AGENT SKILL

speak-performance-tuning

Optimize Speak API performance with caching, audio preprocessing, and connection pooling. Use when experiencing slow API responses, implementing caching strategies, or optimizing request throughput for language learning applications. Trigger with phrases like "speak performance", "optimize speak", "speak latency", "speak caching", "speak slow", "speak audio optimization".

使用场景图像与多媒体
License 许可
MIT
最近更新
2026-11-15
Stars
★ 2.0k
Forks
233

01 前言 Front Matter

name
speak-performance-tuning
author
Jeremy Longshore <jeremy@intentsolutions.io>
license
MIT
version
1.0.0
description
Optimize Speak API performance with caching, audio preprocessing, and connection pooling. Use when experiencing slow API responses, implementing caching strategies, or optimizing request throughput for language learning applications. Trigger with phrases like "speak performance", "optimize speak", "speak latency", "speak caching", "speak slow", "speak audio optimization".
allowed-tools
Read, Write, Edit

03 SKILL.md 正文

Speak Performance Tuning

Overview

Optimize Speak API performance with caching, audio preprocessing, and connection pooling for language learning applications.

Prerequisites

  • Speak SDK installed
  • Understanding of async patterns
  • Redis or in-memory cache available (optional)
  • Performance monitoring in place

Latency Benchmarks

Operation P50 P95 P99
Session Start 200ms 500ms 1000ms
Tutor Prompt 150ms 300ms 600ms
Text Response Submit 100ms 250ms 500ms
Audio Recognition 500ms 1500ms 3000ms
Pronunciation Scoring 800ms 2000ms 4000ms

Audio Optimization

Pre-processing Audio Before Upload

class AudioOptimizer {
  // Optimize audio for Speak's speech recognition
  async optimizeForRecognition(audioData: ArrayBuffer): Promise<ArrayBuffer> {
    // Convert to optimal format: 16kHz mono PCM WAV
    const audioContext = new AudioContext({ sampleRate: 16000 });
    const audioBuffer = await audioContext.decodeAudioData(audioData);

    // Convert to mono if stereo
    const monoBuffer = this.toMono(audioBuffer);

    // Normalize audio levels
    const normalizedBuffer = this.normalize(monoBuffer);

    // Remove silence at start/end
    const trimmedBuffer = this.trimSilence(normalizedBuffer);

    // Encode as WAV
    return this.encodeWav(trimmedBuffer);
  }

  private toMono(buffer: AudioBuffer): AudioBuffer {
    if (buffer.numberOfChannels === 1) return buffer;

    const monoData = new Float32Array(buffer.length);
    const left = buffer.getChannelData(0);
    const right = buffer.getChannelData(1);

    for (let i = 0; i < buffer.length; i++) {
      monoData[i] = (left[i] + right[i]) / 2;
    }

    const ctx = new OfflineAudioContext(1, buffer.length, buffer.sampleRate);
    const newBuffer = ctx.createBuffer(1, buffer.length, buffer.sampleRate);
    newBuffer.copyToChannel(monoData, 0);
    return newBuffer;
  }

  private normalize(buffer: AudioBuffer): AudioBuffer {
    const data = buffer.getChannelData(0);
    let max = 0;

    for (let i = 0; i < data.length; i++) {
      max = Math.max(max, Math.abs(data[i]));
    }

    if (max > 0 && max < 0.9) {
      const factor = 0.9 / max;
      for (let i = 0; i < data.length; i++) {
        data[i] *= factor;
      }
    }

    return buffer;
  }

  private trimSilence(buffer: AudioBuffer, threshold = 0.01): AudioBuffer {
    const data = buffer.getChannelData(0);
    let start = 0;
    let end = data.length;

    // Find start
    for (let i = 0; i < data.length; i++) {
      if (Math.abs(data[i]) > threshold) {
        start = Math.max(0, i - 1000); // Keep small buffer
        break;
      }
    }

    // Find end
    for (let i = data.length - 1; i >= 0; i--) {
      if (Math.abs(data[i]) > threshold) {
        end = Math.min(data.length, i + 1000);
        break;
      }
    }

    const trimmedLength = end - start;
    const ctx = new OfflineAudioContext(1, trimmedLength, buffer.sampleRate);
    const newBuffer = ctx.createBuffer(1, trimmedLength, buffer.sampleRate);
    newBuffer.copyToChannel(data.slice(start, end), 0);
    return newBuffer;
  }
}

Streaming Audio for Real-time Recognition

class StreamingRecognizer {
  private chunks: ArrayBuffer[] = [];
  private processingPromise: Promise<void> | null = null;

  async streamAudioChunk(chunk: ArrayBuffer): Promise<PartialResult | null> {
    this.chunks.push(chunk);

    // Process in batches to reduce API calls
    if (this.chunks.length >= 5 || this.shouldProcess()) {
      return this.processAccumulated();
    }

    return null;
  }

  private async processAccumulated(): Promise<PartialResult> {
    const combinedSize = this.chunks.reduce((sum, c) => sum + c.byteLength, 0);
    const combined = new ArrayBuffer(combinedSize);
    const view = new Uint8Array(combined);

    let offset = 0;
    for (const chunk of this.chunks) {
      view.set(new Uint8Array(chunk), offset);
      offset += chunk.byteLength;
    }

    this.chunks = [];

    const result = await speakClient.speech.recognizeStream(combined);
    return result;
  }
}

Caching Strategy

Response Caching for Static Content

import { LRUCache } from 'lru-cache';

// Cache tutor prompts and audio URLs
const promptCache = new LRUCache<string, TutorPrompt>({
  max: 500,
  ttl: 60 * 60 * 1000, // 1 hour
  updateAgeOnGet: true,
});

// Cache vocabulary lookups
const vocabularyCache = new LRUCache<string, VocabularyEntry>({
  max: 10000,
  ttl: 24 * 60 * 60 * 1000, // 24 hours
});

async function getCachedVocabulary(
  word: string,
  language: string
): Promise<VocabularyEntry> {
  const key = `${language}:${word}`;
  const cached = vocabularyCache.get(key);
  if (cached) return cached;

  const entry = await speakClient.vocabulary.lookup(word, language);
  vocabularyCache.set(key, entry);
  return entry;
}

Redis Caching for Distributed Systems

import Redis from 'ioredis';

const redis = new Redis(process.env.REDIS_URL);

async function cachedWithRedis<T>(
  key: string,
  fetcher: () => Promise<T>,
  ttlSeconds = 3600
): Promise<T> {
  const cached = await redis.get(key);
  if (cached) {
    return JSON.parse(cached);
  }

  const result = await fetcher();
  await redis.setex(key, ttlSeconds, JSON.stringify(result));
  return result;
}

// Cache user progress
async function getUserProgress(userId: string): Promise<UserProgress> {
  return cachedWithRedis(
    `speak:progress:${userId}`,
    () => speakClient.users.getProgress(userId),
    300 // 5 minutes
  );
}

Audio Asset Caching

// Pre-fetch and cache audio assets
class AudioAssetCache {
  private cache: Map<string, ArrayBuffer> = new Map();
  private preloadQueue: Set<string> = new Set();

  async preloadLessonAudio(lessonId: string): Promise<void> {
    const lesson = await speakClient.lessons.get(lessonId);

    // Pre-fetch all audio for the lesson
    const audioUrls = lesson.items.map(item => item.audioUrl);

    await Promise.all(
      audioUrls.map(async (url) => {
        if (!this.cache.has(url) && !this.preloadQueue.has(url)) {
          this.preloadQueue.add(url);
          const response = await fetch(url);
          const buffer = await response.arrayBuffer();
          this.cache.set(url, buffer);
          this.preloadQueue.delete(url);
        }
      })
    );
  }

  async getAudio(url: string): Promise<ArrayBuffer> {
    const cached = this.cache.get(url);
    if (cached) return cached;

    const response = await fetch(url);
    const buffer = await response.arrayBuffer();
    this.cache.set(url, buffer);
    return buffer;
  }
}

Connection Optimization

import { Agent } from 'https';

// Keep-alive connection pooling
const agent = new Agent({
  keepAlive: true,
  maxSockets: 10,
  maxFreeSockets: 5,
  timeout: 60000,
});

const client = new SpeakClient({
  apiKey: process.env.SPEAK_API_KEY!,
  appId: process.env.SPEAK_APP_ID!,
  httpAgent: agent,
  timeout: 30000,
});

Request Batching

import DataLoader from 'dataloader';

// Batch vocabulary lookups
const vocabularyLoader = new DataLoader<string, VocabularyEntry>(
  async (words) => {
    // Batch API call
    const results = await speakClient.vocabulary.batchLookup(words);
    return words.map(word => results.find(r => r.word === word) || null);
  },
  {
    maxBatchSize: 50,
    batchScheduleFn: callback => setTimeout(callback, 50),
  }
);

// Usage - automatically batched
const [word1, word2, word3] = await Promise.all([
  vocabularyLoader.load('hola'),
  vocabularyLoader.load('buenos'),
  vocabularyLoader.load('días'),
]);

Performance Monitoring

interface SpeakMetrics {
  operation: string;
  duration: number;
  success: boolean;
  audioSize?: number;
}

async function measuredSpeakCall<T>(
  operation: string,
  fn: () => Promise<T>,
  metadata?: Record<string, any>
): Promise<T> {
  const start = performance.now();
  try {
    const result = await fn();
    const duration = performance.now() - start;

    // Log metrics
    console.log({
      operation,
      duration,
      success: true,
      ...metadata,
    });

    // Track in metrics system
    metrics.histogram('speak_api_duration', duration, { operation });
    metrics.increment('speak_api_success', { operation });

    return result;
  } catch (error) {
    const duration = performance.now() - start;

    console.error({
      operation,
      duration,
      success: false,
      error,
      ...metadata,
    });

    metrics.histogram('speak_api_duration', duration, { operation });
    metrics.increment('speak_api_error', { operation });

    throw error;
  }
}

// Usage
const result = await measuredSpeakCall(
  'speech.recognize',
  () => speakClient.speech.recognize(audioBuffer),
  { audioSize: audioBuffer.byteLength }
);

Output

  • Reduced API latency
  • Audio preprocessing pipeline
  • Caching layer implemented
  • Request batching enabled
  • Connection pooling configured

Error Handling

Issue Cause Solution
Cache miss storm TTL expired Use stale-while-revalidate
Audio too large No compression Optimize audio format
Connection exhausted No pooling Configure max sockets
Memory pressure Cache too large Set max cache entries
Batch timeout Too many items Reduce batch size

Examples

Quick Performance Wrapper

const withPerformance = <T>(name: string, fn: () => Promise<T>) =>
  measuredSpeakCall(name, () =>
    cachedWithRedis(`cache:${name}`, fn, 300)
  );

Resources

Next Steps

For cost optimization, see speak-cost-tuning.

04 同仓库更多 Skills

Dicklesworthstone
Dicklesworthstone/pi_agent_rust

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/fonts that you can apply to any artifact that has been creating, or can generate a new theme on-the-fly.

视觉设计
★ 2.0k
Dicklesworthstone
Dicklesworthstone/pi_agent_rust

web-artifacts-builder

Suite of tools for creating elaborate, multi-component claude.ai HTML artifacts using modern frontend web technologies (React, Tailwind CSS, shadcn/ui). Use for complex artifacts requiring state management, routing, or shadcn/ui components - not for simple single-file HTML/JSX artifacts.

前端开发
★ 2.0k
Dicklesworthstone
Dicklesworthstone/pi_agent_rust

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.

测试与质量前端开发
★ 2.0k
Dicklesworthstone
Dicklesworthstone/pi_agent_rust

neuropixels-analysis

Neuropixels neural recording analysis. Load SpikeGLX/OpenEphys data, preprocess, motion correction, Kilosort4 spike sorting, quality metrics, Allen/IBL curation, AI-assisted visual analysis, for Neuropixels 1.0/2.0 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.

★ 2.0k
Dicklesworthstone
Dicklesworthstone/pi_agent_rust

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.

Skill与智能体开发
★ 2.0k
Dicklesworthstone
Dicklesworthstone/pi_agent_rust

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.

Skill与智能体开发
★ 2.0k