Definition of Tonality:
Tonality is a measure of the proportion of tonal components in the spectrum of a signal and allows a distinction between tones and noises. Tones consist mainly of tonal components which show in the spectrum as pronounced peaks. Noise and broadband noises have no or little tonality.
Calculation is based on publications by Terhardt and Aures and is achieved via short time spectra, arrived at via an FFT of 4096 points and a Hanning window. In the first pass, a search is made for spectral lines Si which are larger than both their respective neighbors Si ± 1.
Only the lines which are at least 7 dB larger than the lines Si ± 2 and Si ± 3 are taken into account. The seven-line groups with the indexes i-3 to i+3 found in this way as pure tones are removed from the spectrum.
Only up to a single tonal component per critical band is taken into account. A search is then made for narrowband noise in the remaining spectrum of bandwidth smaller than the critical bandwidth at this location. This is because such signals also create an impression of tonality, although only to a small extent.
The so-called remaining level is calculated for the tonal components thus found. This is the level of the components minus:
1. resting threshold
2. noise power in each critical band, given by the power of the remaining spectral lines after subtraction of the tonal components
3. excitation level resulting at this position from the other tonal components.
A constant in tonality calculation K achieves that a value of 1 tu results from a sine tone of 1 kHz and 60 dB.
Definition of Tone to noise Ratio:
Just like the Prominence ratio analysis, the purpose of this analysis is to identify tonal components in a signal and to present them as numeric values. The tone-to-noise ratio is defined as the power of a tone compared to the noise power of the frequency group surrounding the tone frequency.
In most implementations of this method, the tone has to be manually detected in the spectrum by the user, and the width of the tone must be marked manually as well, which causes some uncertainty. In the ArtemiS implementation, both tasks are handled automatically.
The determination of the frequency groups and frequency limits complies with ECMA-74 (9th edition). If several tones are found within one frequency group, they are treated as specified in ECM-74 If the two loudest tones are close to each other, the power of the weaker one is added to that of the stronger one; if they have a considerable distance, they are treated as separate tones.
According to ECMA-74, tones are considered as prominent if their tone-to-noise ratio is at least 8 dB. Below 1 kHz, this threshold is increased by 2.5 dB/octave. Detected tones exceeding this threshold are displayed as single number values.
According to ECMA-74, the frequency resolution of the FFT spectrum should not exceed 1 percent of the tone frequency. Therefore, a suitably large FFT length sould be chosen for signals with low tones. |