Museum audio guide content production: voiceover, translation and AI

Voice casting should match the tone of the collection, not simply the language. A contemporary art museum and a natural history museum need different narrators even in the same language. The brief should specify pace, sentence emphasis style, the required stop durations, and whether the narrator should read the audio description variant in a different register from the standard tour.

For multilingual fleets, the temptation is to use the same voice agency for all languages and let the agency source each narrator. That works for common European languages. For languages with fewer professional museum narrators available, plan extra lead time. A narrator who is wrong for the collection tone will require re-recording, which is more expensive than a longer casting process.

A recording brief for museum audio guide narration should cover: target stop duration, words per minute, pause positions (marked in the script), how proper nouns and artist names are pronounced, which words carry the principal emphasis in each sentence, whether sound effects or music beds will be mixed under the narration (so the narrator does not fill pauses they should leave empty), and who will approve the first take before the full session continues. Sending the brief in advance of the session, not at the start of it, is the single cheapest improvement most productions can make.

Remote recording is now standard practice for many language versions. A narrator records at a home or local studio, files are delivered as WAV, and the production team edits and mixes centrally. Remote recording is faster to schedule and removes travel costs, but requires the narrator to have a professional acoustic setup. For the primary language version, where a content director is likely to attend and give real-time direction, an in-studio session usually produces a more consistent result on the first pass. For additional languages where the script is fixed and the direction brief is detailed, remote delivery is often the practical choice.

Machine translation has improved substantially for common language pairs and is a reasonable starting point for translation drafts. It is not a final product for museum audio guides. Machine translation misses idiomatic phrasing, cultural context, and the difference between a sentence that scans on paper and one that lands cleanly when spoken. The appropriate workflow is machine-translated draft, reviewed and edited by a native-speaking translator with audio guide experience, then read aloud by the reviewer before passing to the narrator. This combines machine speed on the first pass with human judgement on the result.

Translated tour scripts are derivative works of the source scripts. The museum should confirm before production starts who holds the rights to the source script, whether those rights extend to translation and adaptation, and whether the translator's work belongs to the museum or the supplier. This is particularly relevant when the original script was written by a freelance curator or an external content agency. Rights disputes that surface after recording are expensive to resolve.

Most museum audio guide hardware requires mono MP3 or WAV files. Stereo files use twice the storage without a meaningful audio benefit through a single earpiece. A typical specification is 44.1 kHz sample rate, 128 kbps MP3, mono. Check the device specification before post-production begins: mixing to stereo and then converting to mono can introduce phase cancellation artefacts if the tracks were recorded with any spatial processing. For tablet-based guides that support headphone-out plus speakers, stereo files may be appropriate for specific immersive stops.

Audio files should be named consistently before upload: stop number, language code, tour variant, and version number. A naming convention such as 04_EN_standard_v2.mp3 makes it possible to confirm at a glance which stops have been delivered in which languages and whether the correct version is in the upload folder. Files with inconsistent names or with spaces in the filename cause preventable CMS import errors. Establish the naming convention at the start of production, not at the upload stage.

Quality review for museum audio guide content should happen in the gallery, not in a meeting room or post-production studio. Play each stop through the actual device and headset in the actual exhibit space, with the room at representative occupancy. Check that the narration is intelligible over ambient noise, that the stop duration fits the exhibit dwell time, that the cue-out lands at the right moment, and that any pronunciation of artist names, locations or technical terms has been confirmed by the curator. A studio playback that sounds clean can still fail in a stone-floored gallery with school groups passing through.

After a content update, the museum should confirm that the new version has reached the fleet before the next visitor session. The CMS should show sync status per device, or per device group, so operations staff can see at a glance whether any device missed the update and needs to return to a charger before handout. Checking two or three devices manually by playing the updated stop is faster than trusting that a sync completed correctly without verification.

Partial updates reduce the amount of content staff have to reopen when one stop changes. A museum should be able to change the script for a single stop, record or generate new audio, upload the file, and push the update to the fleet without replacing or re-approving content for unaffected stops. This makes it practical to correct a factual error quickly, add a temporary stop for a visiting work, or remove a stop when an exhibit is moved to storage. If the CMS requires a full tour republish every time a single file changes, routine corrections become slower.