TL;DR
This paper investigates how glossectomy (tongue surgery for cancer) affects speech acoustics. While vowel sounds remain largely similar between patients and healthy individuals, the study reveals significant differences in fricative sounds (/s/ and /sh/), suggesting altered tongue constriction patterns post-surgery. This research is crucial for understanding speech impairment after cancer treatment and guiding future rehabilitation efforts.
What Problem Does This Paper Address?
Oral and tongue cancers are a significant health concern, and glossectomy, the surgical removal of cancerous tumors from the tongue, is a common treatment. This surgery inevitably alters the tongue's volume, shape, muscle structure, and motility, which are fundamental to producing clear speech. While previous studies have explored speech intelligibility after glossectomy, there's a need for a deeper understanding of the specific acoustic changes that occur. This paper aims to bridge that gap by comparing the acoustic properties of vowels and fricatives produced by glossectomy patients and healthy control subjects, seeking to identify consistent acoustic differences and their potential articulatory causes.
Methods
The researchers employed a comparative acoustic analysis of speech samples. For vowels (/iy/, /eh/, /ah/, /uw/), they measured formant frequencies (F1, F2, F3) using a WaveSurfer formant tracker, manually correcting trajectories for accuracy. For fricative sounds (/s/, /sh/), they analyzed spectral properties using the center of gravity and skewness, calculated from the power spectral density (PSD) estimated via Welch's method. The center of gravity indicates the mean frequency of the spectral peak, while skewness quantifies the asymmetry of the PSD. These acoustic features were then compared across different subject groups.
Datasets
The study utilized a database comprising:
- 18 normal control subjects.
- 16 glossectomy patients (some with pre-surgery data, some with post-surgery data).
All subjects were native American English adult speakers. The patients had T1 or T2 tumors located in the lateral part of the tongue. Speech data was collected by having subjects pronounce 16 VCV nonsense words (4 vowels x 4 consonants) and was downsampled to 22 kHz. The paper also references classic vowel formant data from Peterson et al. (1952) and Hillenbrand et al. (1995) for comparison.
Key Insights and Ideas
- Vowel Acoustics Largely Preserved: Despite surgical alterations, the fundamental acoustic characteristics of vowels (/iy/, /eh/, /ah/, /uw/) produced by glossectomy patients are remarkably similar to those of normal speakers. This suggests that the vocal tract's overall resonance patterns for vowels are relatively robust to partial glossectomy.
- Fricative Acoustics Significantly Altered: The study found significant differences in the spectral properties of fricative sounds (/s/ and /sh/) between patients and controls. Specifically, the center of gravity for these sounds was consistently lower in patients.
- Posterior Constriction Hypothesis: The observed differences in fricatives are hypothesized to stem from a more posterior constriction in the vocal tract for glossectomy patients. This could be due to changes in tongue motility or the physical shortening of the tongue post-surgery, leading to a longer front cavity and thus lower spectral peak frequencies.
- Pre- vs. Post-Surgery: While there are some differences between pre- and post-surgery data, the most consistent and significant acoustic differences were observed between the patient group as a whole and the normal control group, even for patients who had not yet undergone surgery. This suggests that the presence of the tumor itself, or the discomfort it causes, might also influence speech production.
Key Results
- Vowels: Average formant patterns (F1-F3) for the four analyzed vowels were very similar across normal subjects and glossectomy patients. While some minor differences (within 50-80 Hz for F1, up to 240 Hz for F2, and 400 Hz for F3) were noted, no consistent pattern of difference was found.
- Fricatives:
- Center of Gravity: Medians of the center of gravity for /s/ and /sh/ were consistently higher in normal subjects compared to patients. These differences were statistically significant (p < 0.05) for most comparisons, with confidence intervals not overlapping.
- Skewness: While correlated with the center of gravity, skewness differences between normal subjects and patients were less consistent and significant.
Limitations and Future Work
Limitations: * The study focused on patients with T1 or T2 tumors, which are relatively small. Patients with larger tumors (T3-T5) might experience more pronounced speech difficulties. * The database included only two patients with both pre- and post-surgery data, limiting longitudinal analysis. * The paper acknowledges that a 3-D vocal tract shape analysis would be beneficial for further confirming the articulatory explanations.
Future Work: * Expand the database with more subjects, including those with larger tumors and more longitudinal data. * Conduct ANOVA analysis for more rigorous statistical validation of acoustic measurements. * Perform detailed speaker-specific acoustic and articulatory analyses. * Develop vocal tract models to interpret the observed acoustic differences in glossectomy patients.
Summary updated