UCL Discovery
UCL home » Library Services » Electronic resources » UCL Discovery

A speech pattern processing method for Chinese listeners with profound hearing loss.

Wei, Jianing; (1993) A speech pattern processing method for Chinese listeners with profound hearing loss. Doctoral thesis (Ph.D), UCL (University College London). Green open access

[thumbnail of A_speech_pattern_processing_me.pdf]

Download (19MB) | Preview


frequency, voiceless frication, and speech amplitude information for Chinese listeners with profound hearing loss. Two broad problem areas have been investigated. The first concerned the robust analysis of speech signals into frication, voicing, and silence regions. This task was formulated as a pattern classifying recognition problem. The multi-layer perceptron (MLP) method was employed in the development of the pattern classifier. In the preprocessing stage of the algorithm, an input vector was generated which consists of a set of parameters from the output of a wide-band filter-bank, and from calculations of zero- crossing rate, short-time energy and auto-correlation of the speech signal. A feature divergence analysis method was developed and used as a guide-line for selecting useful features from the pre-processor. Both training and testing speech data were hand- labelled. The classifier was initially trained on anechoic speech recordings of three speakers (two male, one female), and tested both on new recordings from the same speakers used for training and also with speech from three other speakers. The algorithm was then additionally further trained and tested using reverberant speech, and speech with a babble noise background. Satisfactory results were obtained from three different recording conditions. The classifier was implemented on a Masscomp 6000 computer, but the algorithm can be ported to real-time DSP processors. The second task in this study concerned the investigation of the effectiveness of the speech pattern processing scheme for Chinese hearing impaired listeners as an aid to lipreading. A facility for dubbing off-line speech pattern sounds from the Masscomp computer onto video recordings in synchrony with the video-image was specially developed and used in the study. Lipreading tests were then carried out using this facility and employing the compound speech patterns of voice fundamental frequency, voiceless frication excitation information, and amplitude information. The voice fundamental frequency can be obtained from the laryngograph signal or the MLP-Tx algorithm developed in the Department, and the voiceless frication information is extracted by the algorithm developed in this study and coded by combining aperiodic low frequency sound. Responses from both normal and hearing impaired Chinese listeners for lexical tone, consonant, and connected speech were obtained. The perceptual analysis method was used in the analysis of consonant perception. The results indicate that this speech pattern encoding scheme is effective as an aid to lipreading when applied to tonal languages such as Chinese.

Type: Thesis (Doctoral)
Qualification: Ph.D
Title: A speech pattern processing method for Chinese listeners with profound hearing loss.
Open access status: An open access version is available from UCL Discovery
Language: English
Additional information: Thesis digitised by ProQuest.
URI: https://discovery.ucl.ac.uk/id/eprint/10103942
Downloads since deposit
Download activity - last month
Download activity - last 12 months
Downloads by country - last 12 months

Archive Staff Only

View Item View Item