Plenary lecture III: Enhancement and restoration
Hector Manuel Perez-Meana · 2008
Persons that suffer from diseases such as throat cancer require that their larynx and vocal cords be extracted by a surgical operation, and then require rehabilitation in order to be able to reintegrate to their individual, social, familiar and work activities. To accomplish this, different methods have been used, such as: The esophageal speech, the use of tracheoesophagical prosthetics and the Artificial Larynx Transducer (ALT), also known as electronic larynx. The ALT, which has the form of a handheld device, introduces an excitation in the vocal track by applying a vibration against the external walls of the neck. This excitation is then modulated by the movement of the oral cavity to produce the speech sound. This transducer is attached to the speaker's neck, and in some cases in the speaker's cheeks. The ALT is very easy using even for new patients, although the voice produced by these transducers is unnatural and with low quality, besides that it is distorted by the ALT produced background noise. The Esophageal speech is produced through the compression of the contained air in the vocal tract with the tongue. This air is swallowed and as passing through the esophageal-pharynx segment produces a vibration of the esophageal upper muscle, bringing about the speech. The generated sound is similar to a burp, the tone is commonly very low and the timbre generally harsh. In ALT as well as in esophageal speech, the voiced segments are the most affected part of speech. Several approaches have been proposed to improve the quality and intelligibility of ALT produced, as well as esophageal speech signals. Some of them reduce the ALT produced background noise by using cepstral root subtraction or adaptive filtering. However the speech quality produced by these approaches is still poor. Another approach intended to improve the speech quality estimating the frequency band from 4 KHz to 8 KHz using the frequency band from 300Hz to 4 KHz. Although this approach may be an attractive alternative, it must be still improved. A promising approach is based on speech conversion techniques which carry out a spectral conversion using vector quantization methods. A similar approach based on a pattern recognition approach, has also been proposed, in which, firstly the voiced segments are detected and identified. Then the voiced segments are replaced by their equivalent voiced segments of normal speech while the unvoiced segments are kept without change. Finally the voiced, unvoiced and silence segments are concatenated together to produce the restored speech. These approaches perform fairly well although still present some problems because the spectral conversion reduce a continuous spectral space into a discrete code book, which may produce a distortion that still must be reduced. This speech presents a review of alaryngeal speech enhancement systems, providing also evaluation results to show the improvement in the quality and intelligibility of produced speech.