Analyzes of exemplary spectrograms
Author: Ph.D. Marcin Just (DiagNova Technologies)The influence of distortions and noise
A comparison of the narrowband spectrogram (long letter "a") for a voice with a small amount of non-harmonic distortions and a large amount of noise is shown in Fig. 5.
Fig. 5. Two spectrograms with different amounts of non-harmonic components: a) large; b) small
The spaces between the horizontal lines of harmonics should ideally be white (or more precisely, much brighter than the adjacent lines of "harmonics").
The same samples after the contrast increase operation are shown in Fig. 6. The gray inserts of inharmonics between the horizontal lines of the harmonic components are clearly visible.
Fig. 6. The spectrograms in Fig. 5 after enhancing the contrast. Number of non-harmonic components: a) large; b) small
It is important that in this case the non-harmonic components evenly fill the area between the harmonics, which is clearly visible in Fig. 7. (spectrogram magnification from Fig. 5).
Fig. 7. Spectrogram magnification from Fig. 5. Number of non-harmonic components: a) large; b) small
Such a spectrographic image is characteristic of the presence of noise without additional signal distortions. This may indicate incomplete closure of the folds caused by a small organic change.
For the cases of low-frequency voices F0, the noise analysis can also be performed on a strongly "stretched" broadband spectrogram (Fig. 8).
Fig. 8. Comparison of the energy distribution in the range of individual periods for a noisy (a) and "clean" (b) waveform
The presence of noise is evidenced by the presence of higher harmonics, especially before the moment of closing the vocal folds (the effect is visible as no reduction of the shading towards the right - Fig. 8a).
Poor harmonic structure
In some cases, the excitation signal generated by the folds may be so low in harmonic frequencies that the harmonic structure is limited to two or three lines on the spectrogram, as shown in Fig. 9.
Fig. 9. Voice spectrogram of a person after laryngectomy (the word "małżeństwo")
The formant structure is much clearer here than the harmonic one (Fig. 10).
Fig. 10. The spectrogram from Fig. 9 with the formants' waveform marked
These types of spectrograms are usually associated with the formation of the voice by structures other than the vocal folds.
Subharmonic distortions
In cases of certain pathologies or postoperative conditions (e.g. chordectomy), quite specific disturbances - subharmonics - may appear in the voice. Then, frequencies appear in the voice that are a certain specific fraction of F0: (1/2)F0, (1/3)F0, (1/4)F0 etc. and their multiples, e.g. (7/2)F0. An illustration of the influence of subharmonics (1/2) F0 can be the spectrograms from Figs. 11a – 11c and the periograms from Figs. 12a – 12c.
Fig. 11a. Extremely large influence of subharmonics (1/2) F0 on the spectrogram form (only two fragments without subharmonics)
Fig. 11b. Little influence of subharmonics (1/2) F0 on the spectrogram form (only in one fragment)
Fig. 11c. No subharmonics
Fig. 12a. Extremely large influence of subharmonics on the form of the periogram (only in green fields there are no subharmonics)
Fig. 12b. Little influence of subharmonics on the form of a periogram (red color - subharmonics lines)
Fig. 12b. Little influence of subharmonics on the form of a periogram (red color - subharmonics lines)
It is important that the lines related to subharmonics (successive multiples of the subharmonics) lie in the middle (of course for the fundamental subharmonic frequency equal to (1/2) F0) between the harmonic frequency lines (Fig. 13).
Fig. 13. An enlarged fragment of the spectrogram with visible subharmonic disturbances
The presence of non-harmonic components, especially subharmonics, in the low frequency range indicates a distortion of the speech signal. The deformations are usually caused by uneven work of folds, different work of both folds caused by paralysis or large organic lesion. Small organic changes occur less frequently in this way. The case with which you are dealing also depends on whether the disturbances in the spectrogram are permanent or temporary. In the case of continuous disturbances, there is a high probability of the existence of large changes of the organic type, in the case of temporary changes, fatigue paralysis of the vocal fold or a smaller change of the ballot type can often be found.