Assemblyai Rate Limits

AssemblyAI Rate Limits

Overview

Handle AssemblyAI rate limits with exponential backoff, queue-based throttling, and concurrency management. AssemblyAI auto-scales limits for paid users.

Prerequisites

• assemblyai package installed
• Understanding of async/await patterns

Rate Limit Tiers (Actual)

Async Transcription API

Streaming (WebSocket)

LeMUR

Note: AssemblyAI auto-scales paid limits. At 70%+ utilization, the new session rate limit increases by 10% every 60 seconds with no ceiling cap.

Instructions

Step 1: Exponential Backoff with Jitter

import { AssemblyAI, type Transcript } from 'assemblyai';

const client = new AssemblyAI({
  apiKey: process.env.ASSEMBLYAI_API_KEY!,
});

async function transcribeWithBackoff(
  audioUrl: string,
  options: Record<string, any> = {},
  config = { maxRetries: 5, baseDelayMs: 1000, maxDelayMs: 30000 }
): Promise<Transcript> {
  for (let attempt = 0; attempt <= config.maxRetries; attempt++) {
    try {
      return await client.transcripts.transcribe({
        audio: audioUrl,
        ...options,
      });
    } catch (err: any) {
      if (attempt === config.maxRetries) throw err;

      const status = err.status ?? err.statusCode;
      // Only retry on 429 (rate limit) and 5xx (server errors)
      if (status && status !== 429 && (status < 500 || status >= 600)) throw err;

      const exponentialDelay = config.baseDelayMs * Math.pow(2, attempt);
      const jitter = Math.random() * config.baseDelayMs;
      const delay = Math.min(exponentialDelay + jitter, config.maxDelayMs);

      console.warn(`[${attempt + 1}/${config.maxRetries}] Retrying in ${delay.toFixed(0)}ms...`);
      await new Promise(r => setTimeout(r, delay));
    }
  }
  throw new Error('Unreachable');
}

Step 2: Queue-Based Concurrency Control

import PQueue from 'p-queue';

// Limit to N concurrent transcription jobs
const transcriptionQueue = new PQueue({
  concurrency: 5,           // Max 5 concurrent jobs
  interval: 60_000,         // Per minute window
  intervalCap: 50,           // Max 50 new jobs per minute
});

async function queuedTranscribe(audioUrl: string): Promise<Transcript> {
  return transcriptionQueue.add(() =>
    transcribeWithBackoff(audioUrl)
  );
}

// Process a batch of files
const audioUrls = [
  'https://example.com/audio1.mp3',
  'https://example.com/audio2.mp3',
  'https://example.com/audio3.mp3',
];

const results = await Promise.all(
  audioUrls.map(url => queuedTranscribe(url))
);

console.log(`Completed ${results.length} transcriptions`);
console.log(`Queue size: ${transcriptionQueue.size}, pending: ${transcriptionQueue.pending}`);

Step 3: Batch Processing with Progress

async function batchTranscribe(
  audioUrls: string[],
  onProgress?: (completed: number, total: number) => void
): Promise<Transcript[]> {
  const queue = new PQueue({ concurrency: 5 });
  const results: Transcript[] = [];
  let completed = 0;

  const promises = audioUrls.map(url =>
    queue.add(async () => {
      const transcript = await transcribeWithBackoff(url);
      completed++;
      onProgress?.(completed, audioUrls.length);
      return transcript;
    })
  );

  return Promise.all(promises);
}

// Usage
await batchTranscribe(
  urls,
  (done, total) => console.log(`Progress: ${done}/${total}`)
);

Step 4: Streaming Rate Limit Handling

async function connectStreamingWithRetry(maxRetries = 3) {
  for (let attempt = 0; attempt <= maxRetries; attempt++) {
    try {
      const transcriber = client.streaming.createService({
        speech_model: 'nova-3',
        sample_rate: 16000,
      });

      transcriber.on('error', (error) => {
        console.error('Streaming error:', error);
      });

      await transcriber.connect();
      return transcriber;
    } catch (err: any) {
      if (attempt === maxRetries) throw err;

      // WebSocket code 4008 = session limit
      const delay = Math.pow(2, attempt) * 2000;
      console.warn(`Stream connect failed. Retrying in ${delay}ms...`);
      await new Promise(r => setTimeout(r, delay));
    }
  }
}

Output

• Automatic retry with exponential backoff and jitter
• Queue-based concurrency control with p-queue
• Batch transcription with progress reporting
• Streaming reconnection logic

Examples

Treat displayed limits as configuration inputs that are verified against the live account and provider documentation before rollout. Begin with a low concurrency ceiling, honor server-provided retry guidance, persist idempotency keys, and alert on sustained throttling rather than retrying a workload indefinitely.

Error Handling

Resources

• AssemblyAI Rate Limits
• p-queue Documentation
• AssemblyAI Streaming Limits

Next Steps

For security configuration, see assemblyai-security-basics.