improvements');\n  }\n\n  if (metrics.diarizationErrorRate > 0.05) {\n    recommendations.push('Speaker labeling issues - ensure clear audio separation');\n  }\n\n  if (metrics.realtimeFactor > 0.5) {\n    recommendations.push('Slow processing - consider model optimization');\n  }\n\n  if (metrics.noiseLevel === 'high') {\n    recommendations.push('High background noise - recommend noise reduction');\n  }\n\n  return recommendations;\n}\n```\n\n### Step 2: Audio Quality Optimization\n\n```typescript\n\u002F\u002F src\u002Ftwinmind\u002Fpreprocessing\u002Faudio.ts\nimport ffmpeg from 'fluent-ffmpeg';\n\ninterface AudioPreprocessOptions {\n  targetSampleRate?: number;   \u002F\u002F Default: 16000 Hz\n  channels?: number;           \u002F\u002F Default: 1 (mono)\n  noiseReduction?: boolean;    \u002F\u002F Enable noise reduction\n  normalization?: boolean;     \u002F\u002F Normalize volume\n  format?: string;             \u002F\u002F Output format (mp3, wav)\n}\n\nconst defaultOptions: AudioPreprocessOptions = {\n  targetSampleRate: 16000,\n  channels: 1,\n  noiseReduction: true,\n  normalization: true,\n  format: 'mp3',\n};\n\nexport async function preprocessAudio(\n  inputPath: string,\n  outputPath: string,\n  options: AudioPreprocessOptions = {}\n): Promise\u003Cvoid> {\n  const opts = { ...defaultOptions, ...options };\n\n  return new Promise((resolve, reject) => {\n    let command = ffmpeg(inputPath)\n      .audioFrequency(opts.targetSampleRate!)\n      .audioChannels(opts.channels!);\n\n    \u002F\u002F Noise reduction using highpass and lowpass filters\n    if (opts.noiseReduction) {\n      command = command.audioFilters([\n        'highpass=f=200',     \u002F\u002F Remove low frequency noise\n        'lowpass=f=3000',     \u002F\u002F Remove high frequency noise\n        'afftdn=nf=-25',      \u002F\u002F FFT denoiser\n      ]);\n    }\n\n    \u002F\u002F Volume normalization\n    if (opts.normalization) {\n      command = command.audioFilters([\n        'loudnorm=I=-16:TP=-1.5:LRA=11', \u002F\u002F EBU R128 loudness normalization\n      ]);\n    }\n\n    command\n      .toFormat(opts.format!)\n      .on('end', () => resolve())\n      .on('error', reject)\n      .save(outputPath);\n  });\n}\n\n\u002F\u002F Audio quality assessment\nexport async function assessAudioQuality(filePath: string): Promise\u003C{\n  quality: 'excellent' | 'good' | 'fair' | 'poor';\n  issues: string[];\n  recommendations: string[];\n}> {\n  const issues: string[] = [];\n  const recommendations: string[] = [];\n\n  \u002F\u002F Get audio metadata\n  const metadata = await getAudioMetadata(filePath);\n\n  \u002F\u002F Check sample rate\n  if (metadata.sampleRate \u003C 16000) {\n    issues.push(`Low sample rate: ${metadata.sampleRate} Hz`);\n    recommendations.push('Use 16 kHz or higher sample rate');\n  }\n\n  \u002F\u002F Check bit depth\n  if (metadata.bitDepth && metadata.bitDepth \u003C 16) {\n    issues.push(`Low bit depth: ${metadata.bitDepth} bits`);\n    recommendations.push('Use 16-bit or higher audio');\n  }\n\n  \u002F\u002F Check for clipping\n  if (metadata.peakLevel && metadata.peakLevel > -1) {\n    issues.push('Audio clipping detected');\n    recommendations.push('Reduce recording volume to prevent distortion');\n  }\n\n  \u002F\u002F Check noise floor\n  if (metadata.noiseFloor && metadata.noiseFloor > -40) {\n    issues.push(`High noise floor: ${metadata.noiseFloor} dB`);\n    recommendations.push('Use noise reduction or improve recording environment');\n  }\n\n  const quality = issues.length === 0 ? 'excellent' :\n                  issues.length \u003C= 1 ? 'good' :\n                  issues.length \u003C= 2 ? 'fair' : 'poor';\n\n  return { quality, issues, recommendations };\n}\n```\n\n### Step 3: Model Selection and Configuration\n\n```typescript\n\u002F\u002F src\u002Ftwinmind\u002Fmodels\u002Fconfig.ts\nexport interface ModelConfig {\n  model: 'ear-3' | 'ear-2' | 'ear-3-custom';\n  language?: string;           \u002F\u002F 'auto' or ISO code (en, es, fr, etc.)\n  diarization: boolean;        \u002F\u002F Speaker labeling\n  punctuation: boolean;        \u002F\u002F Auto-punctuation\n  profanityFilter: boolean;    \u002F\u002F Filter explicit content\n  vocabulary?: string[];       \u002F\u002F Custom vocabulary boost\n  speakerHints?: string[];     \u002F\u002F Expected speaker names\n}\n\n\u002F\u002F Optimized configs for different scenarios\nexport const modelConfigs: Record\u003Cstring, ModelConfig> = {\n  \u002F\u002F Standard meeting transcription\n  meeting: {\n    model: 'ear-3',\n    language: 'auto',\n    diarization: true,\n    punctuation: true,\n    profanityFilter: false,\n  },\n\n  \u002F\u002F Technical presentation with jargon\n  technical: {\n    model: 'ear-3',\n    language: 'en',\n    diarization: true,\n    punctuation: true,\n    profanityFilter: false,\n    vocabulary: [\n      'API', 'SDK', 'microservice', 'Kubernetes', 'Docker',\n      'CI\u002FCD', 'serverless', 'GraphQL', 'PostgreSQL'\n    ],\n  },\n\n  \u002F\u002F Call center \u002F customer support\n  callCenter: {\n    model: 'ear-3',\n    language: 'auto',\n    diarization: true,  \u002F\u002F Important for customer vs agent\n    punctuation: true,\n    profanityFilter: true,\n  },\n\n  \u002F\u002F Medical \u002F healthcare\n  medical: {\n    model: 'ear-3-custom',  \u002F\u002F Enterprise custom model\n    language: 'en',\n    diarization: true,\n    punctuation: true,\n    profanityFilter: false,\n    vocabulary: [\n      \u002F\u002F Medical terminology\n      'diagnosis', 'prognosis', 'contraindication',\n      'hematology', 'cardiology', 'oncology',\n    ],\n  },\n\n  \u002F\u002F Lecture \u002F educational\n  lecture: {\n    model: 'ear-3',\n    language: 'auto',\n    diarization: false,  \u002F\u002F Usually single speaker\n    punctuation: true,\n    profanityFilter: false,\n  },\n\n  \u002F\u002F Podcast \u002F interview\n  podcast: {\n    model: 'ear-3',\n    language: 'auto',\n    diarization: true,\n    punctuation: true,\n    profanityFilter: false,\n  },\n};\n\nexport function getOptimalConfig(useCase: string): ModelConfig {\n  return modelConfigs[useCase] || modelConfigs.meeting;\n}\n```\n\n### Step 4: Streaming Optimization\n\n```typescript\n\u002F\u002F src\u002Ftwinmind\u002Fstreaming\u002Foptimized.ts\nexport interface StreamingConfig {\n  chunkDurationMs: number;     \u002F\u002F Audio chunk size\n  overlapMs: number;           \u002F\u002F Overlap for continuity\n  maxBufferMs: number;         \u002F\u002F Maximum buffer before processing\n  interimResults: boolean;     \u002F\u002F Return partial results\n  endpointDetection: boolean;  \u002F\u002F Auto-detect speech endpoints\n}\n\nconst defaultStreamingConfig: StreamingConfig = {\n  chunkDurationMs: 100,        \u002F\u002F 100ms chunks for low latency\n  overlapMs: 50,               \u002F\u002F 50ms overlap\n  maxBufferMs: 5000,           \u002F\u002F 5 second max buffer\n  interimResults: true,\n  endpointDetection: true,\n};\n\nexport class OptimizedStreamingClient {\n  private config: StreamingConfig;\n  private buffer: Float32Array[] = [];\n  private lastInterimResult = '';\n\n  constructor(config: Partial\u003CStreamingConfig> = {}) {\n    this.config = { ...defaultStreamingConfig, ...config };\n  }\n\n  async processChunk(audioChunk: Float32Array): Promise\u003C{\n    interim?: string;\n    final?: string;\n    confidence: number;\n  }> {\n    this.buffer.push(audioChunk);\n\n    \u002F\u002F Check if we have enough data\n    const totalMs = this.buffer.length * (this.config.chunkDurationMs);\n\n    if (totalMs >= this.config.chunkDurationMs * 3) {\n      \u002F\u002F Process accumulated buffer\n      const result = await this.sendToApi(this.concatenateBuffer());\n\n      if (result.isFinal) {\n        this.buffer = [];  \u002F\u002F Clear buffer on final result\n        return {\n          final: result.text,\n          confidence: result.confidence,\n        };\n      } else {\n        \u002F\u002F Interim result - keep some buffer for context\n        this.trimBuffer();\n        return {\n          interim: result.text,\n          confidence: result.confidence,\n        };\n      }\n    }\n\n    return { confidence: 0 };\n  }\n\n  private concatenateBuffer(): Float32Array {\n    const totalLength = this.buffer.reduce((sum, arr) => sum + arr.length, 0);\n    const result = new Float32Array(totalLength);\n    let offset = 0;\n    for (const arr of this.buffer) {\n      result.set(arr, offset);\n      offset += arr.length;\n    }\n    return result;\n  }\n\n  private trimBuffer(): void {\n    \u002F\u002F Keep last 2 chunks for context overlap\n    if (this.buffer.length > 2) {\n      this.buffer = this.buffer.slice(-2);\n    }\n  }\n\n  private async sendToApi(audio: Float32Array): Promise\u003C{\n    text: string;\n    confidence: number;\n    isFinal: boolean;\n  }> {\n    \u002F\u002F Implementation to send to TwinMind streaming API\n    const client = getTwinMindClient();\n    const response = await client.post('\u002Fstream\u002Fprocess', {\n      audio: Buffer.from(audio.buffer).toString('base64'),\n      interim_results: this.config.interimResults,\n    });\n    return response.data;\n  }\n}\n```\n\n### Step 5: Caching and Deduplication\n\n```typescript\n\u002F\u002F src\u002Ftwinmind\u002Foptimization\u002Fcache.ts\nimport crypto from 'crypto';\n\ninterface CachedTranscript {\n  hash: string;\n  transcriptId: string;\n  createdAt: Date;\n  expiresAt: Date;\n}\n\nclass TranscriptCache {\n  private cache = new Map\u003Cstring, CachedTranscript>();\n  private ttlMs = 24 * 60 * 60 * 1000; \u002F\u002F 24 hours\n\n  \u002F\u002F Generate hash of audio content for deduplication\n  async hashAudio(audioUrl: string): Promise\u003Cstring> {\n    const response = await fetch(audioUrl);\n    const buffer = await response.arrayBuffer();\n\n    return crypto\n      .createHash('sha256')\n      .update(Buffer.from(buffer))\n      .digest('hex');\n  }\n\n  get(hash: string): string | null {\n    const cached = this.cache.get(hash);\n\n    if (!cached) return null;\n\n    if (new Date() > cached.expiresAt) {\n      this.cache.delete(hash);\n      return null;\n    }\n\n    return cached.transcriptId;\n  }\n\n  set(hash: string, transcriptId: string): void {\n    this.cache.set(hash, {\n      hash,\n      transcriptId,\n      createdAt: new Date(),\n      expiresAt: new Date(Date.now() + this.ttlMs),\n    });\n  }\n\n  \u002F\u002F Transcribe with deduplication\n  async transcribeWithCache(audioUrl: string): Promise\u003Cstring> {\n    const hash = await this.hashAudio(audioUrl);\n\n    \u002F\u002F Check cache\n    const cachedId = this.get(hash);\n    if (cachedId) {\n      console.log(`Cache hit for audio: ${hash.substring(0, 8)}...`);\n      return cachedId;\n    }\n\n    \u002F\u002F Process new audio\n    console.log(`Cache miss for audio: ${hash.substring(0, 8)}...`);\n    const client = getTwinMindClient();\n    const result = await client.transcribe(audioUrl);\n\n    \u002F\u002F Store in cache\n    this.set(hash, result.id);\n\n    return result.id;\n  }\n}\n\nexport const transcriptCache = new TranscriptCache();\n```\n\n## Output\n- Performance metrics tracking\n- Audio preprocessing pipeline\n- Model configuration for use cases\n- Streaming optimization\n- Caching and deduplication\n\n## Performance Benchmarks\n\n| Metric | Target | Ear-3 Actual |\n|--------|--------|--------------|\n| Word Error Rate | \u003C 10% | ~5.26% |\n| Diarization Error Rate | \u003C 5% | ~3.8% |\n| Real-time Factor | \u003C 0.5x | ~0.3x |\n| First Word Latency | \u003C 500ms | ~300ms |\n| Languages Supported | 100+ | 140+ |\n\n## Error Handling\n\n| Issue | Cause | Solution |\n|-------|-------|----------|\n| High WER | Poor audio quality | Apply preprocessing |\n| Slow processing | Large file | Use streaming API |\n| Wrong language | Auto-detect failed | Specify language explicitly |\n| Missing speakers | Low audio separation | Improve microphone setup |\n\n## Resources\n- [TwinMind Ear-3 Model](https:\u002F\u002Ftwinmind.com\u002Fear-3)\n- [Audio Best Practices](https:\u002F\u002Ftwinmind.com\u002Fdocs\u002Faudio-quality)\n- [Streaming API](https:\u002F\u002Ftwinmind.com\u002Fdocs\u002Fstreaming)\n\n## Next Steps\nFor cost optimization, see `twinmind-cost-tuning`.\n",{"name":5,"author":46,"license":15,"version":47,"description":48,"allowed-tools":49},"Jeremy Longshore \u003Cjeremy@intentsolutions.io>","1.0.0","Optimize TwinMind transcription accuracy and processing speed.\nUse when improving transcription quality, reducing latency,\nor tuning model parameters for specific use cases.\nTrigger with phrases like \"twinmind performance\", \"improve transcription accuracy\",\n\"faster twinmind\", \"optimize twinmind\", \"transcription quality\".\n","Read, Write, Edit","main","90b50ed0778745a80fd5bfb13b1843d45b9e82f7","tests\u002Fext_conformance\u002Fartifacts\u002Fplugins-community\u002Fplugins\u002Fsaas-packs\u002Ftwinmind-pack\u002Fskills\u002Ftwinmind-performance-tuning",233,"https:\u002F\u002Fgithub.com\u002FDicklesworthstone\u002Fpi_agent_rust","https:\u002F\u002Fgithub.com\u002FDicklesworthstone","https:\u002F\u002Fwww.jeffreyemanuel.com\u002F","Building in NY","2026-11-15 01:54:15",{"source":8,"skillName":5,"commands":60,"prompt":70,"zipAvailable":71},[61,64,67],{"tool":62,"command":63},"npx","npx skills add Dicklesworthstone\u002Fpi_agent_rust --skill twinmind-performance-tuning",{"tool":65,"command":66},"bunx","bunx skills add Dicklesworthstone\u002Fpi_agent_rust --skill twinmind-performance-tuning",{"tool":68,"command":69},"pnpm","pnpm dlx skills add Dicklesworthstone\u002Fpi_agent_rust --skill twinmind-performance-tuning","安装技能，技能地址：https:\u002F\u002Fgithub.com\u002FDicklesworthstone\u002Fpi_agent_rust\u002Fblob\u002Fmain\u002Ftests\u002Fext_conformance\u002Fartifacts\u002Fplugins-community\u002Fplugins\u002Fsaas-packs\u002Ftwinmind-pack\u002Fskills\u002Ftwinmind-performance-tuning\u002FSKILL.md",true,[],[74,107,141,164,189,216],{"skillNo":75,"name":76,"nameZh":6,"description":77,"descriptionZh":6,"repositoryFullName":8,"ownerLogin":9,"ownerAvatarUrl":10,"ownerAccountNo":11,"ownerDisplayName":12,"repositoryNo":13,"starsCount":14,"licenseSpdxId":78,"licenseFileUrl":79,"sourceHtmlUrl":80,"categories":81,"keywords":87,"publishedAt":106},"01M6FCR7G5A005P364Z6ZWCJDP","theme-factory","Toolkit for styling artifacts with a theme. These artifacts can be slides, docs, reportings, HTML landing pages, etc. There are 10 pre-set themes with colors\u002Ffonts that you can apply to any artifact that has been creating, or can generate a new theme on-the-fly.","Apache-2.0","https:\u002F\u002Fgithub.com\u002FDicklesworthstone\u002Fpi_agent_rust\u002Fblob\u002Fmain\u002Ftests\u002Fext_conformance\u002Fartifacts\u002Ftemplates-davila7\u002Fcli-tool\u002Fcomponents\u002Fskills\u002Fcreative-design\u002Ftheme-factory\u002FLICENSE.txt","https:\u002F\u002Fgithub.com\u002FDicklesworthstone\u002Fpi_agent_rust\u002Fblob\u002Fmain\u002Ftests\u002Fext_conformance\u002Fartifacts\u002Ftemplates-davila7\u002Fcli-tool\u002Fcomponents\u002Fskills\u002Fcreative-design\u002Ftheme-factory\u002FSKILL.md",[82],{"categoryNo":83,"categoryName":84,"taxonomyType":85,"taxonomyTypeDesc":86},"01JBSEEDCAT000000000000015","视觉设计","SCENARIO","使用场景",[88,89,90,91,92,93,94,95,96,97,98,99,100,101,102,103,26,104,105],"theme factory","factory","theme","docs","toolkit","generate","artifacts","SET","fly","slides","landing","html","styling","artifact","pages","themes","you","fonts","2026-11-11 15:11:14",{"skillNo":108,"name":109,"nameZh":6,"description":110,"descriptionZh":6,"repositoryFullName":8,"ownerLogin":9,"ownerAvatarUrl":10,"ownerAccountNo":11,"ownerDisplayName":12,"repositoryNo":13,"starsCount":14,"licenseSpdxId":78,"licenseFileUrl":111,"sourceHtmlUrl":112,"categories":113,"keywords":117,"publishedAt":41},"01M6FEGPPK0ZTKCCM0R3E3KS33","web-artifacts-builder","Suite of tools for creating elaborate, multi-component claude.ai HTML artifacts using modern frontend web technologies (React, Tailwind CSS, shadcn\u002Fui). Use for complex artifacts requiring state management, routing, or shadcn\u002Fui components - not for simple single-file HTML\u002FJSX artifacts.","https:\u002F\u002Fgithub.com\u002FDicklesworthstone\u002Fpi_agent_rust\u002Fblob\u002Fmain\u002Ftests\u002Fext_conformance\u002Fartifacts\u002Ftemplates-davila7\u002Fcli-tool\u002Fcomponents\u002Fskills\u002Futilities\u002Fweb-artifacts-builder\u002FLICENSE.txt","https:\u002F\u002Fgithub.com\u002FDicklesworthstone\u002Fpi_agent_rust\u002Fblob\u002Fmain\u002Ftests\u002Fext_conformance\u002Fartifacts\u002Ftemplates-davila7\u002Fcli-tool\u002Fcomponents\u002Fskills\u002Futilities\u002Fweb-artifacts-builder\u002FSKILL.md",[114],{"categoryNo":115,"categoryName":116,"taxonomyType":85,"taxonomyTypeDesc":86},"01JBSEEDCAT000000000000009","前端开发",[118,119,94,120,121,122,123,124,125,126,127,128,129,130,131,132,133,134,34,99,135,136,137,138,25,139,140],"web artifacts builder","web","artifacts builder","builder","frontend","css","components","react","component","routing","state management","state","shadcn","file","ui","Tools","Tailwind","single-file","simple","ui components","multi","jsx","suite",{"skillNo":142,"name":143,"nameZh":6,"description":144,"descriptionZh":6,"repositoryFullName":8,"ownerLogin":9,"ownerAvatarUrl":10,"ownerAccountNo":11,"ownerDisplayName":12,"repositoryNo":13,"starsCount":14,"licenseSpdxId":78,"licenseFileUrl":145,"sourceHtmlUrl":146,"categories":147,"keywords":152,"publishedAt":41},"01M6FEE9VTK1W4GY21AKREJXRG","webapp-testing","Toolkit for interacting with and testing local web applications using Playwright. Supports verifying frontend functionality, debugging UI behavior, capturing browser screenshots, and viewing browser logs.","https:\u002F\u002Fgithub.com\u002FDicklesworthstone\u002Fpi_agent_rust\u002Fblob\u002Fmain\u002Ftests\u002Fext_conformance\u002Fartifacts\u002Ftemplates-davila7\u002Fcli-tool\u002Fcomponents\u002Fskills\u002Fsecurity\u002Fwebapp-testing\u002FLICENSE.txt","https:\u002F\u002Fgithub.com\u002FDicklesworthstone\u002Fpi_agent_rust\u002Fblob\u002Fmain\u002Ftests\u002Fext_conformance\u002Fartifacts\u002Ftemplates-davila7\u002Fcli-tool\u002Fcomponents\u002Fskills\u002Fsecurity\u002Fwebapp-testing\u002FSKILL.md",[148,151],{"categoryNo":149,"categoryName":150,"taxonomyType":85,"taxonomyTypeDesc":86},"01JBSEEDCAT000000000000008","测试与质量",{"categoryNo":115,"categoryName":116,"taxonomyType":85,"taxonomyTypeDesc":86},[153,154,155,122,156,157,119,158,92,159,132,160,161,162,163,26,25],"testing","webapp testing","webapp","screenshots","playwright","debugging","browser","logs","applications","behavior","local",{"skillNo":165,"name":166,"nameZh":6,"description":167,"descriptionZh":6,"repositoryFullName":8,"ownerLogin":9,"ownerAvatarUrl":10,"ownerAccountNo":11,"ownerDisplayName":12,"repositoryNo":13,"starsCount":14,"licenseSpdxId":15,"licenseFileUrl":168,"sourceHtmlUrl":169,"categories":170,"keywords":171,"publishedAt":41},"01M6FE20F7ZTQBFWKF4MPJ2SBX","neuropixels-analysis","Neuropixels neural recording analysis. Load SpikeGLX\u002FOpenEphys data, preprocess, motion correction, Kilosort4 spike sorting, quality metrics, Allen\u002FIBL curation, AI-assisted visual analysis, for Neuropixels 1.0\u002F2.0 extracellular electrophysiology. Use when working with neural recordings, spike sorting, extracellular electrophysiology, or when the user mentions Neuropixels, SpikeGLX, Open Ephys, Kilosort, quality metrics, or unit curation.","https:\u002F\u002Fgithub.com\u002FDicklesworthstone\u002Fpi_agent_rust\u002Fblob\u002Fmain\u002Ftests\u002Fext_conformance\u002Fartifacts\u002Ftemplates-davila7\u002Fcli-tool\u002Fcomponents\u002Fskills\u002Fscientific\u002Fneuropixels-analysis\u002FLICENSE.txt","https:\u002F\u002Fgithub.com\u002FDicklesworthstone\u002Fpi_agent_rust\u002Fblob\u002Fmain\u002Ftests\u002Fext_conformance\u002Fartifacts\u002Ftemplates-davila7\u002Fcli-tool\u002Fcomponents\u002Fskills\u002Fscientific\u002Fneuropixels-analysis\u002FSKILL.md",[],[172,173,174,175,176,177,38,178,34,179,180,181,182,183,184,185,26,186,187,188,25,23],"analysis","neuropixels analysis","spike","metrics","data","open","user","recordings","Neural","motion","visual","quality metrics","unit","sorting","curation","recording","load",{"skillNo":190,"name":191,"nameZh":6,"description":192,"descriptionZh":6,"repositoryFullName":8,"ownerLogin":9,"ownerAvatarUrl":10,"ownerAccountNo":11,"ownerDisplayName":12,"repositoryNo":13,"starsCount":14,"licenseSpdxId":78,"licenseFileUrl":193,"sourceHtmlUrl":194,"categories":195,"keywords":199,"publishedAt":41},"01M6FDP0PDK6F1GGX395C7625Z","skill-creator","Guide for creating effective skills. This skill should be used when users want to create a new skill (or update an existing skill) that extends Claude's capabilities with specialized knowledge, workflows, or tool integrations.","https:\u002F\u002Fgithub.com\u002FDicklesworthstone\u002Fpi_agent_rust\u002Fblob\u002Fmain\u002Ftests\u002Fext_conformance\u002Fartifacts\u002Ftemplates-davila7\u002Fcli-tool\u002Fcomponents\u002Fskills\u002Fproductivity\u002Fskill-creator\u002FLICENSE.txt","https:\u002F\u002Fgithub.com\u002FDicklesworthstone\u002Fpi_agent_rust\u002Fblob\u002Fmain\u002Ftests\u002Fext_conformance\u002Fartifacts\u002Ftemplates-davila7\u002Fcli-tool\u002Fcomponents\u002Fskills\u002Fproductivity\u002Fskill-creator\u002FSKILL.md",[196],{"categoryNo":197,"categoryName":198,"taxonomyType":85,"taxonomyTypeDesc":86},"01JBSEEDCAT000000000000023","Skill与智能体开发",[200,201,202,203,204,205,206,207,208,209,210,211,212,213,26,25,214,23,215],"skill creator","creator","workflows","create","update","capabilities","knowledge","tool","integrations","specialized","guide","new skill","existing skill","users","Existing","new",{"skillNo":217,"name":218,"nameZh":6,"description":219,"descriptionZh":6,"repositoryFullName":8,"ownerLogin":9,"ownerAvatarUrl":10,"ownerAccountNo":11,"ownerDisplayName":12,"repositoryNo":13,"starsCount":14,"licenseSpdxId":15,"licenseFileUrl":220,"sourceHtmlUrl":221,"categories":222,"keywords":224,"publishedAt":41},"01M6FDM8Y7VK7NYG9SM7QPA3PW","notebooklm","Use this skill to query your Google NotebookLM notebooks directly from Claude Code for source-grounded, citation-backed answers from Gemini. Browser automation, library management, persistent auth. Drastically reduced hallucinations through document-only responses.","https:\u002F\u002Fgithub.com\u002FDicklesworthstone\u002Fpi_agent_rust\u002Fblob\u002Fmain\u002Ftests\u002Fext_conformance\u002Fartifacts\u002Ftemplates-davila7\u002Fcli-tool\u002Fcomponents\u002Fskills\u002Fproductivity\u002Fnotebooklm\u002FLICENSE","https:\u002F\u002Fgithub.com\u002FDicklesworthstone\u002Fpi_agent_rust\u002Fblob\u002Fmain\u002Ftests\u002Fext_conformance\u002Fartifacts\u002Ftemplates-davila7\u002Fcli-tool\u002Fcomponents\u002Fskills\u002Fproductivity\u002Fnotebooklm\u002FSKILL.md",[223],{"categoryNo":197,"categoryName":198,"taxonomyType":85,"taxonomyTypeDesc":86},[218,225,226,227,228,229,230,159,231,232,34,233,234,235,236,25,237],"google","browser automation","gemini","automation","auth","code","library","management","only","document","query","Citation","source"]