Audio-guide transcripts. Match every stop.

Give each listening stop a readable companion

An audio guide can let visitors choose a stop, skip ahead or return to an object later. Its transcript needs to remain attached to the right recording whichever order they choose. Polyglot can transcribe authorized audio locally on an Apple Silicon Mac, let you review the text alongside playback and export a plain-text transcript. Your guide’s website or publishing system handles the visitor-facing page and navigation.

This workflow is for museums, heritage sites and independent guide producers who own their recordings or have permission to process and distribute them. It addresses separate audio stops, rather than captions over a continuous walking-tour film or an installed exhibit video. Polyglot does not build an audio-guide app, trigger playback by location or publish a tour.

W3C’s transcript guidance explains the value of a readable equivalent to recorded audio, including relevant non-speech information. The Whitney’s public guides illustrate numbered stops, language choices and different kinds of text and audio material. These are useful reference patterns, not a claim that the Whitney uses Polyglot or accepts its exports.

Map the visitor label to the actual recording

Before transcription, make a stop map in a separate document. For each stop, record the number printed on the visitor label, the displayed title, the audio filename and revision, its spoken language, and the intended transcript page. Keep a stable internal identifier if display numbers can change. A filename such as final-audio.mp3 is not enough to distinguish two stops or two revisions.

Check recording and distribution permission for narration, interviews, music and included archival material. Public availability of someone else’s guide is not permission to download, translate or republish it. Use the owner-approved final audio; hold restricted or unresolved material out of the workflow.

The fictional map below catches a mismatch that a successful transcription cannot detect. The words in the wrong recording may be accurate, but they still belong to the wrong object. Fix the source pairing before reviewing text.

Fictional stop map
Stop 12 / Harbour bell
Audio: stop12-en-v03.mp3
Text: stop12-en-v03.txt
Page: /guide/harbour-bell/en

Stop 13 / Signal lamp
Assigned audio: stop12-en-v03
HOLD: wrong recording
Action: owner supplies stop13

Check each stop independently;
visitor order is not fixed.

Check the finished audio against the script

Import one approved stop recording and confirm that its spoken language is supported for transcription. Review the full result while listening. Use an approved script or object record to check names, dates and unfamiliar terms, but let the actual recording establish what was said. W3C notes that a script may need changes to match finished audio.

If recognition missed a word, correct the text. If the narrator says the wrong date, return the issue to the owner; do not silently make the transcript contradict the recording. When the owner supplies replacement audio, review that new version and update the stop map. Keep the earlier approved files identifiable until the replacement is accepted.

Add speaker identification and meaningful sound information where needed. A bell used to demonstrate a signal may carry information that speech recognition misses. Review and complete those descriptions manually in the exported transcript; Polyglot does not automatically recognize speakers or describe sounds. A transcript also does not supply an unspoken description of the object. Prepare that description separately when required.

Keep translated text and recorded languages distinct

A translated transcript of English narration is different from the transcript of a separately recorded Spanish narration. A translated text track can accompany the English source if clearly labeled. It must not be presented as a checked transcript of Spanish audio that nobody has reviewed.

When separate language recordings exist, pair each with its own reviewed text and revision. A narrator may shorten an introduction or adapt a reference, so the files need not share the same duration or cue timing. Transcribe each supported recording separately. Translation targets and speech-recognition languages are different capabilities; verify support before starting.

Have a competent language reviewer check object names, quoted speech, directions and references to other stops. Keep visitor labels and transcript headings consistent across the guide. Translating a subtitle file preserves a relationship to its source recording; it does not create a new spoken narration or approve the meaning.

Publish readable text, then test every route to it

Export the reviewed plain text and format it in the guide’s own publishing system. Use a clear stop number and title, readable paragraphs and relevant speaker labels. Put editorial introductions, current visitor information and visual descriptions in clearly labeled sections so readers can distinguish them from the recorded words. W3C recommends useful structure and making transcripts easy to find from their media.

For a guide that already supplies an approved script matching the audio, formatting and publishing that script may be sufficient. Polyglot is useful when you need to transcribe an existing recording, reconcile a changed take or review translated text locally. It does not replace a museum’s publishing platform or accessibility review.

Open each stop through the actual visitor entry point: the guide index, printed link or QR code if used, and language selector. Confirm the number, title, audio version and text all match. Test stops out of order, return to the index and check the transcript on a narrow phone screen. The reader should be able to reach the text without first listening to the recording.

If the receiving player supports synchronized text, export its accepted SRT or WebVTT file and test seeking and timing against that exact stop recording. Do not assume a plain-text upload creates timed highlighting or that every audio player supports subtitle files. Keep the readable transcript available through the guide’s approved route.

Record the final page links, audio and text revisions, language reviewer and remaining issues in the external stop map. Verify replacements at the public destination before calling the handoff complete. Local processing does not publish files, set access permissions or certify the finished guide’s accessibility.

Keep exploring.