Struct anira::ProcessingSpec¶

struct ProcessingSpec¶

Specification for preprocessing and postprocessing parameters.

The ProcessingSpec struct defines the processing pipeline configuration for transforming data between the host application and neural network inference.

Streamable vs Non-Streamable Tensors:

Streamable tensors: Time-varying data (e.g., audio) that flows continuously
- Have non-zero preprocess_input_size and postprocess_output_size
- Managed through ring buffers for real-time processing
Non-streamable tensors: Static parameters or control values
- Have zero preprocess_input_size or postprocess_output_size
- Stored in thread-safe internal storage

Public Functions

ProcessingSpec() = default¶: Default constructor creating an empty processing specification.

inline ProcessingSpec(std::vector<size_t> preprocess_input_channels, std::vector<size_t> preprocess_output_channels, std::vector<size_t> preprocess_input_size, std::vector<size_t> postprocess_output_size, std::vector<size_t> internal_model_latency)¶

Constructs a complete ProcessingSpec with all parameters.

Parameters:

preprocess_input_channels – Number of input channels for each input tensor
preprocess_output_channels – Number of output channels for each output tensor
preprocess_input_size – Samples count required for preprocessing for each input tensor (0 = non-streamable)
postprocess_output_size – Samples count after the postprocessing for each output tensor (0 = non-streamable)
internal_model_latency – Internal model latency in samples for each output tensor

inline ProcessingSpec(std::vector<size_t> preprocess_input_channels, std::vector<size_t> preprocess_output_channels)¶

Constructs a minimal ProcessingSpec with only channel information.

Creates a processing specification with only input and output channel counts. Other parameters are left empty and will be computed automatically by InferenceConfig.

Parameters:

preprocess_input_channels – Number of input channels for each input tensor
preprocess_output_channels – Number of output channels for each output tensor

Constructs a ProcessingSpec with channel and size information.

Creates a processing specification with input/output channels and buffer sizes. Internal model latency defaults to zero for all tensors.

Parameters:

preprocess_input_channels – Number of input channels for each input tensor
preprocess_output_channels – Number of output channels for each output tensor
preprocess_input_size – Samples count required for preprocessing for each input tensor (0 = non-streamable)
postprocess_output_size – Samples count after the postprocessing for each output tensor (0 = non-streamable)

inline bool operator==(const ProcessingSpec &other) const¶

Equality comparison operator.

Parameters:: other – The ProcessingSpec instance to compare with
Returns:: true if all members are equal, false otherwise

inline bool operator!=(const ProcessingSpec &other) const¶

Inequality comparison operator.

Parameters:: other – The ProcessingSpec instance to compare with
Returns:: true if any members are not equal, false otherwise

Public Members

std::vector<size_t> m_preprocess_input_channels¶: Number of input channels for each input tensor.

std::vector<size_t> m_postprocess_output_channels¶: Number of output channels for each output tensor.

std::vector<size_t> m_preprocess_input_size¶: Samples count required for preprocessing for each input tensor (0 = non-streamable)

std::vector<size_t> m_postprocess_output_size¶: Samples count after the postprocessing for each output tensor (0 = non-streamable)

std::vector<size_t> m_internal_model_latency¶: Internal latency in samples for each output tensor.

std::vector<size_t> m_tensor_input_size¶: Total size (elements) of each input tensor (computed from shape)

std::vector<size_t> m_tensor_output_size¶: Total size (elements) of each output tensor (computed from shape)