Listening identity session

Original fictional scene: the voice at the radio desk

Neri has both hands inside the station's map drawer when a ferry calls for the evening departure list. Nera sits at the radio desk. The call interrupts a conversation between them: Neri has just found an amended berth number, and Nera is checking the passengers against a handwritten sheet. Neither character is introduced again when the transmission begins.

“Neri, take the list,” comes through the speaker. Nera reaches for the sheet, then stops. It could be a request to the person beside the drawer, or the continuation of the desk conversation. Neri assumes that the radio operator has finished addressing the desk. The list stays between them while the ferry waits for an acknowledgment.

Nera finally asks, “The amended berth list, or the passenger list?” The answer restores the task before either person changes their name. Neri takes the map-room work; Nera acknowledges the passengers. The scene can preserve two established identities while giving a listener enough action and role information to follow the handover.

This is an original fictional scene, not an existing recording or a reported listener test. It illustrates an attribution problem: whom does a line address, and what action follows? We have not established that these spellings are acoustically confusable. A performance, the recording, the context and the offered choices all matter. The story supplies a reason to inspect a cue, not evidence of a measured cause.

If Neri were renamed Sol in the script but the audio file were unchanged, the new spelling would not repair the recording. If the operator named the task and then the recipient, the scene might become clearer without a rename. That is a creative hypothesis to compare in a new condition. Keep an original condition and new responses rather than editing an old answer key to make the previous choices appear correct.

Ask for identity, not remembered spelling

This workbench plays creator-selected local clips and records which declared identity an anonymous session selects. It answers a different question from asking someone to reproduce a spelling. A listener can identify a character while misspelling the name, or spell a visible option perfectly while attributing a voice to the wrong person. Do not substitute one observation for the other.

Use it for an audio-drama handover, a motion-comic cue or a voiced game's introduction. Supply public identity labels, a private trial key and your own clips. The tool does not generate recordings or pretend the initial fixtures contain spoken names. It does not recruit listeners, randomize a study, establish independent participants or predict recognition for a population. The default ledger is empty.

There is no best-name score. The private matrix contains recorded choices under the supplied conditions. It can help you decide what to inspect next, but it does not isolate the effect of a name from the rest of a performance. A wrong selected identity is a concrete observation; “the consonants caused it” is an additional explanation that these counts alone do not prove.

Prepare the options before opening the response view

Each identity has a stable ID and a public option label. Each trial declares its own ID, neutral clip ID, condition, neutral prompt, intended identity and allowed choices. There must be two to twelve allowed choices including the intended identity. Exact duplicate labels within one trial are rejected because two indistinguishable options would make the selection contract unclear. No pronunciation is inferred from a label.

The supplied trial order is fixed. It is not a randomized design, and repeated trials or replays are not treated as independent people. Use neutral prompts such as Clip 1. Do not put the intended character into the prompt, clip ID, condition or filename you show to a listener elsewhere. A technically hidden key cannot repair a question that announces its own answer.

Select a local file for each clip ID in the preparation area. Files must be nonempty, no larger than twenty megabytes each and have a readable duration above zero and at most ten minutes. The actual browser must support the file. A filename extension alone does not prove that it can play. Metadata failures remain unavailable; the software does not substitute a silent successful trial.

Local object URLs provide playback in this page. The File API describes access to selected file objects and object URLs; this implementation makes no upload request for your clips. A local SHA-256 content token identifies the selected bytes within the private playback record. Neither the audio nor its filesystem name is put into the project. That token is an implementation reference, not a rights check or a certification of what the clip says.

A session freezes the question it actually asked

Enter a new anonymous session token, then open the response view. Use a token such as session-01, not a person's name, email or contact details. The session stores the exact current identities and trials. Opening creates an empty session record; it does not create a listener answer or count a trial as observed. A token is not proof of a unique person.

The ordinary response view hides the author input, intended identities, guide and private analysis. It shows the neutral prompt, player controls and allowed labels. This is display separation for a cooperative exercise. The source and editable records still exist in the page. It is not a secure exam or protection against a participant who inspected the key beforehand.

An existing open token can resume only when its frozen protocol exactly matches the current one. A changed label, intended identity, trial order or option set changes that comparison. Closed sessions with a different protocol are retained but excluded from the current matrix. The software does not silently reclassify yesterday's selection using today's preferred target.

The same principle applies when a project is restored. The recorded protocol and responses can return, but local audio access is cleared. Reselect the files. An old playback record remains an editable record of the earlier session; it does not mean the current tab has loaded those bytes or played the clip again. Save or export your closed session before leaving the page; the live response view is not an automatic persistent recorder.

Playback has its own states

Use Play clip to begin, or the native audio controls. The page records media play starts and covered endings. An identity choice or an uncertain answer becomes available only after an ended event with sufficient browser-reported played coverage. Merely loading metadata, moving a seek cursor or having a duration value is not a completed exposure in this workflow.

The coverage gate uses the media element's played ranges, allowing a tolerance no greater than fifty milliseconds or five percent of a short clip. That tolerance is an internal implementation choice, not a hearing threshold. Coverage can be accumulated across the ranges the browser reports. A covered ending is not a guarantee of one uninterrupted pass, audible output, attention or comprehension. The HTML media specification documents those browser states; it does not turn them into psychological evidence.

Replay from start records an explicit replay request and reloads the local source. Native pause and resume can create more play starts, so starts are not labeled as full replays. A failed play request cannot become a completed answer. The current asset, trial and session are checked before asynchronous events affect the live state; a late event from another clip must not enable the next trial's answer.

Skip records unplayed when there has been no play start. After an attempted start it records skipped, preserving the playback state, including failure or incomplete coverage. An uncertain response requires the same ended coverage as an identity choice, but contributes no selected-identity cell. This keeps a listener's uncertainty different from a trial that was never presented successfully.

Read the matrix without inventing a denominator

Close the session to return to private analysis. Only closed sessions with an exactly matching frozen protocol enter the current counts. The matrix separates conditions and intended identities. Its columns are selected identities. A response increments one intended-by-selected cell; it does not increase another condition's count because a name happens to be shared.

The choice denominator counts identity choices only. Uncertain, skipped, unplayed and missing remain separate counts. Missing means a configured trial in a matching closed session has no response row. It is not an incorrect answer. Before any identity choice is recorded for a row, the table shows a dash rather than suggesting that zero mistakes establishes perfect recognition.

The recorded audio content matters as well as the question labels. If the same trial ID has play starts with different content tokens across matching closed sessions, the affected row is marked Unavailable and its identity-choice counts are withheld. Original responses remain in the private ledger. This is not zero observed choices. Inspect the tokens and establish distinct recording conditions before comparing; another file under the same clip label is not silently treated as the same performance.

For example, suppose a synthetic development session selects b for a trial intended as a, then leaves the second trial unrecorded before closing. The first row has one selected-b count and one incorrect attribution; the second has a missing observation. This example explains accounting only. It is not included as a real listener in the default ledger and supports no conclusion about Neri or Nera.

An exact recorded row can be undone while the session is open. Undo resets the live playback state for the trial and preserves unrelated records. If someone edited that row or closed the session, the guarded operation refuses to overwrite it. You can inspect the private session data, but editing those records is an authored assertion, not a way to certify that an event occurred.

Choose a revision that changes the relevant cue

When a cell draws attention, listen to the actual clip beside the action and permitted options. Consider whether an address, task cue, interruption or performance needs revision. Do not rename every character on the strength of one session. Keep the changed recording in a separate condition and retain both question contracts if you intend to compare them. The workbench does not balance different listening contexts for you.

The private handoff contains the frozen protocols, responses, playback states and count definitions. It excludes raw audio and filenames, but author-written labels or prompts can still identify people or expose story information. Review it before sharing. No message is sent to a listener, no contact list is built and no participant identity is verified.

Use the pronunciation handoff tool when performers need approved readings and scene preparation. Use spelling recall when the question is what a reader can reproduce in writing. Use cast separation for a declared textual heuristic before recording. These tools can support the same production, but none is evidence that another task has already been tested. The useful outcome here is a traceable listening-choice record and a narrower creative question for the next revision.

Does an ended clip prove the listener heard it?

No. It records browser media events and played coverage. Muting, output problems, attention and comprehension are not established. The records remain editable local assertions.

Why are uncertain and skipped not wrong choices?

They answer different questions. Uncertain follows a completed presentation without selecting an identity. Skipped can follow an attempted playback, and unplayed has no start. None invents a selected person.

Can I change the intended identity after collecting responses?

The session retains the exact original protocol. A changed target, label, order or choice set excludes that old session from the current matrix; it is not silently rescored.

Will a saved project reopen the original audio files?

No. It retains protocols and records without audio or filenames. Restoration clears live audio access and asks for file reselection. Save or export closed sessions before leaving; the response view is not autosaved.