Characterising vowel phonation by fundamental spectral normalisation of LX-waveforms

Objective spectral reference standards for normal vowel phonation are described. Application in larynx cancer monitoring is envisaged. 120 individuals contributed to a database of vowels lad and /i/ in the form of trans-larynx impedance time series captured using an electrolaryngograph. The impedance signals are used for iterated power spectral estimation followed by spectral intra-pooling for each individual. Pooling of spectra across many individuals is complicated by significant variations in the precise frequencies and powers of the fundamental, harmonics and any other characteristic peaks that may be present. Fundamental-harmonic normalisation (FHN) of individual spectra circumvents these obstacles by normalising all powers relative to that of the fundamental and by transforming the entire frequency range into floating point multiples of the fundamental-frequency fO. Population pooling then results in stable FHN-spectral patterns and associated characteristic distributions of fo values. For females it is FHN-spectral pattern that matters most. For males the fO distribution is also highly characteristic. Introduction The choice of medical treatment, driven by economic considerations and issues of accepted practice, is increasingly based on objectively assessed outcome. A particular example driving this study is the competition between radiotherapy and surgery as treatment of choice for larynx cancer. Since both techniques are equally successful for local control of the disease, quality of life after treatment has become an influencing factor. In this context radiotherapy has obvious advantages for the maintenance of vocal fold function compared to the surgical removal of tissues in laryngectomy. However, conservative and reconstructive surgical techniques now claim to offer similar advantages. The result is a heightened interest in the relative degree of conservation offered by the two therapeutic modalities. To the patient the most direct evidence of normal vocal fold functionality is 'voice quality'. To the expert such evidence is to be found in measures of glottal waveform that are free from the complicating factors introduced by tract resonance. In tum this has spotlighted the lack of concise but characteristically detailed objective reference standards for normal glottal waveforms, or for that matter voice quality, against which a patient can be compared. Pilot studies for assessing patient vocal fold function before and at intervals after radiotherapy have already shown promising results, even though they are based on relatively unrefmed standards (1,2]. This report describes what is believed to be significant progress in the generation of improved standards that will underpin further clinical investigations Theoretical Background Speech and language therapists, SALTs, subjectively assess and score voice quality using a range of parameters (3]. Many parameters are simply descriptive, e.g. whisper, although some have been adopted from the physical sciences to produce a hybrid terminology, e.g. fundamental frequency and shimmer (rather than variance). All are capable of interpretation in terms of spectral content in the frequency domain [4]. The periodic glottal waveform produced by the vibrating vocal folds of the larynx is the driving force behind the production of the complex human acoustic waveform. As such the glottal waveform is an indication of the functional integrity of the vocal folds which, as already explained, is of particular interest in the management of larynx cancer. In this and the wider medical context SALTs increasingly supplement their assessments with objective measurements that are closely correlated with glottal waveform, in particular those from electrolaryngography [5]. MAVEBA 1999, Firenze, Italy 1 Models and Analysis of Vocal Emissions for Biomedical Applications