[aguirreSL]

Adds four utilities that the ITA-Toolbox has (ita_multitone, ita_xfade_spk, ita_cutImpulseTrain and ita_velocity_from_pressure_gradient):

1. pytta.generate.multitone(freqs, amplitude=1, phase=0, samplingRate=None, fftDegree=None): a sum of sines at the frequencies freqs rounded to the nearest DFT bin, so that every tone has a whole number of periods. The logarithmic and fixed factor spacings of ita_multitone are freqs = numpy.geomspace(...), and the phase can be one for all tones or one per tone.
2. crossfade_spectra(signal_low, signal_high, freqs): the spectrum of signal_low below freqs[0] and of signal_high above freqs[1], with raised cosine weights between them that add up to 1, as ita_xfade_spk.
3. cut_impulse_train(signal, threshold=0.7, length=1, start_before=0.004, ref_channel=0): the impulses of an impulse train, detected on one channel with the threshold threshold * L_max + (1 - threshold) * L_rms of ita_cutImpulseTrain and cut at the same samples in every channel. It returns one SignalObj per channel, whose channels are the impulses, and the 0-based start samples.
4. velocity_from_pressure_gradient(signal, distance, rho=1.2, c=343, normalized=False): the particle velocity midway between two microphones from Euler's equation, v = (p0 - p1) / (jωρd), times ρc if normalized.

The ITA code was read for its semantics only and nothing was copied. Indices are 0-based.

The tests in tests/test_dsp_misc.py print the measured differences:

1. The DFT of a multitone is A N / 2 at the rounded bins and zero elsewhere (1.1e-9 and 2.7e-9 of A N / 2).
2. The crossfade keeps the spectrum of the low signal below freqs[0] and of the high one above freqs[1] (2.2e-8), and a signal crossfaded with itself comes back exactly.
3. cut_impulse_train gives the start samples and the cuts of ita_cutImpulseTrain exactly, on a train of four decaying bursts in two channels from tests/golden/make_cut_impulse_train_ita.m (MATLAB R2026b).
4. For a plane wave the velocity is p / (ρc) at the midpoint times sin(kd/2) / (kd/2), the error of the finite difference (5.2e-8 of p / (ρc)).

multitone with a NumPy array as freqs, as in its docstring, needs #59, since SignalObj compares the first local variable of the calling function with the new object; a list works without it. Checked with Python 3.12, numpy 1.26 and scipy 1.17. The 7 failures of development (3 in test_h5io.py, 4 in test_pyttasave.py) are the same with and without this change.

Co-Authored-By: Claude Fable 5.1 noreply@anthropic.com