modusa 0.4.30__tar.gz → 0.4.32__tar.gz

{modusa-0.4.30 → modusa-0.4.32}/PKG-INFO

@@ -1,6 +1,6 @@
 Metadata-Version: 2.1
 Name: modusa
-Version: 0.4.30
+Version: 0.4.32
 Summary: A modular signal analysis python library.
 Author-Email: Ankit Anand <ankit0.anand0@gmail.com>
 License: MIT

{modusa-0.4.30 → modusa-0.4.32}/pyproject.toml

@@ -16,7 +16,7 @@ dependencies = [
 ]
 requires-python = ">=3.11"
 readme = "README.md"
-version = "0.4.30"
+version = "0.4.32"
 
 [project.license]
 text = "MIT"

{modusa-0.4.30 → modusa-0.4.32}/src/modusa/__init__.py

@@ -9,6 +9,9 @@ from modusa.tools import load_ann, save_ann
 from modusa.tools import dist_plot, hill_plot, plot, fig
 
 # Synthsizing related
-from modusa.tools import synth_f0
+from modusa.tools import synth_f0, synth_clicks
 
-__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__ = "0.4.32" # 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-0.4.32}/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, synth_clicks
+
+# Audio features
+from .audio_stft import stft

modusa-0.4.32/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

modusa-0.4.32/src/modusa/tools/synth.py

@@ -0,0 +1,112 @@
+#!/usr/bin/env python3
+
+#---------------------------------
+# Author: Ankit Anand
+# Date: 22/10/25
+# Email: ankit0.anand0@gmail.com
+#---------------------------------
+
+import numpy as np
+
+def synth_f0(f0, f0t, sr, nharm=0):
+	"""
+	Synthesize f0 contour so that you can
+	hear it back.
+
+	Parameters
+	----------
+	f0: ndarray
+		- Fundamental frequency (f0) contour in Hz.
+	f0t: ndarray
+		- Timestamps in seconds
+	sr: int
+		- Sampling rate in Hz for the synthesized audio.
+	nharm: int
+		- Number of harmonics
+		- Default: 0 => Only fundamental frequency (No harmonics)
+
+	Returns
+	-------
+	ndarray
+		- Syntesized audio.
+	sr
+		- Sampling rate of the synthesized audio
+	"""
+	
+	# Create new time axis
+	t = np.arange(0, f0t[-1], 1 / sr)
+	
+	# Interpolate the f0 to match the sampling time points
+	f0_interp = np.interp(t, f0t, f0)
+	
+	# Compute phase by integrating frequency over time
+	phase = 2 * np.pi * np.cumsum(f0_interp) / sr
+	
+	# Start with fundamental
+	y = np.sin(phase)
+	
+	# Add harmonics if requested
+	for n in range(2, nharm + 2):  # from 2nd to (nharm+1)th harmonic
+		y += np.sin(n * phase) / n**2  # dividing by n to reduce harmonic amplitude
+	
+	# Normalize output to avoid clipping
+	y /= np.max(np.abs(y))
+	
+	return y, sr
+
+def synth_clicks(event_times, sr, freq=500, nharm=4, click_duration_ms=5):
+	"""
+	Generate a train of short sine wave clicks with harmonics at specified event times.
+
+	Parameters
+	----------
+	event_times : array-like
+		- Times of events in seconds where clicks should be placed.
+	sr : int
+		- Sampling rate in Hz.
+	freq : float, optional
+		- Fundamental frequency of the sine click in Hz. Default is 500 Hz.
+	nharm : int | None
+		- Number of harmonics to include (including fundamental). Default is 4.
+	click_duration_ms : float | None
+		- Duration of each click in milliseconds. Default is 5 ms.
+	
+	Returns
+	-------
+	np.ndarray
+		- Audio signal with sine wave clicks (with harmonics) at event times.
+	int
+		- Sampling rate of the generated click audio.
+	"""
+	
+	n_samples = int(np.ceil(sr * event_times[-1]))
+	y = np.zeros(n_samples, dtype=np.float32)
+	
+	# Single click length
+	click_len = int(sr * click_duration_ms / 1000)
+	if click_len < 1:
+		click_len = 1
+		
+	t = np.arange(click_len) / sr
+	window = np.hanning(click_len)
+	
+	# Generate harmonic sine click
+	sine_click = np.zeros(click_len)
+	for n in range(1, nharm+2):
+		sine_click += (1 / n**2) * np.sin(2 * np.pi * freq * n * t)
+		
+	# Apply window
+	sine_click = sine_click * window**2
+	
+	for event_time in event_times:
+		start_sample = int(event_time * sr)
+		end_sample = start_sample + click_len
+		if end_sample > n_samples:
+			end_sample = n_samples
+		y[start_sample:end_sample] += sine_click[:end_sample - start_sample]
+		
+	# Normalize to avoid clipping if clicks overlap
+	y /= np.max(np.abs(y))
+	
+	return y, sr
+

modusa-0.4.30/src/modusa/tools/synth.py

@@ -1,55 +0,0 @@
-#!/usr/bin/env python3
-
-#---------------------------------
-# Author: Ankit Anand
-# Date: 22/10/25
-# Email: ankit0.anand0@gmail.com
-#---------------------------------
-
-import numpy as np
-
-def synth_f0(f0, f0t, sr, nharm=0):
-	"""
-	Synthesize f0 contour so that you can
-	hear it back.
-
-	Parameters
-	----------
-	f0: ndarray
-		- Fundamental frequency (f0) contour in Hz.
-	f0t: ndarray
-		- Timestamps in seconds
-	sr: int
-		- Sampling rate in Hz for the synthesized audio.
-	nharm: int
-		- Number of harmonics
-		- Default: 0 => Only fundamental frequency (No harmonics)
-
-	Returns
-	-------
-	ndarray
-		- Syntesized audio.
-	sr
-		- Sampling rate of the synthesized audio
-	"""
-	
-	# Create new time axis
-	t = np.arange(0, f0t[-1], 1 / sr)
-	
-	# Interpolate the f0 to match the sampling time points
-	f0_interp = np.interp(t, f0t, f0)
-	
-	# Compute phase by integrating frequency over time
-	phase = 2 * np.pi * np.cumsum(f0_interp) / sr
-	
-	# Start with fundamental
-	y = np.sin(phase)
-	
-	# Add harmonics if requested
-	for n in range(2, nharm + 2):  # from 2nd to (nharm+1)th harmonic
-		y += np.sin(n * phase) / n**2  # dividing by n to reduce harmonic amplitude
-	
-	# Normalize output to avoid clipping
-	y /= np.max(np.abs(y))
-	
-	return y, sr