graph LR
Inference_Orchestrator["Inference Orchestrator"]
Chunk_Processor["Chunk Processor"]
Chunk_Merger["Chunk Merger"]
Model_Optimizer["Model Optimizer"]
Offset_Computer["Offset Computer"]
Correlation_Computer["Correlation Computer"]
Inference_Orchestrator -- "prepares model using" --> Model_Optimizer
Inference_Orchestrator -- "calls" --> Chunk_Processor
Inference_Orchestrator -- "calls" --> Chunk_Merger
Chunk_Merger -- "calls" --> Offset_Computer
Offset_Computer -- "calls" --> Correlation_Computer
Architectural analysis of the resemble-enhance project's core audio enhancement functionality within the inference module, focusing on concrete, fully qualified code references (QNames).
The top-level entry point for performing audio enhancement. It orchestrates the entire inference pipeline, including model preparation, audio chunking, processing each chunk, and merging the results.
Related Classes/Methods:
Processes individual segments (chunks) of audio data using the loaded ML model (Denoiser and/or Enhancer). It's the core processing unit for audio enhancement on a per-chunk basis.
Related Classes/Methods:
Combines processed audio chunks into a single, coherent output, ensuring proper alignment and handling overlaps. It aggregates and reconstructs the full enhanced audio from processed segments.
Related Classes/Methods:
Modifies the loaded deep learning model by recursively removing weight normalization layers, which is often a pre-inference optimization step to improve performance or compatibility.
Related Classes/Methods:
Utility function to determine correct merging points for audio chunks, crucial for seamless reconstruction.
Related Classes/Methods:
Utility function for correlation calculation, likely used in determining optimal chunk overlaps or alignment.
Related Classes/Methods: