[aguirreSL]

Adds to pytta.functions nine editing utilities that the ITA-Toolbox has (ita_envelope, ita_start_IR, ita_time_shift, ita_append, ita_normalize_dat, ita_normalize_spk, ita_amplify_to, ita_zerophase, ita_randomize_phase and ita_quantize):

1. envelope(signal): the magnitude of the analytic signal, computed on twice the signal length, as ita_envelope.
2. start_ir(signal, threshold=20): the 0-based start sample of each channel of an impulse response, by ISO 3382-1:2009, A.3. It is the last sample before the peak that lies 20 dB below it, moved back to the last such sample before an earlier one 6 dB above that level (an oscillating onset), as ita_start_IR.
3. time_shift(signal, shift=None, unit='time'): a cyclic shift in seconds or samples, one for all channels or one per channel. A fraction of a sample is applied as a linear phase. With no shift, the earliest start of the response moves to the first sample.
4. append(signal1, signal2): concatenation in time.
5. normalize(signal, domain='time', per_channel=False): the signal scaled so that its largest magnitude, in time or in the spectrum, is 1, with the gain removed.
6. amplify_to(signal, level): the signal scaled so that the mean RMS of its channels is level dB, re 20 µPa for channels in Pa.
7. zero_phase(signal) and randomize_phase(signal, seed=None): the magnitude spectrum with zero or random phase.
8. quantize(signal, bits=24): rounding to 2^(bits - 1) + 1 intervals across the peak, halves away from zero, as ita_quantize.

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

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

1. start_ir gives the starts of ita_start_IR on three decaying responses with a weak lead-in, on golden data from tests/golden/make_dsp_edit_ita.m (MATLAB R2026b).
2. The envelope of an AM tone follows 1 + m cos(2π fm t) away from the ends (3.9e-3) and the envelope of ita_envelope for the same samples (1.2e-7).
3. An integer shift is a roll, two shifts of half a sample make one (3.6e-8), and the automatic shift brings the start of the response to sample 0.
4. normalize brings the peak to 1 and returns it, per channel too, and amplify_to sets the level of a signal in Pa to 70 dB.
5. zero_phase and randomize_phase keep the magnitude spectrum, and the zero-phase signal is even.
6. append concatenates and rejects different sampling rates, and quantize gives the samples of ita_quantize exactly.

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