Speech technology and multimodal methods

• Grant number: 09/18242-5
• Support Opportunities: Regular Research Grants
• Start date: April 01, 2010
• End date: September 30, 2012
• Field of knowledge: Engineering - Electrical Engineering - Telecommunications
• Principal Investigator: Miguel Arjona Ramírez
• Grantee: Miguel Arjona Ramírez
• Host Institution: Escola Politécnica (EP). Universidade de São Paulo (USP). São Paulo , SP, Brazil

Abstract

This research plan is made up of multiple themes connected to speech analysis, speech coding, speech recognition, speaker identification and multibiometric person identification. Speech analysis, in addition to its own importance, also provides signal and parameter representations that are necessary for the other themes. Novel types of autoregressive analysis will be explored for coding the short-term spectral envelope of wideband speech signals, mainly, by means of vector quantization techniques and Gaussian mixture models. These novel analysis techniques could lead to applications for the representation of the speech excitation signal, which will be mainly represented by the modulation spectra of filterbank components. Several parametric representations will also be applied to speaker identification tasks, where speech recognition could possibly be used as an aid. Speaker recognition tasks will combine through data fusion the individual decisions about subject identification obtained from signals acquired through microphones, stethoscopes and cameras for visible light and long infrared radiation. All of these signals should be classified by vector quantizers and neural networks. (AU)