[aguirreSL]

Adds pytta.ecma418 with the two methods of ECMA-418-1:2024 for prominent discrete tones, the same methods as in ECMA-74 Annex D:

1. tone_to_noise_ratio(signal, tone_frequency=None, resolution=None): tone-to-noise ratio (TNR), Clause 11 and Annex A.
2. prominence_ratio(signal, tone_frequency=None, resolution=None): prominence ratio (PR), Clause 12 and Annex B.
3. critical_band(frequency): the edges of the critical band, Clause 10, used by both.

With a tone frequency, each method returns a dict with the frequency, the ratio in dB, the criterion in dB and whether the tone is prominent (11.5 and 12.6). Without one, it searches the spectrum and returns one such dict per candidate. The spectrum is a Welch average with a Hann window, with a line spacing below 0.25 % of the tone frequency for TNR and below 1 % for PR (Clause 7). The implementation covers the bandwidth of Formula (2), the tone band of A.1, the masking noise of A.2, the primary and secondary tones within the proximity spacing of Formula (14), and the outer bands of Tables 2 and 3, with the lower band referred to 100 Hz below 171.4 Hz (B.8).

The standard leaves the identification of tones to the operator (8.2 c). In the search mode, a line is a candidate when it is the highest within 15 % of its critical band and at least 10 dB above the median of that band. The audibility checks of 11.8 and 12.8 are left out, since they need a calibrated level and a listener.

Regarding the text of the standard, four points are worth checking in review:

1. In the copy used, the powers of ten of C_L,2 and C_U,2 in Tables 2 and 3 are lost. The values -6.90e-5, -8.20e-6, 7.70e-5 and 2.16e-5 were used, since with them Formulae (21) and (22) are continuous at 1600 Hz (1275.5 Hz and 2002.6 Hz from both sides) and f1,L is 20.0 Hz at 171.4 Hz, as 12.4 states.
2. Formula (21) is printed as C_L,0 + C_L,0 f_t + C_L,2 f_t^2. The second term was read as C_L,1 f_t.
3. Formula (A.7) is printed with X_t,p - X_t,s. The sum was used, as in Formula (15) and the text of A.6.
4. Formula (23) is printed "for f_t <= 171.4 Hz". It was applied above 171.4 Hz, as the sentence before it says.

The tests in tests/test_ecma418.py print the measured differences:

1. The worked values of the standard: the critical bandwidth at 1000 Hz is 162.217 Hz (162.2 in the standard), with band edges 922.18 Hz and 1084.39 Hz (922.2 and 1084.4). The bandwidth at 500 Hz is 117.255 Hz (117.3), and the proximity spacings at 150 Hz and 850 Hz are 23.00 Hz and 63.84 Hz (23 and 63.8).
2. The criteria of 11.5 and 12.6 at 100, 500, 1000 and 5000 Hz, exact to 1e-3 dB.
3. TNR of a tone in white noise against the analytic value 10 lg[(A^2/2) / (N0 Δfc)]: +0.025 dB at 300 Hz and +0.062 dB at 2 kHz.
4. PR against the analytic band powers: -0.046 dB at 150 Hz, where the lower band rule of B.8 applies, and -0.056 dB at 1500 Hz.
5. Two tones at 600 Hz and 625 Hz are combined, and 600 Hz and 545 Hz, beyond the proximity spacing of 40.7 Hz, are not. The expected values of the two cases are 1.6 dB apart.
6. Pure noise gives no candidate and no prominent tone, and a 2 kHz tone in noise is found once, within 1 Hz.
7. Invalid input is rejected.

No external implementation was found to compare with: SQAT and the ITA-Toolbox have no TNR or PR.

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