Config¶
config
¶
Configuration dataclass for PyGaborSTM.
Default values match Bellur & Elhilali (2017).
Config
dataclass
¶
Config(use_gpu: bool = False, sample_rate: int = 16000, n_filters: int = 128, f_min: float = 180.0, octaves: float = 5.3, tau_ms: float = 8.0, frmlen_ms: float = 16.0, rates: ndarray = (lambda: copy())(), scales: ndarray = (lambda: copy())(), resolution: str = 'low', rsf_frame_size_ms: int = 500, rsf_frame_shift_ms: int = 10)
Configuration for the full PyGaborSTM pipeline.
| PARAMETER | DESCRIPTION |
|---|---|
use_gpu
|
If True, use CuPy for GPU acceleration. Silently falls back to NumPy with a warning if CuPy is unavailable.
TYPE:
|
sample_rate
|
Audio sample rate in Hz. Must match the input signal.
TYPE:
|
n_filters
|
Number of cochlear (gammatone) channels in the spectrogram stage.
TYPE:
|
f_min
|
Lowest filter center frequency in Hz.
TYPE:
|
octaves
|
Total frequency range in octaves.
TYPE:
|
tau_ms
|
Time constant (ms) for the leaky-integration stage of the spectrogram.
TYPE:
|
frmlen_ms
|
Spectrogram frame length in ms (downsampling factor after y5).
TYPE:
|
rates
|
Base set of temporal modulation rates (Hz). Used as anchor points;
TYPE:
|
scales
|
Base set of spectral modulation scales (cycles/octave). Anchor
points for
TYPE:
|
resolution
|
Density multiplier for the rate and scale grids. Each level doubles the number of kernels along both axes.
TYPE:
|
rsf_frame_size_ms
|
RSF integration window length in ms.
TYPE:
|
rsf_frame_shift_ms
|
RSF hop size in ms.
TYPE:
|