cochl-sense-sound-event-detection — Classification AI Model
Cochl.Sense listens to a recording about once a second and tags what it hears from more than 100 sounds across eight groups, from emergencies and human sounds to animals, music and household noise. Runs of the same tag are joined into timed events you can cut a video on. Runs on Cochl's side against your own project key.
What it does
Sound Events (Cochl) scans a recording of up to an hour and reports which sounds occur and when. Cochl.Sense recognises more than 100 tags in eight groups, including emergency sounds such as glass breaking, screams and sirens, human sounds, animals, music, vehicles and household noise.
Cochl tags 2-second windows that advance 1 second at a time. This node keeps the windows above your confidence threshold and joins runs of the same tag into one event, so the output is a list of timed segments that Video Trim and the EDL nodes take directly.
When to use it
- Finding the moments with glass breaking, alarms or screams in long security or dashcam audio
- Cutting a highlight reel at applause, cheering or laughter
- Logging where music, speech and background noise start and stop in a recording
- Tagging pet and animal sounds in home video for search or editing
Details
- CategoryClassification
- Year2026
- LicenseCustom (Cochl Terms of Service)
Compliance & Provenance
- ProviderCochl · Specialized
- EU AI Act RiskMinimal Risk
- Art. 50 TransparencyNot applicable
Inputs & Outputs
- SoundInputsound
Recording to scan for sound events (up to one hour)
- Sound eventsOutputsegments
Timed sound events (feeds the Video Trim node)
Parameters
- cochl_min_confidencefloatdefault 0.5range 0.05–0.95
Minimum confidence for a sound to count (0.05-0.95)
Windows whose Cochl confidence is below this are ignored. Default 0.5. Lower to catch faint or distant sounds at the cost of more false hits; raise to keep only clear ones.
- cochl_merge_gap_secfloatdefault 1range 0–5advanced
Join two runs of the same sound when the gap is shorter than this (seconds)
Two runs of the same tag separated by less than this many seconds become one event. Default 1.0. 0 keeps every run separate; raise it to treat an intermittent sound (barking, a stuttering alarm) as one event.
Limitations
- Up to one hour of audio per run. mp3, wav, flac and ogg are sent as they are; other formats are converted first.
- Cochl analyses 2-second windows that advance 1 second at a time, so event start and end times are accurate to about one second, not to the frame.
- Only the tags Cochl.Sense knows are detected; it does not describe sounds outside its list.
- Runs on your own Cochl project key. Without one configured in the workspace the node cannot run, and usage is billed by Cochl to that project.
Tags
Alternatives in Audio Understanding
- Audio Insights (Cochl)
Cochl.Sense audio insights: a summary of what a recording contains - setting, notable sounds, language and what is said. Needs Cochl key.
- Speech Analysis (Cochl)
Cochl.Sense speech analysis: speaker-separated transcript in 100 languages, naming voices registered in your Cochl dashboard. Needs Cochl key.
- Speech Recognition
Recognize speech in a video or audio file and convert it to text.
- Speech Recognition (Clips)
Recognize speech in each clip of a video list and convert it to timed transcripts.
- Transcribe (Gemini)
Google Gemini 3.5 Transcribe: speech to text in 85+ languages, with optional speaker labels and custom vocabulary. Needs API key.