Alarms are designed to be unmistakable, which makes them the most constrained sounds in a library: they have to cut through noise, repeat indefinitely, and be identified before anyone thinks about them. That helps when prompting, because only a few features matter: the pattern, the pitch, and whether the listener is inside the building or outside it.
Every clip shows the exact prompt that produced it. The render script reads its prompts from the same file as this page. Copy one, change a word and generate your own.
building fire alarm sounding, harsh repeating electronic tone echoing down a corridor, seamless loop
police siren wailing and passing the listener at speed, doppler shift through the pass, city street
air raid siren winding up to a long mournful wail and falling away, distant across a city, outdoor
submarine dive klaxon, harsh two-tone mechanical horn repeating, enclosed metal interior, seamless loop
car alarm going off in a quiet street at night, repeating cycle of chirps and horn blasts, echoing
heavy vehicle reversing beeper, single high electronic beep repeating steadily, outdoor site, seamless loop
smoke detector low battery chirp, single piercing beep every thirty seconds, quiet house at night
spacecraft red alert klaxon, urgent synthetic two-tone alarm over a low warning drone, metal interior, seamless loop
ambulance siren wailing several streets away, muffled and echoing between buildings, night city
industrial warning horn blasting three times before machinery starts, huge factory reverb
These models respond to physical description. Name the material, the action and the space.
building fire alarm sounding, harsh repeating electronic tone echoing down a corridor, seamless loop
The corridor places it in a building. Alarms are almost always heard through a lot of hard architecture.
police siren wailing and passing the listener at speed, doppler shift through the pass, city street
A siren that does not move sounds parked. Asking for the pass and the doppler in one render gives you the whole event.
ambulance siren wailing several streets away, muffled and echoing between buildings, night city
"Between buildings" produces the slapping urban reflections that place a siren in a city instead of open air.
air raid siren winding up to a long mournful wail and falling away, distant across a city, outdoor
"Winding up" names the spin-up of a rotating mechanical siren, which is what distinguishes it from an electronic tone.
An alarm is recognised by its rhythm more than its timbre: the two-tone European siren, the American wail, the single repeating reversing beep, the three-blast industrial warning. Describe the pattern and the model gets the character right even if the tone is approximate. Describe only the tone and you tend to get a generic alarm that matches nothing in particular.
An alarm heard inside the building it protects is loud, hard and close, with corridor reflections. Heard from outside, it is muffled, distant and mixed with the street. These are different assets and filtering does not turn one into the other. Because the alarm itself is so repetitive, the space carries most of the information about where the scene is, which makes it the most valuable clause in the prompt.
An alarm scene usually has three parts: the moment it starts, the sustained state, and the moment it stops or is silenced. Make the sustained state a steady seamless loop you can run under the whole scene, and make the start and the stop one-shots. A single clip with all three cannot loop or be extended, which is what makes stock alarm clips awkward.
The models run on your machine, which changes the cost, the privacy and what you can automate.
Weights are fetched once. After that, generation, the library and the API all work with the network off. Everything lives in one SQLite file and one folder of WAVs under ~/.soundbox, and nothing is uploaded.
Soundbox runs an MCP server in the same process, with tools for generating and listing sounds. Generated files come back as absolute paths, so an agent can copy them into a project.
Every model's terms are shown before you download it, and non-commercial weights need an explicit confirmation. Soundbox is MIT. The licence of the model that made a sound governs what you may do with it.
claude mcp add --transport http soundbox http://127.0.0.1:17494/mcpcurl -X POST http://127.0.0.1:17494/api/generate \
-H 'Content-Type: application/json' \
-d '{"model":"audiocpp-stable-audio-3-sfx","prompt":"8-bit coin pickup blip","duration_s":1}'