Journal
A founder’s note

Keeping Up Is Not the Same as Speaking Clearly

Shadowing, listening blind spots, and the honest limits of TextShadow

Jiu Pang Hoy on sentence practice, word-level audio comparison, your last three attempts, and why green scores cannot replace rhythm or real conversation.

Jiu Pang Hoy

Jiu Pang Hoy

I used to treat keeping up with a recording as evidence that my shadowing practice was going well. The reference voice kept moving, I kept speaking, and finishing the passage felt like progress.

Then I turned off the reference and listened to my recording alone. Some endings were unclear. Some vowels did not sound the way I thought they had. I had reached the end of the sentence without putting the emphasis where I meant it to go.

The problem was not one poor attempt. It was practising without making room to hear what needed attention.

This is not an argument against shadowing. It is an argument for feedback—and for being honest about what a tool can tell you.

1. Following the audio can leave little attention for your own voice

Strict shadowing overlaps listening and speaking. TextShadow’s main practice loop is different: listen to a sentence, record it, then replay and compare. It is closer to sentence imitation or listen-and-repeat. Related names do not make these identical tasks.

When I listen, understand and speak together, “don’t fall behind” can take over. I might keep the pace without noticing a missing consonant or a word squeezed too short. The difference between how speaking feels and how the recording sounds is a useful reminder: feeling fluent is not the same as checking clearly.

I would not reduce this to everyone being fooled by bone conduction, or prescribe a universal fraction-of-a-second delay. Difficulty, familiarity and purpose matter. Slowing down enough to notice a detail is more useful than chasing every syllable.

2. Make feedback small enough to listen to

A sentence first, then your attempt

TextShadow starts simply: hear a synthesised sentence model, then record your version when you are ready. Listen again, try again, or return to Reading to hear the passage continuously. You do not have to speak over the model, and I do not claim this bypasses understanding or rewires your brain on demand.

What comes after the number?

The app reports sentence results and marks words red, yellow or green to suggest places to inspect. A colour is a model’s practice cue, not a final verdict on whether a word is right or wrong. Microphones, noise, accent variation and alignment errors can affect it. A score cannot diagnose your tongue position.

Tap a word and you can compare two sounds, with an important distinction:

  • Listen plays a dictionary or locally cached word reference, generated on-device if no usable recording exists. It is not a slice of that word from the current sentence demo, and its voice may differ from the sentence voice you chose.
  • My Voice plays the estimated word region from your sentence recording when audio and usable timing are available. Those boundaries are algorithmic estimates, not manually verified millisecond annotations.

This can help you notice differences, but an isolated word is not the only valid model for connected speech, reductions or coarticulation. If a slice sounds wrong, the room is noisy, or the colour disagrees with your ear, return to the full sentence.

Three attempts per sentence, not an unlimited archive

The current Android app keeps the latest three practice results for the same sentence within an article. Where saved audio is available, you can replay those attempts. This is not three recordings for the whole app, or permanent audio storage: cleanup, storage limits or a failed audio save can leave a result without playable sound.

I use the history to ask concrete questions: did I keep the ending this time? Is the emphasis clearer? A colour change can draw attention to something; it does not prove that a motor habit has been learned.

3. Practise something you will actually need to say

Repeating a supplied paragraph and finding your own words under pressure are different skills. A polished repetition does not automatically become an answer to an unexpected question.

That is why bringing your own text matters to me. Paste an English email draft, a project update, an article or a personal introduction. Import an EPUB you have the right to use. On Android, you can also select English text and send it to TextShadow through the system share sheet.

Sharing a web page often sends its URL, not its contents. Select the actual text before sharing; if your browser cannot share selected text, copy and paste it. The app organises practice by paragraphs and sentences. It is not a universal web-page extractor or a guarantee of perfect sentence boundaries. Check the imported material.

Two sentences for tomorrow’s meeting give practice a purpose. The software does not understand your job; you know why those sentences matter.

4. Green is not the finish line

Clear words and natural sentence rhythm are related, but different. Stressing every article and preposition to chase green marks can flatten a sentence. A synthetic reference is a practice aid, not the only acceptable accent or a substitute for human intention.

My suggested loop is:

  1. Choose real material. Take a short passage you understand and expect to use. Check the import and its meaning.
  2. Record, then inspect a detail. Compare one or two doubtful words with your own recording. Ask a teacher or proficient speaker when you need another judgment.
  3. Return to the sentence. Listen for stress, pauses, reductions and intent. Do not give every word equal weight.
  4. Put the text away. Explain the idea differently, or practise answering a follow-up question. This retelling is your exercise; TextShadow does not currently assess spontaneous conversation or retelling ability.

Once speech resources are installed, core reading aloud, recording and pronunciation assessment work offline on your device. Imported text and recordings are not uploaded to our servers. Resource downloads, purchases, entitlement checks and external links can still require a connection. Some scoring and recorded-word replay have free usage limits; Pro removes the relevant limits. Offline operation is not a promise that every operation is network-free or that privacy has no limits.

I want TextShadow to be a useful acoustic mirror, not a reason to chase a streak or a number. It can shorten the distance between listening, recording and comparing. Taking that practice into a real conversation still asks something of you.

Further reading

Hamada on shadowing and a review of automatic pronunciation assessment. These provide background, not evidence from a controlled trial of TextShadow’s learning outcomes.

TextShadow / Android

Get it on Google Play

TextShadow for Android is available on Google Play. Import a short passage to start reading.

About & support

Meet the developer, see what you can try and find help.

About & support