Open Design Tools for Speech Signal Processing (R43/R44)

This Funding Opportunity Announcement (FOA) seeks applications to design, test, and disseminate acoustic signal processing tools that have the potential to accelerate research studies and facilitate translation of novel algorithms into use with hearing aids, cochlear implants, personal sound processors, and consumer electronic devices.

Open hardware and software designs that are widely shared and readily reconfigured by the end user have become increasingly popular with many engineering communities, including those engaged in biomedical research. Open designs allow users to enhance specific features and retain other aspects of a shared design without modification. Proprietary designs, by contrast, are subject to commercial interests that can restrict the level of documentation provided on its design.  Commercial constraints also encourage manufacture of instrument architectures that support broadly applicable techniques as opposed to tools that can be reconfigured as needed to track an evolving set of scientific questions.

Open designs allow for rapid implementation of new features driven by the needs of a single researcher or a large user community.  Study results can be published and replicated with greater clarity by reference to the original base instrument design. System enhancements can be widely shared and benefit from collaborative design and validation procedures that acknowledge contributions from all team members, innovators and users alike. The net result is that the overall pace of research can be accelerated when open design tools are used. NIDCD held a workshop in October, 2014, to discuss the potential value of a research tool that employs open design principles to provide continued enhancements in the computing power available in a portable form factor and to accelerate research and development of bold new algorithms for the field of acoustic signal processing for speech enhancement and noise reduction.

The summary of the discussion from this workshop is available on the NIDCD website (http://www.nidcd.nih.gov/openspeech), which describes specific needs articulated by the research community. Participants agreed that access to a shared and open tool would lower barriers for hardware and software refinement, accelerate studies with novel acoustic processing algorithms, and facilitate translation of these advances into widespread use with hearing aids, cochlear implants, and consumer electronics devices.  The panel of experts also noted that development of this tool would require considerable effort and substantial infrastructure to support both the initial development phase and an extended period of interactions across different research laboratories to establish widespread use.

This Funding Opportunity Announcement (FOA) invites applications with plans to design, test, and disseminate open design signal processing tools with the potential to accelerate research and translation for acoustic psychophysical research. This will require development of algorithms, test data, and validation procedures for use at the benchtop and for use with human subjects. Curation and dissemination of software as both open source code and executable binary code will be required. Portable hardware systems will be required to run this software in a variety of research studies. Applications should provide initial plans for the design of this hardware along with a roadmap for selling enhanced revisions that provide significant increments in computational power and other enhancements needed to support research studies such as wireless connectivity.

This FOA and its companion open to the broader research community will be used to support grant awards that will run concurrently.  Some overlap is expected for development efforts supported under different awards to different groups. This is intentional and necessary to foster a high level of activity across the research community and to encourage constructive competition between different approaches in order to identify the merits and costs for different approaches. Both FOAs will support research and development to create and disseminate libraries of open source software and signal processing frameworks that can be reconfigured by users to accelerate the pace of their individual research programs. Open source code must be easily reconfigurable and clearly documented.  Compiled "system images" that are validated and ready for use on hardware available for purchase must be disseminated as well.

Responsive SBIR applications will describe plans to design, vend, and enhance hardware platforms on an ongoing basis throughout the award. Plans for continued redevelopment of the hardware platforms to provide substantial increments in the computational power over the project period will be needed.  System designs that allow users to extend or reconfigure the hardware can be proposed. SBIR Awards made under this FOA, as compared to the companion FOA, will be uniquely suited to hardware design and vending efforts, as well as servicing user requests to resolve issues arising from individual attempts to use these tools.

The open design approach will require a unique emphasis on outreach and dissemination activities to ensure widespread and effective use of these research tools by outside laboratories. Curation of websites with user support forums, answers to common user questions, and the means to distribute software updates is encouraged. Other activities intended to support effective use and reconfiguration of this research tool by outside laboratories can be proposed. Agile software development methods are encouraged in order to gain feedback from users, build consensus among users, and identify emerging needs based on ongoing research. A "waterfall model" of sequential design steps leading to distribution of a final product near the end of the project will not be considered responsive to this funding opportunity.

Activities encouraged under this FOA include, but are not limited to, development of:

• A comprehensive system that includes all functions widely recognized as essential features for a hearing aid, such as acoustic feedback cancellation and multiband amplitude compression. These functions should be provided as both source code and a compiled, binary image that can be run on various target platforms to support research studies of hearing aid signal processing.
• Function libraries that support areas of active research for speech signal processing, such as algorithms for various forms of speech enhancement, noise reduction, dynamic gain adjustment, frequency lowering, and other types of acoustic processing, which can be linked into the comprehensive system noted above.
• Development of validation procedures will be needed to demonstrate proper operation of functions in the code library as originally written, and to support future efforts to port software to new hardware, optimize specific sections of code, or develop alternative processing algorithms. Performance benchmarks will be needed to quantify important differences between alternate implementations of similar functions.
• A variety of system architectures may be needed to provide different operating modes such as very short audio latency with restricted amounts of processing vs. relatively long latencies that allow more comprehensive processing algorithms to operate. Audio latency, e.g. the time delay between presentation of a raw sound at the microphone input and the processed variant at a speaker, will require careful consideration throughout the design and development activities.