Split View: 영어 스피킹 쉐도잉 시스템 설계: 90일 루틴과 측정 지표

English Speaking Shadowing System Design: 90-Day Routine and Measurement Metrics

• Scientific Basis for Why Shadowing Is Effective for Speaking
• The 4-Stage Shadowing Progression Model
  • Stage 1: Mumbling - Weeks 1-2
  • Stage 2: Synchronized Shadowing - Weeks 3-4
  • Stage 3: Prosody Shadowing - Weeks 5-8
  • Stage 4: Content Shadowing - Weeks 9-12
• 90-Day Training Program Details
  • Material Selection Criteria
  • Weekly Training Plan
  • Actual Daily 15-Minute Routine (Phase 3 Example)
• Measurement Metrics System
  • 4 Core Metrics
  • Automated Pronunciation Accuracy Measurement Script
  • Weekly Log Template
• Plateau Breakthrough Strategies
  • Plateau Signals and Countermeasures
  • Minimal Pairs Focused Training
• Tool Comparison: Shadowing Apps and Platforms
• Shadowing vs Other Speaking Training Methods
• Real-World Scenario: Office Worker A's 90 Days
• Quiz
• References

Scientific Basis for Why Shadowing Is Effective for Speaking

Shadowing is a training method where you listen to a native speaker's audio and repeat it with a 0.5-1 second delay. It was systematized by Professor Kadota Shuhei, a Japanese interpretation training expert, in second language acquisition research and is currently used as a core training method in interpreter training programs worldwide.

The reasons why shadowing is effective can be found in neuroscience.

Working Memory Training: Processing listening and speaking simultaneously strengthens the phonological loop. This is the core mechanism explained in Baddeley's working memory model.

Motor Learning: Repeatedly moving the mouth, tongue, and vocal cords in the same pattern forms muscle memory. It works on the same principle as piano practice. Initially done consciously, it becomes automated through repetition.

Prosody Internalization: English rhythm, stress, and intonation patterns cannot be learned word by word. They are naturally internalized by repeating entire sentences as a whole.

According to Stephen Krashen's Input Hypothesis, "comprehensible input" is the key to language acquisition. Shadowing adds "forced output" to this, bridging the gap between input and output.

The 4-Stage Shadowing Progression Model

Shadowing requires only 15 minutes per day, but skipping stages reduces its effectiveness by half. Follow these 4 stages in order.

Stage 1: Mumbling - Weeks 1-2

Goal: Getting accustomed to the rhythm and flow of speech

Method:

1. Listen to the audio and move your lips, following along in a quiet voice
2. You don't need to say every word accurately. The goal is to ride the rhythm
3. Don't look at the script

What to measure at this stage: The proportion of 15 daily minutes where you "followed along without stopping." 30-40% is normal at first.

Stage 2: Synchronized Shadowing - Weeks 3-4

Goal: Speaking simultaneously with the native speaker

Method:

1. Repeat the same audio at least 5 times before starting to shadow
2. Speak as simultaneously as possible with the native speaker's utterance (within 0.5 second delay)
3. Check with the script while shadowing

What to measure at this stage: Synchronization rate (proportion of words spoken simultaneously with the native speaker). Move to the next stage when you reach 60% or higher.

Stage 3: Prosody Shadowing - Weeks 5-8

Goal: Accurately reproducing stress, rhythm, and intonation

Method:

1. Shadow while recording yourself
2. Compare the original with your own recording
3. Focus particularly on stress placement, linking, and reduction

Key Point - English pronunciation patterns that Korean speakers often miss:

Stage 4: Content Shadowing - Weeks 9-12

Goal: Shadowing naturally while understanding the content

Method:

1. Shadow without a script while simultaneously grasping the content
2. Immediately after shadowing, summarize what you heard in your own words (Retelling)
3. Measure your first-attempt shadowing success rate with new material

90-Day Training Program Details

Material Selection Criteria

Material selection is half the battle. Choosing the wrong material wastes 90 days.

Characteristics of good material:

• Speed of 120-150 words per minute (for beginner-intermediate level)
• 2-5 minutes per session (too long causes loss of focus)
• Content with available scripts (subtitles)
• Topics related to your interests or work

Recommended material (by difficulty):

Weekly Training Plan

# 90-Day Shadowing Program Configuration
program:
  name: '90-Day Shadowing System'
  daily_minutes: 15
  phases:
    - name: 'Foundation'
      weeks: 1-2
      stage: 'Mumbling'
      material: 'VOA Learning English'
      daily_routine:
        - listen_without_speaking: 3min
        - mumbling_practice: 10min
        - vocabulary_review: 2min
      target: 'Rhythm following rate 40% or above'

    - name: 'Sync'
      weeks: 3-4
      stage: 'Synchronized'
      material: 'TED-Ed'
      daily_routine:
        - full_speed_listening: 2min
        - synchronized_shadowing: 10min
        - script_check: 3min
      target: 'Synchronization rate 60% or above'

    - name: 'Prosody'
      weeks: 5-8
      stage: 'Prosody Shadowing'
      material: 'TED Talks'
      daily_routine:
        - shadowing_with_recording: 10min
        - compare_with_original: 3min
        - note_problem_sounds: 2min
      target: 'Intonation similarity self-assessment 7/10 or above'

    - name: 'Content'
      weeks: 9-12
      stage: 'Content Shadowing'
      material: 'NPR / BBC'
      daily_routine:
        - blind_shadowing: 8min
        - retelling_in_own_words: 5min
        - new_vocabulary_log: 2min
      target: 'First-attempt shadowing success rate 70% or above'

Actual Daily 15-Minute Routine (Phase 3 Example)

[0:00-0:30]  Play today's material - Listen through once (grasp key points)
[0:30-2:00]  1st shadowing - Start recording
[2:00-4:00]  Play recording + Compare with original - Note 3 differences
[4:00-8:00]  Focused repetition on problem sections (at least 5 times)
[8:00-12:00] 2nd shadowing - Record full pass
[12:00-14:00] Compare 2nd recording vs original - Confirm improvements
[14:00-15:00] Note 3 expressions learned today (add to vocabulary app)

Measurement Metrics System

If you evaluate by "feeling," you won't know if your skills have improved. Record the following 4 metrics weekly.

4 Core Metrics

Automated Pronunciation Accuracy Measurement Script

You can automatically measure the recognition rate of shadowing recordings using the Google Cloud Speech-to-Text API.

import difflib
from google.cloud import speech_v1 as speech

def measure_accuracy(audio_path: str, expected_text: str) -> dict:
    """Measures pronunciation accuracy of a shadowing recording file."""
    client = speech.SpeechClient()

    with open(audio_path, "rb") as f:
        audio = speech.RecognitionAudio(content=f.read())

    config = speech.RecognitionConfig(
        encoding=speech.RecognitionConfig.AudioEncoding.LINEAR16,
        sample_rate_hertz=16000,
        language_code="en-US",
        enable_word_time_offsets=True,
    )

    response = client.recognize(config=config, audio=audio)

    recognized = " ".join(
        result.alternatives[0].transcript
        for result in response.results
    ).lower().split()

    expected = expected_text.lower().split()

    # Calculate word-level match rate
    matcher = difflib.SequenceMatcher(None, expected, recognized)
    accuracy = matcher.ratio() * 100

    # Extract list of missed words
    missed_words = []
    for tag, i1, i2, j1, j2 in matcher.get_opcodes():
        if tag in ("replace", "delete"):
            missed_words.extend(expected[i1:i2])

    return {
        "accuracy_pct": round(accuracy, 1),
        "total_expected": len(expected),
        "total_recognized": len(recognized),
        "missed_words": missed_words[:10],  # Top 10 only
    }

# Usage example
result = measure_accuracy(
    "recordings/2026-03-04-ted-talk.wav",
    "The greatest glory in living lies not in never falling "
    "but in rising every time we fall"
)
print(f"Accuracy: {result['accuracy_pct']}%")
print(f"Missed words: {result['missed_words']}")

Weekly Log Template

import json
from datetime import date, timedelta

def weekly_log(week_num: int, data: dict) -> dict:
    """Records weekly shadowing performance and compares with previous week."""
    log = {
        "week": week_num,
        "period": f"{date.today() - timedelta(days=6)} ~ {date.today()}",
        "metrics": {
            "wpm": data.get("wpm", 0),
            "sync_rate_pct": data.get("sync_rate", 0),
            "pronunciation_accuracy_pct": data.get("accuracy", 0),
            "retelling_score": data.get("retelling", 0),
        },
        "practice_days": data.get("days_practiced", 0),
        "total_minutes": data.get("total_minutes", 0),
        "material_used": data.get("materials", []),
        "problem_sounds": data.get("problem_sounds", []),
        "notes": data.get("notes", ""),
    }

    # Save cumulatively to JSON file
    log_file = "shadowing_progress.json"
    try:
        with open(log_file, "r") as f:
            all_logs = json.load(f)
    except FileNotFoundError:
        all_logs = []

    all_logs.append(log)

    with open(log_file, "w") as f:
        json.dump(all_logs, f, indent=2, ensure_ascii=False)

    return log

# Usage example
weekly_log(5, {
    "wpm": 105,
    "sync_rate": 65,
    "accuracy": 72,
    "retelling": 5,
    "days_practiced": 6,
    "total_minutes": 90,
    "materials": ["TED: The power of vulnerability"],
    "problem_sounds": ["th sound", "r/l distinction", "word-final consonants"],
    "notes": "Linking has improved a lot, but th pronunciation is still difficult",
})

Plateau Breakthrough Strategies

After 4-6 weeks of shadowing, almost all learners experience a plateau. Don't quit at this point. A plateau is the process of the brain integrating new patterns.

Plateau Signals and Countermeasures

Minimal Pairs Focused Training

Practice the phoneme pairs that Korean speakers find most difficult for an additional 5 minutes daily.

R vs L:

• right / light
• road / load
• correct / collect
• crowd / cloud
• Sentence: "The right light was really lovely."

B vs V:

• base / vase
• boat / vote
• berry / very
• best / vest
• Sentence: "The best vote came from a very brave person."

F vs P:

• fan / pan
• feel / peel
• coffee / copy
• fast / past
• Sentence: "I feel like peeling a fresh peach."

TH vs S/D:

• think / sink
• three / tree (or free)
• that / dat (Korean-style)
• bath / bass
• Sentence: "I think three thousand is the right number."

Tool Comparison: Shadowing Apps and Platforms

Recommended combination (cost minimized):

1. YouTube (material) + Smartphone recorder (recording) + Google Docs voice typing (recognition rate check)
2. This combination allows you to perform all stages at zero cost

Shadowing vs Other Speaking Training Methods

Best combination: 10 min shadowing + 5 min Retelling (daily) + 30 min Free Talking (twice weekly)

Real-World Scenario: Office Worker A's 90 Days

Profile: 32 years old, IT company PM, TOEIC 780, can listen in English meetings but can't speak

Week 1-2: Started with VOA Learning English. Day 1 synchronization rate 25%. Lost the rhythm most of the time. Shocked by "My mouth doesn't move as much as I expected."

Week 3-4: Switched material to TED-Ed. 15 minutes every morning before work. Synchronization rate rose to 55%. Started recognizing linking patterns in particular (pick it up -> pi-ki-dup).

Week 5-6: Listened to recordings for the first time. Response: "This is my English?" Korean intonation carried over completely. Started focusing on th pronunciation and r/l distinction.

Week 7-8: Plateau. WPM stuck at 110 for 3 weeks. Switched to 1.2x speed TED Talk. Painful for 2 days, but returning to normal speed felt much more comfortable.

Week 9-10: Switched material to NPR Planet Money. Difficult at first due to many economics terms, but work-related vocabulary expanded. Starting Retelling enabled beginning utterances in English meetings with "I think the main point is..."

Week 11-12: First-attempt shadowing success rate 72%. WPM 128. Successfully spoke 3 consecutive sentences in an English meeting. Received feedback from a colleague: "Your English has really improved."

90-Day Final Results:

• WPM: 85 -> 128 (50% improvement)
• Meeting utterance frequency: 0-1 times -> 4-5 times/meeting
• Google Speech-to-Text recognition rate: 58% -> 84%

Quiz

References

• Kadota Shuhei, "Shadowing, Ondoku to Eigo Shutoku no Kagaku" (The Science of Shadowing, Oral Reading, and English Acquisition), Cosmopier, 2015
• Baddeley, A., "Working Memory: Theories, Models, and Controversies," Annual Review of Psychology, 2012
• Krashen, S., "The Input Hypothesis: Issues and Implications," Longman, 1985
• VOA Learning English: https://learningenglish.voanews.com/
• TED Talks: https://www.ted.com/talks
• ELSA Speak: https://elsaspeak.com/
• BBC Learning English - 6 Minute English: https://www.bbc.co.uk/learningenglish/english/features/6-minute-english