Ask a GIS team what holds up a public-facing story map and you will hear about data licensing, basemap credits, or a reviewer who wants one more round on the legend. Almost nobody says audio. Then the narrated tour goes to the communications office, and it comes back with three notes about the sound and none about the map.
Our group builds four or five of these a year for a county planning office. The mapping side is solved work. The audio side has cost us more calendar time than any layer we have ever published, and almost all of it was avoidable. Two quick measurements, plus the vocal remover step described below, now cover what used to eat a week.
Here is how we handle it now, and what it replaced.
The old way: borrow music, hope nobody asks
For a long time the background track came from wherever. Somebody’s library subscription, a clip from a previous project, once a piece a colleague’s nephew had made. It sounded fine. It was also the single item in the whole deliverable that nobody could produce paperwork for.
That caught up with us on a corridor study that got picked up by a regional outlet. Legal asked one question — where did the music come from — and the honest answer took two days to reconstruct and was still incomplete.
The replacement rule is short. Every piece of audio in a published map has a traceable origin, or it does not ship. Not a preference. It is the gate.
The new way, and what each step is actually for
Three pieces of audio go into a typical narrated tour: a spoken track, a music bed, and the transitions between chapters. Each one has a different failure mode.
The spoken track. Field narration gets recorded wherever the subject happens to be. A county engineer talking through a culvert replacement is standing next to a road. A museum curator explaining a historic district is standing under a ceiling speaker playing something. That second case wrecked more of our schedules than any other single problem, because the recording is otherwise good and the music underneath makes it unusable for captioning and impossible to reuse in a second language.
A vocal remover splits that recording into two sides and hands back both. Most people meet this kind of tool as a karaoke thing. That framing is why it took our team an embarrassingly long time to notice that the half nobody talks about — the isolated voice — is the half a mapping shop actually needs, and that a vocal remover is therefore a captioning tool as much as a music one. We keep the voice. The music that was bleeding in goes with the other side, which is exactly where we want it: out of the caption pipeline and out of the rights conversation.
Two limits are worth knowing before any schedule depends on this. Hard, reverberant rooms give a vocal remover trouble, because a voice in a stone corridor keeps ringing after the word ends, and that ring is spread across the whole recording — some of it leaves with the music. And a chorus of overlapping speakers — a public meeting, say — is not what the separation is designed for. One speaker, one room, and the result is clean enough to caption directly. The rule on our team is that somebody listens to the voice side end to end before a single caption line gets written against it, which on a three-minute chapter costs three minutes.
The music bed. We generate our motifs now rather than borrowing them, and the same browser tab that runs the vocal remover handles it. The useful part is not that it is free. The useful part is that the origin is documented from the first minute, which satisfies the gate above without anybody reconstructing anything two days later.
The transitions. This is the part nobody warns you about. A twenty-second chapter change has to land on the map animation, and a bed that runs at the wrong tempo will fight the pan and zoom no matter how good it sounds on its own.
Two numbers we check before anything gets cut
Both of these take under a minute and both have saved a re-edit.
Length, against the chapter. Motifs almost never arrive at the exact length a chapter needs. Rather than crossfading badly, we open the file in a browser-based MIDI editor and move the notes. It accepts standard .mid and .midi files up to two megabytes, shows the part on a piano roll — pitch on the vertical axis, time on the horizontal — and lets you shift a note, change how long it holds, or delete one that was never wanted. Tempo and instrument assignment are adjustable in the same place.
Be clear about what that tool is and is not. It edits musical instructions, not recorded sound. There are no effects, no automation, and it is not standing in for a proper production setup. For our purposes that limitation is the point: the whole job is moving eleven notes so a phrase resolves where the map stops moving.
Tempo, against the animation. Once the length is right, drop the bed into a key and BPM finder and read the tempo back. The analysis runs locally in the browser, which matters when the audio belongs to a client and you would rather it not leave the machine. Accepted files run to twenty megabytes and ten minutes, which covers any bed we have ever used. Treat the reading as an estimate rather than a fact — ambiguous harmony can land on a neighbouring key, and the tool will withhold a result outright when the beat evidence is weak. That withholding is a feature. A blank answer is more honest than a confident wrong one.
Then do the arithmetic against your animation. A four-second pan at 120 beats per minute is eight beats. If the bed sits at 92, that same pan is roughly six beats and lands mid-phrase, and viewers feel it as sloppiness without being able to say why.
What this changed on the schedule
The audio round used to be an unknown that appeared after the map was signed off. Now it runs alongside the cartography and finishes first, because none of the four steps takes more than a few minutes and none of them depends on somebody else’s calendar. Our project template literally has a checkbox reading vocal remover pass, sitting between metadata review and accessibility review, and nobody on the team finds that odd any more.
Three things worth taking away, whatever software your shop standardises on:
– The origin of every audio file is a publishing gate, the same as data provenance. Treat it that way and the legal question never becomes a two-day archaeology project.
– Field narration goes through a vocal remover before captioning, never after. Captions built on a mixed track carry misheard place names and misheard figures into a document with your agency’s name on it.
– Check the tempo against the animation before the edit, not during review. It is a sixty-second check that prevents a rebuild.
Where this approach stops working
If the deliverable is a documentary rather than a map tour, hand it to somebody whose job is sound. Separation is the whole of what the tool does; a vocal remover has never once been a mixing desk. The workflow above is built for short narrated chapters with a bed underneath, which is what a story map is. Push it toward layered scoring, dialogue editing, or anything needing a real mix and it will not hold. You will have spent a week discovering what an audio contractor would have told you in an email.
We publish maps. The audio only has to stop being the reason they are late.