Voice recognition infographics. speech recognition, smart home and voice assistance system concept diagram. Vector 
Project purpose and unfolding, clients.
With the new information and signals processing technologies the modern gadgets and devices become possible. One of such the device is smart speaker. It includes two stages: recognition of some typical phrase, like “OK, Google” or “Hello, Siri” and the working mode: answering all the kind of questions. Like: who is the president of France, or what is the highest mountain pick in Europe ?
The second type of phrases are processed and detected by Amazon recognition engine. And the first phrase is supposed to detect out team.
The project has been realised in 2017..2018 and included the hardware development, mecanical development, software integration and the algorithms development. Some test bench has been also developped and installed.
Principal client are the physical persons, who use this device at home.
Technical realization of project
As it was shown, the software integration part and hardware and mechanical design and implementation have been carried out by SagemCom traditional teams. Concerning the first phrase detection it was distributed between signal processing team inside of SagemCom and subcontractors.
My participation in project.
My participation lasted about 6 months.
During this period the following items have been carried out:
- Literature review of how the phrase is generated by human and which methods one can use to detect the known phrase
- Presentation of literature overview and how work the speech generation and detection for extended team
- Simulation of some typical methods for speech phrase detection under Matlab, using generated test vectors (different phrases, different phrase authors (age, gender, timbre of voice)
- Implementation and test of some methods under C, Linux
Environment and techniques used in project
Environment: C, Linux, Matlab, speech test banc
Technical methods: speach generation & processing (pitches, harmonics, Kepstres processing, Kalman filtering, neuron networks, autocorrelation, spectrum processing
Main results
- Presentation for the team has been done to explain how works speech generation and detections
- Simple algorithm has been developed and tested for test bench to detect the phrases in prensence of noise.
- High detection rate has been proved in presence of different types of noise: rain, television, white noise, football match, speech of another person