Biomedical Engineering Application Brief

Speech Analysis

The Solution

Dr. Roark and colleagues use DADiSP, the graphic display and data management software from DSP Development Corporation, as the core display and analysis tool in their extensive data acquisition and analysis system. Their systems architecture promises a wide application in many areas of health science. The Vocal Motor Control Laboratory (VMCL) has a capability rivaled only by two other laboratories in the world, both of which are mainframe-based.

Physiologic Measurements

The VMCL is based on personal computer technology. A typical acquisition task involves attaching an array of sensors to human volunteers. These sensors record up to 16 time-synchronous physiologic measures at a variety of sampling rates while the subject says a word or sentence. Most digital signal processing software does not permit concurrent differential sampling rates, but DADiSP does. In one session, acquired data might include eight electromyographic (EMG) signals (electrical muscle potential signals recorded by needle electrodes inserted into muscles) sampled at 5 kHz, two laryngeal kinematic signals acquired via instruments inserted through nasal or oral pathways and sampled at 20 kHz, two respiration signals sampled at 200 Hz, and a speech acoustic signal sampled at 10 kHz. Simultaneous video filming of the larynx is performed via endoscopy. Because few volunteers would want to repeat this experience, maximum integrity of the signal quality is vital.

Custom Analysis

Depending on the number of signals involved, a typical experimental session yields 100,000 bytes of digital data per second per subject. A research project might require 25 subjects (10 normal and 15 pathologic) to attend four two-hour experimental sessions. During a project, some 50 gigabytes of physiologic data are acquired and stored directly to magnetic disk. Long-term storage of data is facilitated by the use of optical media. The research team created their own data analysis program with DADiSP. It is a menu-driven, continuous linked command file organized as a relational database. Time-marking, measure extraction, and signal annotation are all automated; one system feature permits extraction of a time subset of one signal in a worksheet for viewing or editing while all other signals in the worksheet are windowed automatically according to user-selected cursors. Acquired measures are updated automatically and stored. Accumulated measures are transferred to spreadsheets for final statistical analysis and comparison.

DADiSP for Health Science Research

Dr. Roark believes that "there is tremendous potential for DADiSP in health science research" because current health science problems require solutions that make use of sophisticated architectures and multidisciplinary cooperation. He states that the success of his research group has depended upon dedicated collaborative effort. The group consists of two electrical engineers, two medical doctors, two speech scientists, and two computer scientists. Since the team has demonstrated that sophisticated research and development tasks can be performed using computer workstations instead of large, expensive mainframe computers, the VMCL is receiving attention from research groups that have relatively small budgets for hardware and software and do not have access.