Harmonized Voice Control (HVC)
Realize a speech-conditioning software module for voice control and hands-free telecommunications in home and car environments on a 16-bit embedded DSP.
| Company: |
Royal Philips Electronics, Philips Semiconductors, International Technology Centre Leuven (ITCL) / Philips Digital Systems Laboratories-Leuven (PDSL-L) |
| Department: |
Audio Competence Centre |
| Supervisors: |
Marc Herregods, Dieter Therssen, Ludo Van Paepegem, Thierry Gouraud |
| Dates: |
02/99 - 07/01 (30 months) |
| Location: |
Leuven, Belgium, with occasional travel to Philips Research Laboratory Eindhoven (NatLab), The Netherlands, and Philips Software Centre, Bangalore, India |
| Team size: |
5 (Leuven), 2 (Eindhoven), 3 (Bangalore) |
| Role: |
DSP Software Engineer |
| Customers: |
Philips Consumer Electronics, Philips Speech Processing, Magnetti Marelli, PSA Peugeot Citroën |
Responsibilities
• Studied and tuned an Acoustic Echo Cancellation and Noise Suppression algorithm.
• Designed a 16-bit fixed-point model and implemented the algorithm on the target processor.
• Specified the interface for integration with the speech recognition engine.
• Programmed an object-oriented voice control component to encapsulate the speech-processing library.
• Participated in the development of a platform for real-time simulation on a target board and authored test cases.
• Recorded multi-channel sound databases and developed tools for algorithm tuning and selection.
• Developed a PC-based prototype for algorithm evaluation and real-time demonstrations.
• Followed-up the cooperation with Philips Software Centre, Bangalore, India, for offshore development of audio features; defined the Configuration Management Plan and audited the process quality.
Functional Environment
Digital Signal Processing (DSP), adaptive filtering, Normalized Least-Mean Square algorithm (NLMS), Fast Fourier Transform (FFT), speech conditioning and recognition, voice command and control, audio coding, fixed-point modeling, throughput and memory trade-off and optimization, system partitioning and chip peripherals specifications
Technical Environment
C, Visual C++, assembly, Philips REAL RD16x, Hitachi SH-4, Philips REAL IDE, WindRiver Tornado, Microsoft Windows, ActiveX, Microsoft Visual Studio, Bullseye C-Cover, NuMega BoundsChecker, The Mathworks Matlab, Cygwin, Standard Audio Development Environment (in-house audio framework), Philips VoCon (speech recognition engine for command and control)
Innovations
• Created floating-point to fixed point C conversion macros to replace Frontier Design’s A|RT library that was used earlier; the new module performed 200 times faster for our requirements.
• Proposed and organized training sessions to leverage the expertise in DSP and acoustics within the organization.
• Recorded in-vehicle multi-channel audio databases for offline algorithm tuning; created an application to normalize and read multi-channel audio data and automatically compute voice recognition scores.
Publications
G. Egelmeer, T. Gouraud, J.-P. Jallet, E. Mellery, M. Sarkar, S. Spronck, and H. Suaudeau, “16-bit implementation of a multi-channel AEC in combination with a speech recognizer” [abstract], submitted to the Philips Conference on Digital Signal Processing, Veldhoven, The Netherlands, Nov. 9-10, 1999. |