A Configurable Accelerator for Keyword Spotting Based on Small-Footprint Temporal Efficient Neural Network
Round 1
Reviewer 1 Report
I think we need to validate and verify the proposed CNN scheme in light of the confidence interval of 95%. Need the distribution for random seed and # of simulations as changed the seed.
Author Response
We sincerely appreciate the reviewer for the thorough and valuable suggestions to examine our manuscript.
Please see the attachment for the response.
Author Response File: Author Response.pdf
Reviewer 2 Report
- Please define acronym "MFCC" at first time its used and other frequently used acronyms.
-It is not very clear how overlap (not using non-overlap) of input audio frames reduces FFT by half ?- perhaps add a bit of clarification around this area [just a soft suggestion] Does it not disturb the raw audio sequence?.
- fix typo 'focu" in section 2.1
Overall , the proposed HW optimizations are fairly novel.
Author Response
We sincerely appreciate the reviewer for the thorough and valuable suggestions to examine our manuscript.
Please see the attachment for the response to comments.
Author Response File: Author Response.pdf
Reviewer 3 Report
The paper presents an improved hardware / software version of a temporal CNN to fit IoT devices and for the Keyword spotting task.
The paper has some novelty and all the innovative aspects are treated in the paper.
The work seems interesting but definitely needs a more stretched description, avoiding too dense and long sentences, in favour of simpler and clearer descriptions.
The paper is not too long, so there is room for additional and more explicit descriptions and explanations. Please expand all the acronyms used first, and make more explicit any passage.
Please make the experimental part stronger, by explicitly justifying the use of the data exploited and whether they are able to challenge the computational power / the performance of your solution when compared to existing approaches.
Please add the limitations of your approach.
Author Response
We sincerely appreciate the reviewer for the thorough and valuable suggestions to examine our manuscript.
Please see the attachment for the response to comments.
Author Response File: Author Response.pdf
Round 2
Reviewer 1 Report
Work done so solely depends on the Editor's decisions. No further comments. Good luck.