Transcript Scripts

The lyra-sdk repo includes runnable scripts in the scripts/ directory for testing transcript features. All transcript scripts run without a YouTube API key.

Prerequisites

No API key is needed. Just run:

npx tsx scripts/transcript-basic.ts

Basic Transcript

Fetches and prints all transcript lines with timestamps.

scripts/transcript-basic.ts

import { transcribeVideo, toPlainText } from '../packages/core/src/modules/transcript.js'

const lines = await transcribeVideo('dQw4w9WgXcQ')

for (const line of lines) {
  console.log(`[${line.offset.toFixed(1)}s] ${line.text}`)
}

console.log(toPlainText(lines))

Expected output:

Fetching transcript for: dQw4w9WgXcQ

Got 61 lines

[0.0s] ♪ We're no strangers to love ♪
[3.6s] ♪ You know the rules and so do I ♪
[6.0s] ♪ A full commitment's what I'm thinking of ♪
...

List Caption Languages

Discovers all available caption tracks for a video.

scripts/transcript-languages.ts

import { listCaptionTracks } from '../packages/core/src/modules/transcript.js'

const tracks = await listCaptionTracks('dQw4w9WgXcQ')

for (const track of tracks) {
  const auto = track.isAutoGenerated ? ' (auto-generated)' : ''
  console.log(`  ${track.languageCode} — ${track.languageName}${auto}`)
}

Expected output:

Fetching caption tracks for: dQw4w9WgXcQ

Found 2 caption track(s):

  en — English (auto-generated)
  es — Spanish

Transcript with Metadata

Fetches transcript lines plus video metadata, then outputs in SRT, VTT, and plain text formats.

scripts/transcript-meta.ts

import { transcribeVideo, toSRT, toVTT, toPlainText } from '../packages/core/src/modules/transcript.js'

const result = await transcribeVideo('dQw4w9WgXcQ', { includeMeta: true })

console.log(result.meta.title)   // Video title
console.log(result.meta.author)  // Channel name
console.log(result.meta.views)   // View count

// Output as SRT
console.log(toSRT(result.lines))

// Output as VTT
console.log(toVTT(result.lines))

// Plain text
console.log(toPlainText(result.lines))

TranscriptClient

Demonstrates the class-based API with shared configuration.

scripts/transcript-client.ts

import { TranscriptClient } from '../packages/core/src/modules/transcript.js'

const client = new TranscriptClient({ lang: 'en' })

const lines = await client.transcribe('dQw4w9WgXcQ')
const tracks = await client.availableTracks('dQw4w9WgXcQ')

Cache Benchmark

Compares no-cache vs InMemoryCache vs FsCache performance across multiple runs.

npx tsx scripts/transcript-cache-bench.ts

How it works

Run 1: No Cache (baseline)

Fetches the transcript 3 times with no caching. Every call makes 3 HTTP requests to YouTube (watch page, Innertube API, transcript XML).

Run 2: InMemoryCache

Creates a new InMemoryCache() and fetches 3 times. The first call is a cold fetch; runs 2 and 3 read from the in-process Map.

Run 3: FsCache

Creates a new FsCache() pointing to a temp directory and fetches 3 times. The first call is a cold fetch; runs 2 and 3 read from JSON files on disk.

Summary metrics

The script calculates averages, cache hit times, and speedup ratios for each strategy.

Sample results

=== Cache Benchmark: dQw4w9WgXcQ ===

--- No Cache ---
  Run 1: 1791ms  (cold — 3 HTTP requests)
  Run 2: 1567ms  (cold — 3 HTTP requests)
  Run 3: 1547ms  (cold — 3 HTTP requests)
  Avg:   1635ms

--- InMemoryCache ---
  Run 1: 1629ms  (cold — 3 HTTP requests)
  Run 2: 0.04ms  (cache HIT — 61 lines from cache)
  Run 3: 0.02ms  (cache HIT — 61 lines from cache)
  Avg:   543ms  |  Cache entries: 1

--- FsCache ---
  Run 1: 1704ms  (cold — 3 HTTP requests)
  Run 2: 0.37ms  (cache HIT — 61 lines from cache)
  Run 3: 0.20ms  (cache HIT — 61 lines from cache)
  Avg:   568ms

=== Summary ===
  No Cache avg:        1635ms
  InMemoryCache avg:   543ms  (3.0x faster overall)
  FsCache avg:         568ms  (2.9x faster overall)
  InMemoryCache hit:   ~0.03ms  (44561x faster than no-cache)
  FsCache hit:         ~0.27ms  (5560x faster than no-cache)

Strategy	Cold fetch	Cache hit	Overall avg
No Cache	~1600ms	N/A	~1635ms
InMemoryCache	~1600ms	~0.03ms	~543ms
FsCache	~1700ms	~0.27ms	~568ms

Retry Benchmark

Tests retry behavior with simulated failures and shows backoff timing.

npx tsx scripts/transcript-retry-bench.ts

=== Retry Benchmark ===

--- Test 1: No retries, success (baseline) ---
  Time:     1299ms
  Attempts: 1

--- Test 2: retries=3, retryDelay=100 — recovers after 1 failure ---
  0ms   → Attempt 1: 503 (retryable, waiting 100ms)
  1489ms → Attempt 2: 200 OK — 61 lines
  Time:     1489ms  (115% of baseline)
  Attempts: 2 (1 failed + 1 succeeded)
  Overhead: ~190ms from retry

--- Test 3: retries=2, retryDelay=200 — all requests fail ---
  50ms  → Attempt 1: 503 (retryable, waiting 200ms)
  300ms → Attempt 2: 503 (retryable, waiting 400ms)
  750ms → Attempt 3: 503 (retries exhausted)
  Time:     604ms
  Attempts: 3 (all failed)
  Backoff:  200ms, 400ms

--- Test 4: retries=5, retryDelay=100 — aborted after ~150ms ---
  Aborted:  true
  Time:     151ms (stopped early by AbortSignal)

Test	Retries	Outcome	Total time	Backoff delays
Baseline	0	Success	~1300ms	—
Recovery	3	Success (retry)	~1490ms	100ms
Exhausted	2	All fail	~604ms	200ms, 400ms
Abort	5	Cancelled	~151ms	stopped early

Prerequisites

Basic Transcript

List Caption Languages

Transcript with Metadata

TranscriptClient

Cache Benchmark

How it works

Run 1: No Cache (baseline)

Run 2: InMemoryCache

Run 3: FsCache

Summary metrics

Sample results

Retry Benchmark

How it works

Test 1: Baseline (no retries)

Test 2: Recovery after failure

Test 3: Exhausted retries

Test 4: AbortSignal cancellation

Sample results

On this page