Hi
@Mr Tech Laden (can't link to your post)
The Megachips mention was interesting. However, just wanted to confirm that you linked the right patent. I had a quick search of the two you linked but didn't find a mention.
Digimarc do have an older patent that mentions both "Megachips" and "neuromorphic processing", I'm guessing this is the one you were talking about? It sure is interesting. From a quick glance it seems to be doing some type of wake word detection, such that upon detection of an actual trigger it activates the second more energy intensive processor. This is a use case Akida excels in and which Brainchip have mentioned many times.
Pure speculation, DYOR
https://appft.uspto.gov/netacgi/nph-Parser?Sect1=PTO2&Sect2=HITOFF&u=%2Fnetahtml%2FPTO%2Fsearch-adv.html&r=1&f=G&l=50&d=PG01&p=1&S1=(%22megachips%22+AND+neuromorphic)&OS=%22megachips%22+AND+neuromorphic&RS=(%22megachips%22+AND+neuromorphic)
| United States Patent Application | 20200133625 |
|---|
| 1 | Kind Code | A1 |
|---|
| 2 | Sharma; Ravi K. ; et al. | April 30, 2020 |
|---|
METHODS AND SYSTEM FOR
CUE DETECTION FROM AUDIO INPUT, LOW-POWER DATA PROCESSING AND RELATED ARRANGEMENTS
AbstractMethods and arrangements involving electronic devices, such as smartphones, tablet computers, wearable devices, etc., are disclosed. One arrangement involves a low-power processing technique for discerning cues from audio input. Another involves a technique for detecting audio activity based on the Kullback-Liebler divergence (KLD) (or a modified version thereof) of the audio input. Still other arrangements concern techniques for managing the manner in which policies are embodied on an electronic device. Others relate to distributed computing techniques. A great variety of other features are also detailed.
| Inventors: | Sharma; Ravi K.;(Portland, OR); Thagadur Shivappa; Shankar;(San Diego, CA); Alattar; Osama M.;(Tigard, OR); Bradley; Brett A.;(Portland, OR); Long; Scott M.;(Portland, OR); Kamath; Ajith M.;(Beaverton, OR); Holub; Vojtech;(Portland, OR); Brunk; Hugh L.;(Portland, OR); Lyons; Robert G.;(Portland, OR); Gurijala; Aparna R.;(Port Coquilam, CA) |
|---|
| Applicant: | | Name | City | State | Country | Type |
|---|
Digimarc Corporation |
Beaverton |
OR |
US |
|
|
|---|
| Family ID: | 53479711 |
| Appl. No.: | 16/665906 |
| Filed: | October 28, 2019 |
1. A method, comprising: obtaining audio input; at a first processor, processing the audio input to discern a characteristic of the audio input; wherein processing the audio input to discern the characteristic of the audio input comprises processing the audio input to discern auxiliary data conveyed by a digital audio watermark signal present within the audio input, the processing of the audio input to discern the auxiliary data comprising: buffering frames of the audio input, transforming the frames into spectral magnitude frames, accumulating spectral magnitude frames into an accumulation buffer, extracting spectral magnitude values corresponding to selected bits of the digital audio watermark signal, and correlating the extracted spectral magnitude values with a predetermined signal to produce a correlation metric; generating an output based upon the processing to discern the characteristic; and controlling an operation of a second processor distinct from the first processor based on the generated output.2. The method of claim 1, wherein the first and second processors are components of an electronic device, the method further comprising generating an audio signal corresponding to sound propagating within an aural environment surrounding the electronic device, wherein the obtained audio input comprises a plurality of samples of the audio signal.3. The method of claim 2, wherein the second processor is a CPU.4. The method of claim 3, wherein the first processor is a digital signal processor.5. The method of claim 3, further comprising processing the audio input while the second processor is in an idle or sleep state.6. The method of claim 5, wherein controlling an operation of the second processor comprises causing the second processor to enter into a higher power state than the idle or sleep state.7. The method of claim 1, wherein processing the audio input to discern the characteristic of the audio input comprises processing the audio input to determine the presence of audio activity within the audio input.8. The method of claim 7, wherein processing the audio input to determine the presence of audio activity comprises determining zero-crossing or short-term energy metrics from the audio input, determining co-occurrence statistics of the zero-crossing or short term energy metrics, and classifying the audio input based on the co-occurrence statistics.[0033] Generally, the sensor interface module 130 may include one or more microprocessors, digital signal processors or other microcontrollers, programmable logic devices, or the like or any combination thereof. The sensor interface module 130 may also optionally include cache or other local memory device (e.g., volatile memory, non-volatile memory or a combination thereof), DMA channels, one or more input buffers, one or more output buffers, and any other component facilitating the functions it supports (e.g., as described above). In one embodiment, the sensor interface module 130 may be provided as the "Sensor Core" (Sensors Processor Subsystem (SPS)) from Qualcomm, the "frizz" from Megachips,or the like or any combination thereof. Although the sensor interface module 130 is illustrated as an individual component, it will be appreciated that the sensor interface module 130 (or portions thereof) may be functionally integrated into one or more other components (e.g., the CPU 102, the communications module 114, the audio I/O module 122, the audio DSP 128, the cue detection module 134, or the like or any combination thereof).[0286] Embodiments of the present technology can also employ neuromorphic processing techniques (sometimes termed "machine learning," "deep learning," or "neural network technology"). As is familiar to artisans, such processors employ large arrays of neuron-like elements--interconnected to mimic biological synapses. Such processors employ programming that is different than the traditional, von Neumann, model. In particular, connections between the circuit elements are weighted according to correlations in data that the processor has previously learned (or been taught). When a pattern of data (e.g., a set of audio, image or other sensor data) is applied to the processor (i.e., to inputs of several of the circuit elements), certain nodes may spike while others remain relatively idle. Each of these nodes may serve as an input to plural other circuit elements, triggering further spiking in certain other nodes--a chain reaction that ultimately provides signals to output nodes to indicate the results of theneuromorphic processing. (In addition to providing output signals responsive to the input data, this process can also serve to alter the weightings, training the network to better respond to certain patterns that it has seen (i.e., processed) before.) Such techniques are well suited for pattern recognition applications, among many others.
(20min delay)