modusa 0.4.30__tar.gz → 0.4.31__tar.gz

{modusa-0.4.30 → modusa-0.4.31}/PKG-INFO

@@ -1,6 +1,6 @@
 Metadata-Version: 2.1
 Name: modusa
-Version: 0.4.30
+Version: 0.4.31
 Summary: A modular signal analysis python library.
 Author-Email: Ankit Anand <ankit0.anand0@gmail.com>
 License: MIT

{modusa-0.4.30 → modusa-0.4.31}/pyproject.toml

@@ -16,7 +16,7 @@ dependencies = [
 ]
 requires-python = ">=3.11"
 readme = "README.md"
-version = "0.4.30"
+version = "0.4.31"
 
 [project.license]
 text = "MIT"

{modusa-0.4.30 → modusa-0.4.31}/src/modusa/__init__.py

@@ -11,4 +11,7 @@ from modusa.tools import dist_plot, hill_plot, plot, fig
 # Synthsizing related
 from modusa.tools import synth_f0
 
-__version__ = "0.4.30" # This is dynamically used by the documentation, and pyproject.toml; Only need to change it here; rest gets taken care of.
+# Audio features related
+from modusa.tools import stft
+
+__version__ = "0.4.31" # This is dynamically used by the documentation, and pyproject.toml; Only need to change it here; rest gets taken care of.

{modusa-0.4.30 → modusa-0.4.31}/src/modusa/tools/__init__.py

@@ -16,4 +16,7 @@ from .plotter import Fig as fig
 from .plotter import dist_plot, hill_plot, plot
 
 # Synthesizing related
-from .synth import synth_f0
+from .synth import synth_f0
+
+# Audio features
+from .audio_stft import stft

modusa-0.4.31/src/modusa/tools/audio_stft.py

@@ -0,0 +1,72 @@
+#!/usr/bin/env python3
+
+#---------------------------------
+# Author: Ankit Anand
+# Date: 23/10/25
+# Email: ankit0.anand0@gmail.com
+#---------------------------------
+
+import numpy as np
+
+def stft(y, sr, winlen=None, hoplen=None, gamma=None):
+	"""
+	Compute spectrogram with just numpy.
+
+	Parameters
+	----------
+	y: ndarray
+		- Audio signal.
+	sr: int
+		- Sampling rate of the audio signal.
+	winlen: int
+		- Window length in samples.
+		- Default: None => set at 0.064 sec
+	hoplen: int
+		- Hop length in samples.
+		- Default: None => set at one-forth of winlen
+	gamma: int | None
+		- Log compression factor.
+		- Add contrast to the plot.
+
+	Returns
+	-------
+	ndarray:
+		- Spectrogram matrix, complex is gamma is None else real
+	ndarray:
+		- Frequency bins in Hz.
+	ndarray:
+		- Timeframes in sec.
+	"""
+	
+	if winlen is None:
+		winlen = 2 ** int(np.log2(0.064 * sr))
+	if hoplen is None:
+		hoplen = int(winlen * 0.25)
+		
+	# Estimating the shape of the S matrix
+	M = int(np.ceil(winlen / 2))
+	N = int(np.ceil((y.size - winlen) / hoplen))
+	
+	# Initialize the S matrix
+	S = np.empty((M, N), dtype=np.complex64) # M X N => freq bin X time frame
+	
+	# We will need a hann window
+	hann = np.hanning(winlen)
+	
+	# Get the frames
+	frames = np.lib.stride_tricks.sliding_window_view(y, window_shape=winlen)[::hoplen] # frame X chunk
+	
+	# Apply window to the frame
+	frames_windowed = frames * hann # frame X chunk
+	
+	# Compute fft for each frame
+	S = np.fft.rfft(frames_windowed, n=winlen, axis=1).T  # transpose to match shape (freq_bins, frame)
+	
+	# Magnitude spectrogram
+	Sf = np.fft.rfftfreq(winlen, d=1/sr) # Frequency bins (Hz)
+	St = np.arange(N) * hoplen / sr # Time bins (sec)
+	
+	if gamma is not None:
+		S = np.log1p(gamma * np.abs(S))
+		
+	return S, Sf, St