Shadowing & self-review method¶
At a glance
Priority: P0 · Complexity: ⅖ · Est. time: 2 h · Phase: 1 · Prereqs: Pronunciation: word stress, rhythm & intonation You're done when: you have a repeatable 20-minute weekly routine (shadow, record, score, log), you have scored at least four recordings with the rubric, and your log shows a trend on at least two dimensions.
Used every Saturday in the drills (Week 1 onwards) and as the focus of Week 6, Week 12 and Week 18.
Why it matters¶
Fluency improves through deliberate practice with feedback, not through hours of exposure. Two techniques close the loop cheaply:
- Shadowing trains the motor and rhythm side: you speak along with a model, copying stress, pace, intonation and linking. It builds the automaticity that frees attention during real speech.
- Self-review trains the monitoring side: you record, listen, score and log. Without it you practise your existing habits harder.
Research on shadowing in second-language learning (largely from Japan, Hamada and others) reports gains in listening and prosody; evidence for large effects on spontaneous speaking is more modest. Treat it as one tool for rhythm and chunking, not a complete method. Combine it with your own production practice (recorded talks), which is where fluency in real work is built.
Core concepts¶
1. What shadowing is (and is not)¶
Shadowing: listen to a short audio clip and repeat it aloud almost simultaneously, a fraction of a second behind, imitating everything: words, stress, pitch, speed, pauses. Related techniques:
| Technique | What you do | Best for |
|---|---|---|
| Listen and repeat | Hear a phrase, pause, repeat it | Beginners, tricky sounds |
| Chunk shadowing | Repeat after each sentence or phrase | Word stress, phrase melody |
| Full (simultaneous) shadowing | Speak with the audio, slightly behind | Rhythm, pace, linking |
| Prosody shadowing | Focus only on pitch and stress; ignore meaning | Intonation |
| Content shadowing | Focus on understanding while shadowing | Listening and chunking |
| Read-and-shadow with transcript | Read the transcript first, then shadow | First pass for hard clips |
| Echo (delayed) repeat | Speak 3-5 s after, from memory | Working memory, summarising |
Not shadowing: copying an accent to hide yours. The purpose is to internalise stress-timed rhythm, natural chunking and the intonation patterns that listeners expect, while remaining yourself.
2. Choosing material¶
Good shadowing clips are:
- 30-90 seconds long, single speaker, clear audio.
- Relevant: conference talks, engineering podcasts, product demos, interviews, earnings-call excerpts.
- A little above your current level in pace (about 140-160 wpm) but not so fast that you can only mumble.
- With a transcript (TED, BBC Learning English and many conference talks provide one).
- Varied accents: at least one US, one UK and one non-native professional speaker, so you learn to understand them all. Your goal is comprehension range, not imitation of one accent.
Sources: TED talks (transcripts available), BBC Learning English (6 Minute English, The English We Speak), conference talks (InfoQ, QCon, Strange Loop, Google I/O) with captions, the "All-In"-style podcasts less so (too fast and overlapping). Use YouGlish to find a specific phrase in context.
3. The 20-minute shadowing routine¶
| Minutes | Step | Details |
|---|---|---|
| 0-3 | Listen | Play the clip twice without speaking; note stressed words |
| 3-6 | Transcript pass | Read the transcript; underline stressed words and mark pauses (/) and pitch falls/rises |
| 6-10 | Chunk shadowing | Repeat phrase by phrase at 100%, then at 85% speed if the player allows |
| 10-14 | Full shadowing x3 | Speak along; first at 85%, then 100% |
| 14-17 | Record | Shadow the clip without headphones bleeding, record your voice and the model separately or in turn |
| 17-20 | Compare and log | Listen to yours against the model; note 2 things: one that improved, one to work on |
Rules:
- Use headphones so you hear the model while hearing yourself only through bone conduction.
- Keep one focus per session (word stress, or pause length, or pitch).
- Repeat the same clip for 3-5 days before moving on: the gain is in the repetition.
- Stop if you are only mumbling: slow the audio.
4. Self-review: the record-score-log loop¶
Record. Weekly, 2 minutes, on a work topic (Saturday drill). Same conditions: same mic, same time of day. Use your phone or laptop voice recorder; save with the date.
Transcribe. Use auto-transcription (Otter, Zoom, Teams, Whisper-based tools) to get words, wpm and a rough filler count. Remember that auto-transcripts may clean fillers and repair grammar, so listen too.
Score with the rubric. Score 1-4 on six dimensions. Be a strict marker.
| Dimension | 1 | 2 | 3 | 4 |
|---|---|---|---|---|
| Clarity | Listener must guess key points | Message present but buried | Clear message, minor detours | Message immediate and memorable |
| Structure | No order | Order, but no signposts | Answer-first with signposts | Pyramid/PREP, crisp close |
| Fluency | Frequent stalls, restarts | Some stalls, fillers >8/min | Fillers 4-8/min, mostly smooth | Fillers <4/min, purposeful pauses |
| Grammar accuracy | Errors impede meaning | Frequent recurring errors | Few errors, none impede | Rare errors; complex structures accurate |
| Vocabulary range | Repetitive, vague | Adequate but generic | Precise, some variety | Precise, varied, natural collocations |
| Pronunciation | Stress/sounds often impede | Stress errors on key words | Mostly clear; a few stress slips | Clear stress and rhythm; easy to follow |
Diagnose with a second listen. Listen once for content and once only for one dimension (for example, stress on technical words). Use the words-per-minute and filler count as data.
Log. One line per recording in docs/log/: date, prompt, length, wpm, fillers/min, six scores, one win, one target for next week, and the top three recurring errors. Trend beats snapshot: compare with your Week 0 baseline and the checkpoint weeks.
5. Getting feedback from others¶
- A trusted colleague. Send the recording (or a 60-second clip) and ask a specific question: "Which words did you have to guess?" and "Where did I lose you?". Narrow questions get honest answers.
- Native or fluent speaker exchange or tutor. One session every 2-4 weeks focused on a recording you send in advance.
- Toastmasters evaluations. A structured, kind, specific format.
- AI tools. Transcription tools reveal what a recogniser hears; if a good ASR mis-hears a word, a human listener may too. Use as a hint, not a judge.
6. Error log: turning mistakes into a syllabus¶
Every recording, write the top three errors: e.g., "dropped -ed on deployed", "stress on ar-chi-TEC-ture", "filler 'so basically' at starts". Weekly, promote the most frequent error to a daily 2-minute drill. Retire it when it does not appear in three consecutive recordings. This log feeds the weeks labelled "personal error log" in the plan (weeks 16, 18, 23).
Resources¶
| Resource | Type | Why this one | Level | Cost |
|---|---|---|---|---|
| TED | video | Clear speakers, downloadable transcripts, and varied accents; ideal 3-minute shadowing clips | intermediate | free |
| BBC Learning English | audio | 6 Minute English has transcripts and vocabulary; The English We Speak has 3-minute idiom clips | intermediate | free |
| YouGlish | interactive | Find exactly the phrase you struggle with, in real speech with captions | all | free |
Praat  |
tool | Free; overlay your pitch contour with the model's to see stress and intonation differences | advanced | free |
| Rachel's English | video | Her "sentence shadowing" style videos show mouth and stress; good for prosody | intermediate | free/freemium |
| Otter | tool | Automatic transcription with words-per-minute; quick self-review data | all | freemium |
| Toastmasters International | community | Human evaluators with a structured feedback format | all | freemium |
| Conference talks with captions (InfoQ, Google I/O, QCon on YouTube) | video | Real engineering talks; content and vocabulary match your work | advanced | free |
| Forvo | reference | Native-speaker pronunciations of single words and names (useful for colleagues' names) | all | free |
Hands-on lab (75 min, then weekly)¶
- Pick a clip (10 min). A 60-second segment of an engineering talk with a transcript (TED or a conference video). Mark stressed words on the transcript.
- Shadow (20 min). Run the 20-minute routine above.
- Record and compare (10 min). Record your shadowing of the clip; listen against the model. Note two differences.
- Free-speech recording (15 min). 2-minute talk: "Explain a technical decision you made last quarter." Get the transcript, compute wpm and fillers per minute.
- Score (10 min). Use the rubric. Be strict. Write the log line.
- Plan (10 min). Choose next week's clip and one focus area.
Expected output: a log file with an entry, a chosen clip for the week, and a personal error list (top three).
Questions¶
L1 — Recall¶
Q1. What is the difference between chunk shadowing and full shadowing?
Answer
Chunk shadowing: repeat after each short phrase (with pauses). Full shadowing: speak simultaneously, slightly behind the audio. Chunk is easier and good for stress and melody; full builds rhythm and speed.
Q2. Name the six rubric dimensions used for self-review.
Answer
Clarity, structure, fluency, grammar accuracy, vocabulary range, pronunciation.
Q3. Why use the same clip for several days?
Answer
Repetition builds automaticity; the first pass is about comprehension, later passes about prosody and speed. Switching clips daily gives exposure but little improvement in production.
Q4. What should be in one line of your practice log?
Answer
Date, prompt, length, wpm, fillers/minute, six scores, one win, one target, top three errors.
L2 — Apply¶
Q5. Your transcript shows 300 words in 2 minutes and 14 fillers. Compute wpm and filler rate, and interpret.
Answer
150 wpm; 7 fillers/min. Pace is fine for a presentation; fillers are in the "score 3 to 2" band (4-8 is 3 by the rubric, so a 3, at the high end). Focus: replace fillers with pauses at sentence starts.
Q6. Choose a shadowing clip for weak word stress on technical terms. Which properties should it have?
Answer
A 60-second engineering talk by a clear speaker with a transcript, containing your target vocabulary (architecture, latency, deploy...), at about 140 wpm; check stress in a dictionary or YouGlish. Focus mode: prosody shadowing with the transcript stress-marked.
Q7. Write two specific questions to ask a colleague when you send a recording.
Answer
"Which words or phrases did you have to work to understand?" and "At which point in the 2 minutes did you lose the main point?" Specific, answerable, and low social cost.
Q8. Your scores: Clarity 3, Structure 2, Fluency 3, Grammar 3, Vocabulary 3, Pronunciation 3. What do you work on next week and why?
Answer
Structure: the lowest score, and it also raises clarity and fluency. Use PREP on all recordings, then re-score. Work on one bottleneck at a time.
L3 — Judge and choose¶
Q9. You have 20 minutes daily. Option A: shadow a TED talk. Option B: record yourself on a work topic and review. Which yields more for the goal "sound clear on calls"?
Answer
Both, in a ratio: shadowing (A) for rhythm and prosody, recording (B) for your real production issues. If forced to choose, B, because it targets your own errors and structure; A is a supplement. A weekly cycle: three shadowing days, two recording days, one review day.
Q10. An AI transcription tool transcribes your word "latency" as "lately". Is that evidence of a pronunciation problem?
Answer
A signal, not proof. It may reflect stress, a shortened vowel or poor audio. Test by listening to the audio, checking context (would a human confuse it?) and re-recording with a carrier phrase ("the latency budget"). If the recogniser fails repeatedly on a key word, treat it as a real clarity risk.
Q11. Is it useful to shadow a speaker whose accent differs from yours?
Answer
Yes for listening range and rhythm, but for pronunciation targets choose speakers who are clear and professional, and keep the goal as intelligibility. You need not adopt their accent. Mixing US, UK and non-native professional speakers builds comprehension of the accents you meet at work.
L4 — Staff-level¶
Q12. Your team wants to run a weekly "speaking club" to improve communication. How do you design it so people actually improve and no one is embarrassed?
Answer
Voluntary, 30 minutes, small groups, psychological safety rules (feedback only on requested dimensions, SBI format). Structure: 2 minutes of speaking each on a work topic, one strength and one target from the listener. Record privately; sharing is optional. Rotate roles. Measure through self-reported rubric scores, not comparisons. Sponsors from leadership should participate to model vulnerability.
Q13. A colleague sends you a recording and says "be honest". The main issue is a monotone delivery and heavy fillers. Give feedback.
Answer
Use SBI and a positive-first but honest structure: "In the second minute (Situation), the pitch stayed level and there were about ten 'so' openers (Behaviour); as a listener I lost which point was the main one (Impact). Try: pause instead of 'so', and raise pitch on the key number. The structure of your three points was clear, which is strong." Offer to re-listen after a second take.
Q14. After 8 weeks your scores plateau at 3 on all dimensions. What do you change?
Answer
Plateaus mean the routine has become comfortable. Increase difficulty: unfamiliar topics, time limits, an audience (a real meeting), less preparation, feedback from a coach. Target a dimension where you are 3 and require a 4 criterion (for example, fillers under 4/min). Shift from practice to real-world stretch: volunteer to present in an all-hands.
Real-world use cases¶
- Preparing a keynote or all-hands: shadow the opening sentences of admired speakers, record your opening, refine.
- New role with a different region: shadow local podcasts to acclimatise to the accent and idioms in your regional team.
- Interview prep: record STAR answers and score with the rubric.
- Onboarding a new hire with English as an additional language: a shared shadowing clip and review routine.
Pitfalls & anti-patterns¶
- Shadowing only, never recording original speech.
- Shadowing too-fast material and mumbling.
- Judging yourself only by feel; no numbers.
- Trying to fix six dimensions at once.
- Perfectionism: re-recording 15 times instead of logging and moving on.
- Copying a speaker's accent instead of their pausing and structure.
- Practising with the same clip forever; add a new one every week.
Checklist¶
- I have a 20-minute routine on my calendar.
- I scored at least four recordings with the rubric and logged them.
- I have an error log and a top-three list.
- I asked a colleague two specific questions about a recording.
- I compare each score with my Week 0 baseline.
- I answered all L3 questions out loud in < 3 min each.