Description:
AI object AO is implemented for running AI/ML applications on target platforms. AI Object performs transformation like STFT(short time fourier transform) on incoming data and performs AI/ML model inference. The model output is sent through the control outputs, which are streamable. Model output will be classification.
By default AI Object does not have any configuration, it will be an empty object and needs to be configured using valid AI/ML model (or session). Valid AI/ML session can be selected by performing below actions.
1- Right click in AI_object -> Open Audio Configuration Panel -> Select one of the available AI sessions -> Apply
2- Save signal flow
2- Save signal flow
After selecting valid AI session AI Object will updated with new configuration. Figure below is AI Object is configured to have 2 control output.
Valid AI model can be imported, exported or saved into GTT using AI Module feature in GTT (AI module is GTT feature like IVP).
AI model will have details like features used and ML algorithm (like CNN, DNN etc.).
Features supported include:
- STFT (Short time fourier transform)
- MFCC (Mel Frequency Cepstral Coefficient)
- MelSpectogram
Configuration:
Audio in: This object supports one audio input
Control outputs: The number of control outputs is selected from additional configuration parameter. Control outputs contains model outputs. Control outputs are stremable.
Additional configuration parameter: This AO has one additional configuration parameter “AI_Metadata“. Additional configuration value is set from configuration when ML model is selected by AO. This additional configuration value decides number of control outputs.