How to Record the Model and Your Shadowing on Separate Tracks
Record model and shadowing on separate Audacity tracks, fix latency first, then compare timing, stress and omissions with a three-listen A/B method.
Put the permitted model audio on Track 1, record your microphone on Track 2 while listening through headphones, correct any global recording latency, then compare MODEL ONLY, SHADOW ONLY and BOTH.
If your model and your voice are baked into one waveform, every mistake arrives wearing a disguise. Was that late entry you—or the computer? Did you omit a word—or did the model simply mask it? Two tracks turn the recording from soup into evidence.
The current Audacity 4 workflow in one minute
Audacity 4.0.0 was released on September 3, 2026, with a rebuilt interface and recording flow, so older Audacity 3 tutorials may show controls in different places. Check the current Audacity changelog if your screen does not match an old video tutorial.
- Import the model audio you are allowed to use so it sits on its own track.
- Put on headphones so the model does not leak back into your microphone.
- Select your microphone and use Audacity's option to record onto a new track while listening to the model.
- Before judging your speaking timing, check whether the whole learner track has been shifted by recording latency.
That is the technical setup. The learning method comes later: two lanes, three listens.
Step 1: put the model on Track 1
The cleanest cross-platform route is to use an audio file you are permitted to record, copy or practise with. In Audacity 4, Import adds an existing audio file to the open project as a new track while leaving the original file unchanged.
Name this track MODEL. Keep the clip short enough that you can actually diagnose it. You are building a pronunciation and timing lab, not archiving a feature film.
If your model comes from a supported video page, FunFluen can help before you open Audacity: use sentence navigation and repeat controls to identify one useful line and understand it well enough to practise. Review FunFluen for choosing and repeating a line before you record. The first action is reviewing the extension listing before installing. FunFluen is not the multitrack recorder in this workflow; Audacity handles the two-track recording.
If you need to capture desktop audio instead
Desktop-audio capture is operating-system dependent. Audacity's current desktop-audio guidance documents WASAPI loopback on Windows and monitor-source routing on Linux, while macOS does not provide the same built-in route and needs additional audio routing.
Use that only for material you are permitted to record. Technical ability to capture playback does not give you permission to bypass DRM or copy protected content. If you already have an authorized model file, importing it is simpler and cleaner.
Step 2: record your SHADOW on Track 2
Now give your own voice a separate lane.
First, use headphones. Audacity's current microphone setup guidance recommends headphones so playback does not get recorded again through the microphone. If you use speakers, your SHADOW track may quietly contain the model too, which ruins the clean comparison.
For one microphone, a mono voice track is normally appropriate. Select the microphone you actually intend to use, then choose Audacity's Record on new track workflow. Audacity creates a new track from the selected input while leaving the existing MODEL track untouched.
Name the new track SHADOW. Start the model, shadow it as you normally would, and let Audacity record only your microphone onto that second track.
Do not worry yet about making the waveforms look identical. Your first goal is much simpler: two clean, independently controllable recordings.
Before you judge your timing: TECH LATE or YOU LATE?
Recording alongside an existing track can introduce technical latency: a delay between when you speak and where the recorded audio lands on the timeline. Audacity 4 includes latency compensation and buffer settings for recording alongside existing material.
This matters because an uncorrected technical delay can make perfectly reasonable shadowing look late. Fix the measuring tool before measuring yourself.
The whole SHADOW track is late by roughly the same amount from beginning to end.
TECH LATE is the first suspect. Check Audacity's latency compensation and your recording setup before judging your language timing. A uniform whole-track offset is different from one particular word entering late.
The tracks begin close together, but the offset grows as the recording continues.
Treat this as a recording/alignment problem first. Check your audio devices and Audacity project/settings before using that recording to judge language timing.
The tracks are globally aligned, but I enter late on the same phrase every time.
YOU LATE becomes a reasonable candidate. Now the recording is useful for language diagnosis. Listen to what happens immediately before that phrase: did you miss a reduction, wait to understand, or need too much planning time?
I can hear a faint copy of the model inside the SHADOW track.
That is bleed, not useful comparison data. Use headphones, lower environmental playback, and record the learner track again.
This TECH LATE / YOU LATE test is a FunFluen practical diagnostic, not an Audacity feature and not an automatic pronunciation score.
The Two-Track A/B Lab: three listens, three different questions
Audacity's per-track Mute and Solo controls let you change what you hear without altering the recorded audio. Use that to run three different listening passes.
| Listen | What to hear | Do not conclude |
|---|---|---|
| MODEL ONLY | Where is the main stress? Where does the phrase group? Which words reduce or link? | Do not decide what you did wrong yet. You are mapping the target. |
| SHADOW ONLY | Which words vanished? Is your stress in the same information area? Do endings weaken or disappear? | Do not compare voice quality or waveform height as if similarity meant correctness. |
| BOTH | After latency is corrected, where do you enter? Where do you separate from the model? Can you recover after a miss? | Do not try to repair five things at once. |
British Council pronunciation guidance similarly recommends listening to a model, repeating, recording yourself and listening back to compare features such as word stress, sentence stress, rhythm, intonation and linking. See the British Council recording guidance.
After the three listens, write down one repair for the next take. One. Your recording is evidence, not a courtroom indictment of your accent.
A waveform is not an accent polygraph
The visual tracks can help with a few coarse questions:
- Did your voice begin much later than expected?
- Did you leave a huge pause where the model continued?
- Did an entire word or phrase disappear?
- Is your learner recording globally shifted relative to the model?
But a waveform alone cannot tell you whether a vowel is intelligible, whether your accent is “good,” whether a stress choice sounds natural, or whether a listener would understand you. Different microphones, gain levels and voices produce different-looking waveforms.
Use your eyes for gross timing clues. Use your ears for the language.
English production check: copy, then create
Use this invented model sentence:
I made a mistake, but I fixed it before the meeting.
Shadow it once, record it on SHADOW, then run the three-listen lab.
Collocation repair: “I did a mistake”
Classification: unusual/non-idiomatic for the intended meaning. A listener will almost certainly understand that you mean you made an error, but standard idiomatic English uses make a mistake: “I made a mistake.” The verb do is natural with many activities and tasks, but not with mistake in this ordinary collocation.
Now mute MODEL and produce a new sentence rather than repeating the memorized line: “I made a mistake in the first version.” If the stress and rhythm you practised can survive new wording, that is more useful than perfect imitation of one saved sentence.
Register check: asking someone to repeat
- “Could you say that again?” — neutral and polite in many everyday situations.
- “Could you repeat that, please?” — a little more formal and explicit.
- “What?” — grammatically valid and natural in some familiar contexts, but it can sound abrupt depending on relationship and tone.
Record two versions and compare not just pronunciation, but the stance you are producing.
Save the project before you flatten your evidence
Audacity 4 projects are saved as .aup4 files. The current Save your project guidance explains that the project preserves tracks, clips and edits so you can reopen the multitrack session.
Ordinary audio Export creates a finished audio file from what the project plays. In other words, export is where your carefully separated lanes can become audio lasagna again.
So save the Audacity project if you want MODEL and SHADOW to remain separately editable. Export a WAV or MP3 only when you actually want a finished listening copy.
One clean next take
Do not keep recording takes without a decision. After the first trustworthy comparison, choose one repair:
- Enter earlier on one recurring phrase.
- Keep one stressed word more prominent.
- Stop dropping one sentence ending.
- Preserve one reduction or linked sequence.
- Recover after a miss instead of freezing.
Then record another SHADOW take against the same MODEL and compare again. If you change the clip, mic, playback speed, target and recording setup all at once, you lose the benefit of separate tracks: you no longer know what caused the difference.
Two lanes, three listens
The setup is only useful if it makes the next language decision clearer.
Keep MODEL and SHADOW on separate tracks. Remove technical latency before judging yourself. Listen MODEL ONLY to map the target, SHADOW ONLY to hear what your own voice actually did, and BOTH to inspect timing and recovery.
Then make one repair and record again.
That is enough engineering. Your English would like the computer back now.
For the broader method of turning real media into practice, see media-based language learning.