AI Music Detection

This FAQ explains how ACRCloud's AI Music Detection feature works for audio content, including prediction thresholds, output interpretation, supported source categories, and performance notes.

---

What does AI Music Detection do?

ACRCloud's AI Music Detection feature helps customers assess whether submitted audio content is more likely to be AI-generated or human-created.

It is designed to support content inspection and risk assessment workflows. Detection results should be interpreted as model outputs based on statistical patterns in the audio, rather than as definitive proof of origin.

---

How does AI Music Detection work?

ACRCloud uses advanced, state-of-the-art detection technology to train specialized models for identifying whether audio content is AI-generated or human-created.

For each inspected audio file, the system returns:

• a binary prediction result

• an ai_probability score

• source_probabilities for supported source categories

Both ai_probability and source_probabilities are generated directly by the model and are not manually adjusted or post-processed by human operators.

---

What threshold determines ai_generated vs. human?

The current threshold is 50%.

• If ai_probability >= 50%, the output is ai_generated

• If ai_probability < 50%, the output is human

---

Is AI Detection generally available?

Yes. AI Detection is generally available and has been officially released.

---

What performance data is available?

The following figures are based on ACRCloud's internal evaluation dataset for the current model version.

Binary Confusion Matrix (Human vs. AI)

• Precision (AI): 99.98%

• Recall (AI): 99.94%

• False Positive Rate: 0.06% of Human

• False Negative Rate: 0.06% of AI

Per-Source Accuracy

• human: 99.91%

• suno: 99.86%

• udio: 99.93%

• sonauto: 98.81%

• mureka: 99.99%

• rdiffusion: 98.00%

These results reflect performance on internal test data and are provided for reference only. Actual performance may vary depending on audio quality, transformations, compression, mixing, editing, and real-world usage conditions.

---

What source categories are currently supported?

The currently supported source categories exposed in the platform include:

• human

• suno

• udio

• sonauto

• mureka

• rdiffusion

We are actively expanding supported source coverage and are working to include additional model sources such as ElevenLabs, MiniMax, and Seed-Music.

Supported source coverage may continue to evolve over time as generative audio models and market usage change.

---

How should ai_probability and source_probabilities be interpreted?

We recommend that customers interpret the output as follows:

• Use the binary prediction for straightforward workflow decisions

• Use ai_probability as a confidence reference

• Use source_probabilities as supporting signals for source attribution analysis

• Consider the result together with other contextual or operational signals where appropriate

Both ai_probability and source_probabilities are direct outputs from the model itself.

Detection results should not be treated as the sole basis for legal, enforcement, or other high-impact decisions without additional review.

---

Notes

• Detection results are probabilistic model outputs, not definitive proof of origin.

• Performance may vary across different content types and real-world conditions.

• Supported source categories may be updated over time as the model evolves.

• Third-party model or product names are used only to describe detection coverage and do not imply partnership, endorsement, or affiliation.

---

Suggested Disclaimer

The AI Detection feature provides probabilistic model outputs based on patterns identified in submitted audio content. Results are intended to assist review and inspection workflows and should not be interpreted as definitive proof of content origin. Performance metrics shown in this document are based on internal evaluation data and may vary in real-world use cases.