TIME IS RUNNING OUT FOR 2026 TAX WRITE-OFFS

Starting Clearly After a Voice Writing Pause

Martel VOX Elite laptop; screen display simplified for illustration.

A clear start after a pause begins with a ready microphone, a comfortable speaking technique, and the intended text destination. A short practice worksheet lets a reporter check those parts of the workflow together. If the first word appears differently from the planned phrase, compare the saved speech, the microphone state at speech onset, and the initial text from that same attempt before choosing the next supported check.

A useful practice exercise repeats one identical phrase at several pause lengths and preserves the observations for each individual performance: the planned words, any saved speech from that attempt, and the first text produced in its selected destination. Use a standalone Dragon destination first and a CAT practice destination in a separately labeled pass. This guide provides a five-pause worksheet for doing that on a Windows 11 voice-writing laptop, with clear notes that an instructor or software support specialist can help interpret.

The matter name, speaker names, phrases, pause lengths, file labels, settings, and observations below are fictional and illustrative. They are not customer results, court records, recognition benchmarks, or guarantees.

Describe the opening phrase precisely

“The first word is missing” can describe several different events. Before changing settings, write down which one you can actually observe.

  • The raw audio starts late. Playback begins after the first consonant, syllable, or entire word.
  • The recorded and written openings differ. The saved sound contains the word; the first standalone text result begins at a different point.
  • Separate practice attempts have different openings. The standalone and CAT destinations contain different results from separate performances; record each attempt’s conditions for review.
  • The text went somewhere else. A focus change makes a word appear missing even though it was inserted in another field or window.
  • The microphone was asleep or off. The reporter spoke before Dragon showed a ready microphone state.
  • A phrase was treated as a command. The words triggered an action or were parsed differently after a pause.
  • Only the display looked delayed. The word appeared later, after the reporter had already begun correcting the line.

Do not combine these into one “accuracy” result. A missing initial T in Today is not the same observation as losing the whole word. Record the smallest unit you can hear and see.

Start with a clean, authorized practice environment

Use invented content in a practice file. Do not troubleshoot with confidential testimony unless your organization’s rules, permissions, recording notices, and retention procedures allow it. Preserve any diagnostic audio only as long as policy permits.

On Windows 11, close unnecessary notifications and applications, but keep the test realistic. Connect the same approved stenomask or microphone path used for practice. Record the exact Dragon product name, edition, version, active user profile, input device, CAT version, and Windows build. If any field is unknown, write Unknown; do not guess.

Confirm that Dragon is using the intended source. The Windows default input and the source selected in a speech application are not always the same. If the CAT application also has an audio or recognition selection, record that separately.

Microsoft’s current Dragon microphone guidance describes green as on, blue as sleeping, and red as off. It also tells the user to wait until the microphone icon is green before speaking after turning the microphone on. Record that visible state when the practice phrase begins so the worksheet reflects the selected listening mode.

Environment field Illustrative entry
Computer Martel VOX Elite voice-writing laptop
Windows Windows 11 Pro; copy the installed build
Dragon Copy exact customer-facing edition and version
Dragon profile Lena-Pause-Practice, fictional
Stenomask/input Record the selected device name
Standalone destination DragonPad or another supported plain practice window
CAT destination Authorized fictional practice job
CAT audio/recognition route Record the selected route
Microphone icon before each phrase Green, blue, red, changing, or not observed
Test date November 19, 2026, illustrative

Restarting can establish a repeatable baseline, but a restart is not a diagnosis by itself. After restarting, allow Windows 11, Dragon, and the authorized CAT application to finish opening before the first measured phrase.

Identify what each evidence source can establish

The worksheet has four principal evidence sources. Keep their relationships explicit. Two recordings of the same printed sentence are separate performances unless you have a supported way to preserve the same utterance at both checkpoints.

1. Planned phrase

Write the exact phrase before speaking. Mark the first word and its first syllable. If the phrase changes between trials, the results cannot be compared cleanly.

2. Saved raw speech

Make a short practice recording with Windows Sound Recorder or another authorized recorder that captures the intended mask or microphone. It establishes what reached that recorder. A waveform can help locate an onset, but listen as well; a faint breath or consonant may not be obvious visually.

Here, raw saved speech means the original saved practice recording before any editing or listening-copy adjustment. It does not promise an unprocessed electrical signal or identical audio processing in every application. Record any known capture processing rather than assuming that two programs receive interchangeable data.

If the supported setup can record the same utterance while Dragon or the CAT application receives it, record the source selection in both applications and label that audio with the attempt ID. Do not change device-sharing settings or add an unsupported route merely to obtain simultaneous evidence. A separate Sound Recorder pass is still useful, but it is a different performance and must be labeled that way.

Dragon’s blue or red microphone state describes Dragon’s listening state. It does not, by itself, stop an independent Sound Recorder recording. If that separate file also loses a sound, record it as another observation requiring its own explanation. A green Dragon icon likewise does not prove that either application selected the intended device.

3. First standalone Dragon result

Dictate into DragonPad or another supported plain destination. Save or copy the first recognition result before correcting it, with the attempt’s microphone state and available audio. A corrected line cannot prove what Dragon produced initially. This is Pass A in the worksheet.

4. First CAT result

Repeat the controlled phrase in a fictional CAT practice job as Pass B. Preserve the first visible result, insertion point, microphone state, selected recognition route, and any audio captured for that particular attempt. A difference from Pass A suggests a destination-related comparison, but a new spoken performance also permits differences in delivery and recognition.

Only describe a word as lost between two processing stages when the supported workflow lets you preserve the same utterance’s earlier and later results. If the earlier result is unavailable, write Earlier result not observed. Do not substitute a successful DragonPad line from a previous attempt for that missing evidence.

For a difficult case, add a fifth column for a screen recording only if policy permits. The four core layers are usually enough to avoid guessing.

Use one identical invented phrase at every interval

The fictional matter is Harbor Lantern Review, and the invented speaker is Talia Merrow. The measured target is always:

Today Talia marked the blue folder.

Repeat that entire sentence at every pause length in both passes. The first token stays Today. Its spelling, capitalization, surrounding words, and intended punctuation remain unchanged, so the comparison does not confuse a different opening word with a different pause.

After completing the baseline, you may investigate another opening sound with a second complete block. For example, use Before lunch Talia checked the index. at all five intervals, and label the new block Before-series. Do not compare only Today at a short pause with only Before at a long pause. To explore both wording and pause length, every selected phrase needs its own full set of intervals.

Do not use names or phrases from an actual matter. Do not train, correct, or alter a production profile until the observation pass is complete.

Run five staggered pauses

Use five pause lengths that are easy to reproduce. The illustrative worksheet uses approximately 0.5, 1, 2, 4, and 8 seconds. These are diagnostic intervals, not product specifications and not thresholds promised by Martel, Microsoft, Nuance, a CAT publisher, or a court reporting school.

For each trial:

  1. Say the same short lead-in: Practice marker.
  2. Stop speaking and start the timer.
  3. Observe the Dragon microphone icon without touching the input device.
  4. At the assigned interval, begin the planned phrase at a normal, consistent volume.
  5. Continue through the whole phrase; do not stop to correct the first word.
  6. Save that attempt’s available audio and first text output; identify separate performances explicitly.
  7. Mark the exact first sound, syllable, or word that is absent or changed.
  8. Rest for a consistent interval before the next trial.

Nuance’s dictation guidance recommends clear, natural speech at a consistent speed and volume. It also notes that longer phrases generally provide more context. Therefore, do not turn each first word into an isolated, exaggerated shout. Speak the complete sentence naturally.

The five-pause observation sheets

These invented entries demonstrate recordkeeping only. For this illustration, an authorized setup preserves the same attempt’s input audio independently while the destination receives speech. The recorder’s selected input is documented, and its recording continues independently of Dragon’s microphone state. A reader whose setup cannot preserve that evidence should write No same-attempt audio instead of filling in an assumed result.

Pass A: standalone Dragon destination

The full planned sentence is identical in every row: Today Talia marked the blue folder.

Attempt Pause Dragon state at onset Same-attempt recorded beginning First standalone text Observation
A-P1 0.5 sec Green Today Talia... Today Talia... Complete opening observed
A-P2 1 sec Green day Talia... day Talia... Opening syllable absent in this recorded performance
A-P3 2 sec Green Today Talia... Today Talia... Complete opening observed
A-P4 4 sec Blue Today Talia... No dictated sentence Independent audio complete; Dragon was sleeping
A-P5 8 sec Green Today Talia... Today Talia... Complete opening observed

A-P4 is retained as an observation but excluded from a comparison intended to hold Dragon’s on-state constant. It must be rerun with a new attempt ID after Dragon is ready. The table does not claim that a four-second pause puts Dragon to sleep; the actual state is a separate recorded variable.

Pass B: CAT practice destination

These are new spoken performances of the same target. Each has its own audio; Pass A’s audio is not reused as evidence for Pass B.

Attempt Pause Dragon state at onset Same-attempt recorded beginning First CAT text Observation
B-P1 0.5 sec Green Today Talia... Today Talia... Complete opening observed
B-P2 1 sec Green Today Talia... Today Talia... Complete opening in this separate performance
B-P3 2 sec Green Today Talia... Today Talia... Complete opening observed
B-P4 4 sec Green Today Talia... Today Talia... Ready state observed throughout onset
B-P5 8 sec Green Today Talia... Talia marked... First CAT word absent; earlier recognition result not observed

A-P2 calls for a closer look at capture and delivery in that attempt. A-P4 documents a listening-state difference while independent audio remains complete. B-P5 calls for route-specific checking of recognition, selected input, focus, and CAT behavior. A-P5’s complete text does not prove that a complete word entered the CAT route during B-P5.

Label loss at the syllable level

Use a small code set so two reviewers describe the same event the same way.

Code Meaning Example with Today
A0 Complete first word Today
A1 Opening consonant absent oday heard or implied
A2 Opening syllable absent day begins the audio or text
A3 Entire first word absent Output begins with Talia
A4 First word substituted To date or another first token
A5 Same utterance verified at an earlier stage, absent at a later stage Supported same-utterance records show Today before a transfer and not after it
A6 Location unresolved No preserved audio or first output

Do not force an uncertain observation into A1 or A2. Use A6 and state what is missing. Precision includes admitting that the evidence cannot support a conclusion.

For B-P5, record A3 for the missing word in the CAT text and Earlier stage unobserved as a separate evidence note. Do not assign A5 merely because A-P5 succeeded. A5 requires an actual earlier result for B-P5’s utterance, preserved through a supported workflow.

Compare audio before changing recognition settings

If the first sound is absent from the raw saved speech, recognition vocabulary changes cannot restore sound that never reached the file. Inspect the physical and acoustic path first.

Check that the mask position stayed consistent, the seal did not shift during the pause, and the reporter did not begin speaking while bringing the mask back into position. Look for a breath, mouth noise, handling sound, or automatic input behavior immediately before the word. Repeat the same phrase without deliberately increasing volume.

The goal is not to make a louder test succeed. The goal is to learn whether normal technique produces a complete onset. If moving the mask or starting softly changes the result, record that as an illustrative technique observation and return to the organization’s approved training process.

The Stenomask Troubleshooting Guide provides broader checks for mask audio and recognition. The pause worksheet focuses on the exact opening sound, repeated at five intervals, with each observation tied to its own performance and evidence source.

Read the Dragon microphone state at the right moment

A green icon observed ten seconds before the phrase is not the same as a green icon at speech onset. Add a timestamp or observer note for the moment the first word begins.

If the icon is blue, the microphone is sleeping. If it is red, the microphone is off. Wake or turn on the microphone using the supported method for the installed Dragon edition, wait for the green state, and then speak. Do not invent an undocumented command. Use the product’s current help when the installed edition differs from a general instruction.

If the icon repeatedly changes state without an intended command, preserve that observation. Check whether a hotkey, button, power policy, profile setting, or authorized application integration is changing the state. Change one variable at a time and repeat the same five-pause sheet.

Distinguish a pause from a command boundary

Dragon uses pauses to help separate commands from ordinary dictation. Nuance’s troubleshooting guidance advises adjusting the pause before a command and speaking command-like words as part of a longer phrase when they are intended as dictation. This means a pause can affect interpretation even when the audio is complete.

Record whether the missing first words resemble a command in the installed edition. Do not assume that every ordinary word is safe or that every command name is active. If the interface changes, a selection moves, or text is formatted instead of inserted, note the action as well as the missing text.

For a control trial, place the suspect word inside a longer invented phrase without changing the word itself. For example, compare Select the blue folder after a long pause with Talia will select the blue folder after the same pause. These are illustrative phrases. Use the installed Dragon documentation to interpret any actual command behavior.

Use standalone recognition before the CAT application

A short standalone pass provides a useful first comparison. Open a supported plain Dragon destination, confirm the microphone is green, and dictate one minute of invented practice material that includes the five pause intervals. Preserve the initial output so an instructor or support specialist can review the same starting point.

Then open the authorized CAT practice job and repeat the same planned phrases. Do not run both windows as uncontrolled destinations at once. Label the files so that the reviewer cannot confuse the standalone and CAT passes.

Read each pass against its own audio first. If that attempt’s recording lacks the opening sound, inspect capture and delivery for the recorded route. If its audio is complete while its text omits the opening word, check the selected recognition input, listening state, profile, and command interpretation. A persistent difference between standalone and CAT attempts gives a reason to inspect the CAT route, but it still needs controlled follow-up. Do not infer that one specific CAT component removed a word without observing the same utterance at the relevant checkpoints.

Check window focus without turning this into a focus guide

Before every CAT trial, click the intended fictional practice field and confirm the insertion point. After speaking, search the open practice windows for the first word before declaring it missing. Record any notification or dialog that appeared at speech onset.

Keep this worksheet focused on the opening word after a controlled pause and the observations available for that attempt. If text consistently appears in another application, record its location and use a separate window-focus rehearsal. That keeps the pause exercise and the text-destination question clear for the person helping you.

Do not correct during the observation pass

Correction changes the evidence. Save the first raw audio, first standalone text, and first CAT text before opening a correction menu, editing the vocabulary, retraining a word, or changing formatting.

After the observation pass, corrections can be tested on a duplicate practice profile or according to the organization’s approved procedure. Record every change and repeat all five pauses. A result after three simultaneous changes cannot identify which change mattered.

Use a simple change ledger:

Change ID Single change Reason Profile/file protected first? Five-pause rerun complete? Outcome
C01 Wait for green icon before phrase A-P4 began in sleep state Yes Yes Record all five rows
C02 Restore consistent mask position A-P2 recording lacked the opening syllable Yes Yes Record all five rows
C03 Repeat both destination blocks B-P5 differs from A-P5 Not applicable Yes Preserve new audio for each performance

All entries are illustrative. A real ledger should contain only actions actually taken.

A worked diagnostic decision

Consider the three exceptions in the illustrative sheets: A-P2, A-P4, and B-P5.

For A-P2, the listener hears day Talia in that attempt’s recording. The target was Today Talia. The first syllable is absent from the recorded performance, and the text begins similarly. Inspect the recording route and ordinary mask delivery before attempting a vocabulary correction. The later complete B-P2 performance does not restore the missing sound to A-P2’s file.

For A-P4, independent audio contains the complete phrase while Dragon’s icon is blue and no dictated sentence appears. That is consistent with Dragon being in its sleeping state. It gives no basis for saying that the independent recorder stopped or clipped the speech. Rerun the row after Dragon is green, preserving both the original exception and the new result.

For B-P5, the CAT-pass audio contains Today, but the first CAT text begins with Talia. Its earlier recognition result was not preserved. A-P5’s complete standalone line was a different utterance, so the missing stage remains unknown. Check the CAT pass’s selected recognition route and input, then repeat with the publisher’s supported diagnostic procedure. Include the exact initial output and that pass’s audio in the support package.

The findings support three next checks, each tied to a recorded observation. Keep the pause length, listening state, and text destination distinct when discussing the worksheet with support; the appropriate next step depends on the specific practice attempt.

Repeatability matters more than one success

After a single-variable change, run the five pauses again in the same order. Then reverse the order on a second pass if time permits. A fixed order can hide warming, fatigue, or a state change that happens later in the session.

Do not calculate a recognition percentage from a few practice attempts and present it as an accuracy result. This is a localization exercise, not a benchmark. Report observations such as No ordinary dictated text appeared in two preserved attempts that began while Dragon was sleeping, with the attempt IDs and conditions attached. Keep those state exceptions separate from a comparison that requires the microphone to remain on.

If the symptom cannot be repeated, preserve the original evidence and record Not reproduced. Do not fabricate a cause to fill the worksheet.

Escalate with a compact evidence package

An effective support package is small enough to review and complete enough to reproduce.

Include:

  • the exact customer-facing Dragon product, edition, version, and build;
  • the Windows 11 edition and build;
  • the active Dragon profile name, using a safe identifier;
  • the input device path and whether the icon was green, blue, or red at speech onset;
  • the exact invented phrase and assigned pause length;
  • the saved raw practice audio, if policy allows;
  • the uncorrected standalone output;
  • the uncorrected CAT output and CAT version;
  • the A0–A6 loss code;
  • the one-variable changes already tested; and
  • the time and order of each trial.

Do not send confidential case audio, passwords, license keys, customer names, or unrelated logs. Follow the receiving organization’s secure support route.

Make the pause timing reproducible

Approximate timing is acceptable only when it is recorded as approximate. For a stronger comparison, use a visible timer operated by a second person or an audible practice cue that is not routed into the recognition input. Do not tap the mask, cable, or laptop to mark the pause; the marker itself could change the input path.

Start timing after the final sound in Practice marker. and stop timing when the first sound of the measured phrase begins. The timer does not need laboratory precision. It needs a consistent rule that another reviewer can follow. Write about 2 seconds when that is what was measured. Do not report 2.000 seconds merely because a spreadsheet can display three decimal places.

If the reporter cannot watch the microphone state and a timer simultaneously, use an observer. The observer records icon color at speech onset and does not coach or interrupt the phrase. On a solo test, a permitted screen recording may provide timing evidence, but only if its storage and privacy implications are acceptable.

Create filenames that preserve the order without relying on Windows sorting by a vague creation date. An illustrative set is:

  • A-P1_0-5sec_audio.wav
  • A-P1_0-5sec_standalone.txt
  • B-P1_0-5sec_audio.wav
  • B-P1_0-5sec_CAT.txt
  • A-P2_1sec_audio.wav
  • B-P2_1sec_audio.wav

When Sound Recorder was run separately from dictation, use a separate block label such as R-P1 and record Separate performance in the sheet. An A-P1 audio file must refer to A-P1’s actual performance, not to an earlier successful recording with the same printed target.

Keep the original files read-only when possible and place annotations in a separate worksheet. If a file must be converted for an approved support route, retain the original and label the derivative clearly.

Add a no-pause reference without changing the five-pause set

A short reference phrase spoken as part of continuous dictation can show whether the first token is generally difficult for the active profile. It is not one of the five measured pauses and should not replace any of them.

For example, dictate: The practice continues and today Talia marked the blue folder. The target today now occurs mid-utterance. If it is complete there but incomplete after a pause, pause-boundary timing remains relevant. If it is changed in both positions while the audio is complete, the word, profile, pronunciation, or recognition context deserves separate inspection.

Use the reference only as a branch in the diagnostic tree. Do not train the word between the reference and measured trials, and do not claim that one mid-sentence success proves the whole input path is correct.

Separate session start from an ordinary pause

The first phrase immediately after opening Dragon, waking the microphone, switching profiles, or opening the CAT job is a session-start trial. A pause that occurs after several minutes of successful dictation is an in-session trial. Label them differently.

This distinction matters because an application can still be loading, a profile can still be initializing, or a destination can still be acquiring focus at session start. In contrast, an in-session loss may follow sleep state, technique drift, a notification, or a command boundary. Do not pool the two conditions in one row.

An illustrative worksheet might add S for session start and I for in-session: S-P3 is the two-second pause immediately after opening the practice environment, while I-P3 is the same pause after five minutes of continuous invented dictation. Repeat the phrase order so that a different first word does not become the hidden variable.

Keep the USB and power path stable during the first pass

Do not move the stenomask connection from a direct laptop port to a hub halfway through the observation pass. Do not connect or disconnect charging power, a dock, an adapter, or another audio device between P1 and P5 unless that change is the variable being tested.

Windows 11 may enumerate audio devices differently after a connection changes. A newly connected device can become the visible default while Dragon or the CAT application retains another source. If a cable, adapter, or port must be changed, end the first pass, record the change, re-confirm every selected input, and begin a new labeled pass.

The same rule applies to battery and power-saving experiments. If a reporter suspects that a device sleeps after silence, create a separate pass using the unchanged sentence at all five intervals and record the power condition. Do not infer a sleep timer from one lost word. Observe the device state that is actually available and escalate manufacturer-specific questions through the approved support route.

Listen for pre-word evidence

When reviewing raw speech, begin playback a few seconds before the first word. A file clipped to start exactly at the expected word can hide useful evidence. Listen for the end of Practice marker., room tone, the reporter’s inhale, mask movement, a click, or a sudden noise-floor change.

These sounds do not prove a cause. They help describe the transition into speech. For example, a complete inhale followed by a missing opening consonant differs from a file that begins halfway through the word. A handling sound followed by a faint onset differs from a clean, complete onset in the audio associated with an incomplete text result.

Keep listening conditions consistent. Use the same approved headset, comfortable level, and playback application. Do not increase gain until background noise becomes misleading, and do not edit the raw file before the first review. If a reviewer applies a copy-only level adjustment, label that listening copy and preserve the original.

Use an interpretation matrix, not a single diagnosis box

The evidence layers can be summarized without overclaiming.

Evidence from the identified attempt First text in that attempt Useful next branch
Opening sound absent from its recording Opening sound or word absent Capture path and delivery in the recorded route
Complete audio; Dragon sleeping No ordinary dictated text Supported wake procedure and ready-state rerun
Complete audio; Dragon ready First word absent Selected input, recognition, profile, command interpretation, destination
Complete audio and complete first word Complete opening Mark A0 and retain the conditions
Audio absent or from another performance First word absent Obtain better same-attempt evidence if supported
Same utterance’s earlier text preserved and complete Later result lacks the word Inspect the observed transfer or transformation

Use this matrix to organize a conversation with an instructor or support specialist. Several conditions can be relevant in one session. Each row points to a narrow next check, with the recorded observations available for the person helping you interpret the supported workflow.

Audit the worksheet for hidden changes

Before interpreting the five rows, compare these fields across all of them:

  • same Dragon profile;
  • same input device and physical connection;
  • same mask position instruction;
  • same Windows 11 power condition;
  • same standalone destination for the standalone pass;
  • same CAT practice job and insertion field for the CAT pass;
  • same automatic punctuation and command settings;
  • same correction status, with no corrections during the observation pass;
  • same room or documented room change; and
  • same speaker unless speaker identity is the explicit variable.

Highlight any row where a field changed. The row may still be useful, but it cannot be compared as if only pause length differed.

Also check whether the planned first word changed capitalization, punctuation, or surrounding context. Today, Talia... and Today Talia... are not identical targets. Preserve the exact target string rather than relying on memory.

Plan a safe second pass

The second pass should answer one question raised by the first. Choose one of these branches:

  1. State branch: wake or turn on the microphone, wait for green, and repeat all five rows.
  2. Technique branch: restore the approved mask position and repeat at the same natural volume.
  3. Standalone branch: keep the input unchanged and repeat only in the supported plain destination.
  4. CAT branch: keep the complete standalone evidence and repeat in a fresh fictional CAT field with confirmed focus.
  5. Connection branch: move to an approved direct port, re-confirm the selected input, and begin a completely new labeled pass.

Do not choose the branch that is easiest merely to obtain five successful rows. Choose the branch that tests the strongest evidence from the first pass.

After the second pass, compare row to row. If all five rows change after one controlled adjustment, record that association. If only one row changes, preserve the mixed result. If the pattern moves to another pause length, the evidence does not support a simple fixed-duration threshold.

Know when to stop the practice test

Pause the exercise and ask the appropriate support team for guidance if you cannot confirm the selected input, the application’s ready indication, or the saved practice copy. Also pause when recording permissions need clarification. Preserve the latest observations and the current settings so support can recommend the next step from a clear starting point.

Also stop when fatigue changes normal delivery. Repeating a difficult opening word dozens of times can alter articulation and mask position. A short, well-labeled set is more useful than a long set whose conditions drift.

The objective is a concise practice record that another person can review. An open question can be documented clearly: Recorded opening complete; first written word differs in two practice attempts; settings unchanged; guidance requested. Keep the exact planned and observed phrases alongside that note so support can interpret the difference.

Discussing your workflow with Martel

The Martel VOX Elite voice-writing laptop package and the Voice Writers Vox Dragon Laptops category provide the current package choices. Consult the current listing and written order for the configuration available when you buy. This worksheet can accompany a discussion about the software, input, setup, and support your workflow needs.

Bring the planned phrase, recorded observations, and relevant practice evidence so the conversation can address your chosen software route and setup. An instructor or the appropriate support team can use those details to recommend a supported next check. Keep confidential case material out of a general purchasing inquiry.

Frequently asked questions

1. Why can the first word disappear after a pause?

The opening sound may not reach a recording, Dragon may be sleeping or off, the phrase may be interpreted at a command boundary, or the destination may receive different text than expected. Preserve the evidence available for the affected performance and identify any unobserved stage. Do not infer a cause from the final sentence alone.

2. How long should the pause be for this test?

Use repeatable staggered intervals. This guide’s 0.5, 1, 2, 4, and 8 seconds are illustrative diagnostic values, not universal limits or promised thresholds.

3. Should I speak the first word louder?

No. Use clear, natural speech at a consistent volume. An exaggerated first word can hide a timing or mask-position problem and makes the trial unlike normal voice writing.

4. What does a green Dragon microphone icon mean?

It indicates that the microphone is on. Confirm green at the moment speech begins, particularly after turning on or waking the microphone.

5. What does a blue Dragon microphone icon mean?

It indicates sleeping state in Microsoft’s current Dragon microphone guidance. Wake the microphone using the supported method for the installed edition, wait for green, and then speak.

6. What does a red Dragon microphone icon mean?

It indicates that the microphone is off. Turn it on using the supported method and wait until it is ready before beginning the measured phrase.

7. Why save raw audio if Dragon already shows text?

Raw audio shows whether the opening sound reached a file. If the sound is absent there, changing recognition vocabulary cannot reconstruct it.

8. Can Windows Sound Recorder be used for the raw layer?

Yes, when organizational policy permits and it captures the intended input. Label whether it recorded the same utterance as the text attempt or a separate performance. Dragon sleeping does not by itself stop that independent recording, and a separate successful recording does not prove what reached Dragon in another attempt.

9. Why test in DragonPad or a plain destination first?

It provides a simpler destination baseline. If a later CAT attempt differs, check that attempt’s own audio, input selection, listening state, and text first. The contrast can justify checking the CAT route, but separate performances alone cannot prove that a complete recognized word was removed downstream.

10. Why preserve the first output before correcting it?

The correction changes the evidence. The uncorrected result is needed to compare recognition with the raw audio and CAT output.

11. What if only the first consonant is missing?

Record it as an opening-consonant loss, not as a whole missing word. Listen for the exact onset and use the A1 code or a precise written note.

12. What if the first syllable is missing?

Use the A2 code and write the heard remainder. In the main practice sentence, recording Today as the planned word and day as the heard opening gives the reviewer a precise comparison. Keep the whole sentence unchanged across the pause intervals so the opening syllable remains comparable.

13. What if the whole first word is missing?

Use A3, identify the first surviving word, and compare raw audio, standalone text, and CAT text. The first surviving token helps locate the boundary.

14. What if the word appears in standalone Dragon output but not the CAT file?

First determine whether the two results belong to the same utterance. Separate passes establish different outcomes and call for a controlled rerun. Use A5 only when a supported record preserves the same utterance’s earlier complete text and later incomplete text. Otherwise identify the missing first word and mark the earlier stage unobserved.

15. Could a command-like word cause the issue?

Possibly. Pauses help Dragon distinguish commands from dictation. Record any action or formatting change, then compare the suspect wording inside a longer invented phrase using current help for the installed edition.

16. Should I change several settings at once?

No. Protect the profile and practice files, make one documented change, and rerun all five pauses. Multiple simultaneous changes prevent a useful conclusion.

17. Should I create a recognition percentage?

Not from this worksheet. The unusual phrase set is designed to localize a first-token problem, not to rank systems or predict production accuracy.

18. What if the problem cannot be reproduced?

Record Not reproduced and preserve the original evidence. Include what differed between the original event and the practice attempt; do not invent a cause.

19. Can I use an actual deposition for the test?

Use invented practice content unless applicable permissions and confidentiality, recording, retention, and security rules clearly allow real material. A fictional phrase is usually easier to share safely.

20. Should I test after restarting Windows 11?

A restart can provide a clean baseline, but it is not a diagnosis. Allow the authorized applications to finish opening, record the environment, and still run the layered comparison.

21. What should I send to support?

Send the exact versions, device route, icon state at onset, invented phrase, pause length, permitted raw audio, first standalone and CAT outputs, loss code, and one-variable change log through the approved secure route.

22. Why keep the first word identical at every pause length?

Changing both the first word and the pause makes the cause ambiguous. The opening sound, word familiarity, or delivery could explain the difference. Complete the five intervals with one sentence before testing another sentence as a new full block.

23. What if running a separate audio recorder changes the behavior?

Stop that comparison and record the added recorder as a changed condition. Do not force device sharing or modify capture settings without supported guidance. A separate audio-only rehearsal can still help, but label it as a different performance and explain that simultaneous evidence was unavailable in the normal route.

24. Does this Dragon procedure also describe EclipseVox Speechmatics processing?

The pause worksheet is written for Dragon. In an EclipseVox Speechmatics session, follow the instructions for that speech-processing option and create a separate baseline. The version list associates local Speechmatics with Eclipse 10 and Eclipse 11; in Eclipse 12, either local processing or cloud recognition can be selected. For the local Enhanced option, the required purchase is the software publisher’s Boost Box or one of its qualifying laptops. Identify that device on the Speechmatics worksheet before rehearsing pauses.

Close the test with a precise finding

Finish the worksheet with the attempt IDs, the exact missing sound or word, the evidence actually preserved, and the next supported check. Write A-P4: complete independent audio; Dragon sleeping; no dictated text when that is what you observed. Write B-P5: complete associated audio; first CAT word absent; earlier recognition text unobserved when the earlier stage remains unknown.

Those statements preserve useful facts without filling gaps. Keep the original exception and the rerun together, restore any deliberately changed practice settings, and use the remaining evidence to decide whether to seek help with capture, recognition, or the destination workflow.

Martel Electronics