In film, a heartbeat signals subjectivity: when you hear one, you are inside a character instead of watching them. Nobody hears a heartbeat externally, so its presence means the camera has moved inward. Getting it right is mostly tempo and body. Too fast and it sounds like a drum; with no chest resonance it sounds like a kick sample.
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.
human heartbeat at rest, slow double thump about once per second, deep chest resonance, very close and internal, seamless loop
heartbeat accelerating steadily from calm to racing, deep internal thumps growing faster and harder
heartbeat racing in panic, fast hard double thumps with blood rushing in the ears, internal perspective, seamless loop
single deep heartbeat thump with a long low chest resonance, internal perspective, dry
hospital heart monitor beeping steadily then dropping to a continuous flat tone, clinical room, close
hospital heart monitor beeping steadily once per second, thin electronic tone, quiet ward, seamless loop
person holding their breath then releasing it in a long shaky exhale, very close mic, quiet
high pitched tinnitus ringing after an explosion, thin sine tone with all other sound muffled and distant
heartbeat heard underwater, muffled low thumps with water pressure and no high frequencies, submerged
heartbeat slowing gradually and weakening until it stops, deep internal thumps spacing out, dark and quiet
These models respond to physical description. Name the material, the action and the space.
human heartbeat at rest, slow double thump about once per second, deep chest resonance, very close and internal, seamless loop
"Double thump" is the anatomically correct shape: a heart makes two sounds per beat, and a single thump sounds like a drum.
heartbeat accelerating steadily from calm to racing, deep internal thumps growing faster and harder
Writing the acceleration into one render keeps the tempo curve smooth. Automating it afterwards produces audible steps.
high pitched tinnitus ringing after an explosion, thin sine tone with all other sound muffled and distant
Naming what happens to everything else (muffled, distant) creates the subjective perspective. The tone alone is only a tone.
hospital heart monitor beeping steadily once per second, thin electronic tone, quiet ward, seamless loop
The monitor carries the same information from outside the body, and cutting between the two is a cheap, effective perspective change.
A resting heart is around one beat per second, a frightened one closer to two, and a racing one faster still, where it starts to lose its double-thump definition. That covers the whole emotional range without any other parameter. Prompt the rate ("about once per second", "fast", "racing"), and if a scene needs a change, render it as one continuous clip. Crossfading between two loops never quite hides the join.
A heartbeat without low-frequency body is a kick drum, and audiences hear the difference even if they cannot name it. "Deep chest resonance" and "internal perspective" reliably produce it, and they matter more than volume: a quiet heartbeat with real body sits under dialogue well, while a loud thin one fights everything. If yours is not working, add body before level.
The heartbeat alone does not put an audience inside a character; the contrast with the exterior does. Subjective sequences muffle the exterior, roll off its top end, and often add tinnitus or held breath, and the heartbeat sits in the space that clears. Generate the interior layer here and treat the exterior as a separate decision. If everything else stays normal, the heartbeat sounds like score.
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}'