Hive's AI-Generated Audio Detection API takes an audio file as input and determines whether or not that audio is AI-generated. The model has one head. This head contains two classes: ai_generated and not_ai_generated. Each class has a confidence score, which ranges from 0.0 - 1.0 and indicates how sure the model is of that class prediction.
The model breaks every input audio clip into 20 second chunks and provides a classification for each of these smaller segments. If the audio clip is less than 20 seconds in total length, it will only include one classification. Each classification also includes the timestamp of the start of that chunk.
{
"status": [
{
"status": {
"code": "0",
"message": "SUCCESS"
},
"_version": 2,
"response": {
"output": [
{
"classes": [
{
"score": 0.9996376037597656,
"class": "not_ai_generated"
},
{
"score": 0.000362391525413841,
"class": "ai_generated"
}
],
"time": 0
},
{
"classes": [
{
"score": 0.999784529209137,
"class": "not_ai_generated"
},
{
"score": 0.00021547614596784115,
"class": "ai_generated"
}
],
"time": 20
},
{
"classes": [
{
"score": 0.9993382096290588,
"class": "not_ai_generated"
},
{
"score": 0.0006617384497076273,
"class": "ai_generated"
}
],
"time": 40
},
{
"classes": [
{
"score": 0.9996871948242188,
"class": "not_ai_generated"
},
{
"score": 0.00031287287129089236,
"class": "ai_generated"
}
],
"time": 60
},
{
"classes": [
{
"score": 0.9999006986618042,
"class": "not_ai_generated"
},
{
"score": 9.92821587715298e-05,
"class": "ai_generated"
}
],
"time": 80
}
]
}
}
]
}