samplesPerChunk

samplesPerChunk is a parameter used in digital signal processing and data streaming to specify the number of samples that are grouped into a single processing block, or chunk. The value influences memory usage, latency, and throughput because operations are often performed on fixed-size blocks rather than on arbitrary-length streams.

In audio applications, chunking aligns with Fourier transform sizes, buffering, and audio I/O. Values are often powers of two (for FFT efficiency) such as 256, 512, 1024, or 2048, but the optimal size depends on the algorithm and system constraints.

Larger chunks reduce per-sample overhead and improve vectorization and cache performance but increase latency and buffering requirements. Smaller chunks lower latency and can improve responsiveness in real-time processing but increase overhead and may reduce processing efficiency.

In practice, samplesPerChunk is chosen based on the processing stage: input buffering, feature extraction (e.g., STFT or Mel-spectrogram), or streaming encoders. Some frameworks require a fixed chunk size, others permit dynamic resizing or padding to a target length to match downstream blocks.

Related concepts include frame size, window length, block size, and batch size. The term is not standardized across libraries and may be called blockSize, frameSize, or chunkSize depending on context.

Example contexts include audio processing libraries that process 1024 samples per frame for STFT, or streaming codecs that process 2048-sample blocks for encoding.