Maintenance release v1.0.3

This maintenance release contains:

• Fix incorrect use of restrict keyword #61
• KWS unit test switched from bit exact checking to Noise to Signal ratio checking.
• Most of the double precision floating-point operations in SpeexLib library code replaced by single precision.
• Changes in unit test to clarify the use of Noise-to-Signal ratio instead of Signal-to-Noise (SNR) ratio
• Documentation update: Score calculation formula uses 1/1.5 scaling rather than 1*1.5.
• Documentation update: Clarify that only float version of SpeexDSP is supported.
• Documentation update: KWS unit test description (changed to use Noise-to-Signal ratio)
• Other minor documentation improvements.

Fix an issue in the KWS NN model

The Key Word Spotting (KWS) model in ee_nn_tables.c is extracted from a reference TensorFlow Lite Micro NN model. Due to a limitation of the script used for the conversion (an accuracy issue), the model is not fully matching the reference model. This release updates the KWS model to fully match the reference model. The corresponding data for unit level testing (kws_expected.c) is also updated.

Fixed potential memory bound error.

PR #58 fixes a potential bug where a multiply goes out of bounds.

First release.

This is the first release of the AudioMark(tm) benchmark.