import { AutoProcessorMediaNode, ProcessorNodeSettings } from "./common";
import { NamedChannelLayout, Resolution, SampleFormat, SampleRate } from "./types";
import { FrameRate } from "../types";
/**
 * @public
 * The data type of the elements in a tensor.
 */
export type EmbeddedAIElementDataType = "FLOAT" | "UINT8" | "INT8" | "UINT16" | "INT16" | "INT32" | "INT64" | "STRING" | "BOOL" | "FLOAT16" | "DOUBLE" | "UINT32" | "UINT64" | "COMPLEX64" | "COMPLEX128" | "BFLOAT16";
/**
 * @public
 * Represents a single output tensor from an inference operation.
 */
export interface EmbeddedAITensor {
    /** The name of the output tensor, as defined by the model. */
    name: string;
    /** The data type of each element in the tensor. */
    dataType: EmbeddedAIElementDataType;
    /** The shape of the tensor. */
    dimensions: number[];
    /** The raw tensor data. The interpretation of these bytes depends on the dataType. */
    data: Uint8Array;
}
/**
 * @public
 * The pixel format that the model expects.
 */
export type EmbeddedAIPixelFormat = "rgb";
/**
 * @public
 * Settings for video-input models
 */
export interface EmbeddedAIVideoInput {
    /** The input type */
    inputType: "video";
    /** The target resolution that Norsk should scale the video to before sending it to the inference engine. */
    targetResolution: Resolution;
    /** The pixel format that the model expects. Norsk will handle the conversion. */
    pixelFormat: EmbeddedAIPixelFormat;
    /** The frame rate that the model expects. Norsk will handle the conversion. */
    frameRate: FrameRate;
}
/**
 * @public
 * Settings for audio-input models
 */
export interface EmbeddedAIAudioInput {
    /** The input type */
    inputType: "audio";
    /** The channel layout */
    channelLayout: NamedChannelLayout;
    /** The sample rate */
    sampleRate: SampleRate;
    /** The sample format */
    sampleFormat: SampleFormat;
    /** How many samples per inference call */
    numSamplesPerInference: number;
}
export type EmbeddedAIInputType = EmbeddedAIVideoInput | EmbeddedAIAudioInput;
/**
 * @public
 * Base settings common to all data types
 */
interface EmbeddedAISettingsBase extends ProcessorNodeSettings<EmbeddedAINode> {
    /** The path to the .onnx model file on the Norsk server's filesystem. */
    modelPath: string;
    /** The input to the model */
    modelInput: EmbeddedAIInputType;
    /**
     * Callback invoked with the results of the AI inference for each frame.
     * @param detections An array of objects detected in the frame.
     */
    onInferenceResult: (tensors: EmbeddedAITensor[]) => void;
}
/**
 * @public
 * The union of all possible input type configurations for the EmbeddedAINode
 */
export type EmbeddedAINodeSettings = (EmbeddedAISettingsBase & {
    /** Provide the model with 8-bit unsigned integer pixel data [0, 255]. */
    dataType: "uint8";
}) | (EmbeddedAISettingsBase & {
    /** Provide the model with 32-bit float pixel data, normalized to the range [0.0, 1.0]. */
    dataType: "float32_normalized";
}) | (EmbeddedAISettingsBase & {
    /** Provide the model with 32-bit float pixel data, applying a custom scaling factor. */
    dataType: "float32_custom";
    /** The custom scale factor to multiply each pixel component by. */
    normalizationScale: number;
});
/**
 * @public
 * A media node that can run a generic AI model for tasks like object detection.
 */
export declare class EmbeddedAINode extends AutoProcessorMediaNode<"audio" | "video"> {
}
export {};
//# sourceMappingURL=embeddedAI.d.ts.map