SED save the average response time of rescue services from 7 minutes to less than 5 seconds. Helps to safe more lives.

Every second matters

Function diagram

Step 0: Adaptation

The SED analyzes the common background sounds from the environment in which it is located and adapts.  

Step 1: Detection

The SED detects unwanted sound event that is atypical in the environment.  

Step 2: Extraction

The SED processes the input audio signal and extracts the audio features used to train the classifier.     

Step 3: Classification

The SED classifies the sound event and recognizes whether it falls into one of the undesired categories.  

Step 4: Result

The SED triggers an alarm if the event is evaluated as undesirable.  

Step 5: Notification

The SED informs about the event and help can be on the way immediately.  

System schema

Step 0: Adaptation
Step 1: Detection

The SED analyzes the common background sounds from the environment in which it is located and adapts.  

The SED detects unwanted sound event that is atypical in the environment.

Step 2: Extraction

The SED processes the input audio signal and extracts the audio features used to train the classifier. 

Step 3: Classification

The SED classifies the sound event and recognizes whether it falls into one of the undesired categories.  

Step 4: Result

The SED triggers an alarm if the event is evaluated as undesirable.  

Step 5: Notification

The SED informs about the event and help can be on the way immediately.  

Immediate response saves lives.

The average processing time of a human reaction when evaluating a dangerous situation is around 7 minutes. The SED processing time is 2 seconds. This time can be crucial.

Solutions

Software solution

The SW solution is developed with high emphasis on speed, efficiency and enable third parties to easily port it to their (low power) embedded hardware platforms. 

We worked very hard to get current state-of-the-art Machine Learning techniques with optimized deep neural network architectures running on low-power MCUs in a real time.

Our software comes with a very intuitive user application. Both, the software and the application can be customised to fit your needs. In case you do not want integration into your own product, it is not a problem, we also have our own devices that are ready for your immediate order!

Hardware solution

Our device is ready to operate in both indoor and outdoor situations.  It is designed to be waterproof (IP-67) and can withstand extreme temperatures (-40°C - 70°C). 

Our software solution comes to you fully integrated and ready for immediate use.Integrated GSM/GPRS module ensures operability anywhere where the network is available. 

The user application also comes with the device to provide all the information from connected devices in real time.

Distributors

Sponsors and partners

Bowi

Valvara

City Plzeň

Client reviews

Jablotron

We have been using JALUD's glass breaking detection solution for the past three years and we couldn't be happier with it. We integrated their software into our devices and this process took place to the satisfaction of both parties. 

Therefore, we can not recommend them enough.

Keep up the good work!

Kde SED funguje

FAQ

Do we need powerful HW?

No, the algorithm is highly optimized for use in embedded devices. For our client requirements we tested the SW with STM32F7 CPU, memory requirements were around 250-300kB and time of classification within 600ms.

What is the length of sound event to be analyzed?

It depends on the model settings, but standard length that has proven to work the best is around 0.5s

What type of sound the SED can analyze?

Currently, we are using high precision deep learning techniques to process the sound signal. When we have enough sound data to train the classifier, we can analyze any type of impulsive sound event (such as car crash, bomb blast, gunshot, human screaming, baby cry and many others) .

Are there any ethical issues with the SED?

No, the sound is processed directly on the device and sound events are not stored or further send to the cloud. Unlike cameras, our solution does not interfere with people's privacy.

How do I know that there has indeed been an undesired sound event reported by the SED?

The recognition quality is at a high level, but for some types of sounds (eg human screaming) a false alarm can be expected. Our clients usually install the SED together with the camera system, and therefore we are preparing a variant of the SED together with the camera module, which will ensure that you have a visual overview of the reported dangerous situation.

Do we need powerful HW?

The whole cycle of event: from capturing the event, classification, waking up the GSM, sending the event to the server and handing over the event to the client takes between 2-4 s.

Contact

JALUD Embedded s.r.o. 
Nepomucká 1355/26​​1
CZ-326 00 Pilsen
Czech Republic

Basilio Paz
Tel: 
+420 773 092 900 

Thank you! Your message has been sent.
Unable to send your message. Please fix errors then try again.