Acoustic analyses

Acoustic analysis is an instrument that is often used for objective evaluation of voice quality. The parameters used for the evaluation of voice quality often consist of fundamental frequency, perturbation and noise measures. There are numerous software programs available for acoustic analyses. Performing acoustic analyses in tracheoesophageal speech is more difficult in tracheoesophageal voice than in normal voice. The problems are caused by the irregularity of the voice. These irregularities are sometimes very severe and cause errors in the pitch-extraction (see figure 1. below). It is therefore important to check the results of pitch-extraction by means of visual inspection of the pitch markers placed by the program.

Figure 1. Pitch-extraction results of a sustained /a/ of a tracheoesophageal speaker using the Computerized Speech Lab with CSL software (bottom) and the Multi Dimensional Voice Program software (top). The pitch-markers (see arrows at the right hand side) are correctly placed with MDVP and not with CSL.

In order to obtain a first impression of voice quality, provide information to other clinicians, and prepare for further acoustic analyses, acoustic signal typing is a useful instrument. These acoustic signal types are based on a narrow-band spectrogram displaying the harmonics in the voice signal. Van As et al. describe an acoustic signal typing system that can be used to group voice samples of tracheoesophageal speech can be categorized into four different signal types. (Van As 2001). In figure 2, these four signal types and the criteria for their use are shown. Research shows that these signal types are related to the overall judgment of voice quality as good, reasonable, or poor, which makes them useful as an adjunct to perceptual evaluation. Apart from these signal types, also objective acoustic measures can be calculated. The following 6 measures have shown to be relevant in relation to voice quality: fundamental frequency, standard deviation of fundamental frequency, percentage of voiced, harmonics-to-noise ratio, glottal-to-noise ratio, and band energy difference. For a detailed description of these parameters see Van As (Van As 2001).

Figure 2. Categorization of tracheoesophageal voice quality into 4 different signal types, based on the narrow-band spectrogram. For clarification each figure shows the entire 2 second at the top, followed by a 0.1 s selection from the middle of the vowel, the curve of pitch extraction, the narrow-band (100 ms) spectrogram, and the long-term average spectrum.