Nowadays, with the rapid development of science and technology, artificial intelligence has gradually become an essential for customers when selectting products. Voice recognition, voiceprint authentication, face recognition and other technologies are also applied to various industries.

At present, a number of artificial intelligence engine manufacturers announce that their speech recognition technology achieves an accuracy of over 95%—in an ideal environment. In actual use environment, however, the accuracy rate will decrease. One of the reasons that manufacturers believe that affects the recognition rate is the unwanted noise in the raw data.

To solve the problem, VoiceCyber releasesed AI recognition rate enhancing solution, which processes the raw data by both hardware and software, to minimize mutual interference between neighboring call center agents, in order to provice clean voice data to the artificial intelligence engine, thus maximizing the artificial intelligence recognition rate.


Program Advantages

  • MiniNoise professional seat noised cancelling headset with directional microphone design – eliminates sounds outside a certain angle and accurately shields the sound of adjacent seats
  • Advanced noise reduction technology – eliminates ambient background sound, customer voice echo and other noises
  • Through the software to compare the characteristics of sound data—eliminating invalid sounds, delivering pure sound sources to artificial intelligence engines
  • Set hardware and software parameters based on the customer’s site environment – provide optimal data for artificial intelligence engines


Noise-reducing Headset Improves Speech Recognition Accuracy

For the same call, both the noise reduction headset and the normal headset are used for recording, and the two sets of audio files are sent to the same speech engine for recognition.

It can be seen that the recognition rate of voice data processed by the noise reduction headset is close to 100%. The recognition result of normal headsets not only appears words that did not exist in the original conversation, but also lacks a lot of call content.


Results Comparison

Sample Recognition Accuracy Statistics

Total number of words Correct number of words Accuracy
Original 65 65 100%
Noise Reduction Microphone 68 60 88.2%
Normal Microphone 79 59 74.7%

Through the comparison and analysis we can see that the use of noise reduction microphones in this sample increased the recognition accuracy by nearly 13% . There is no doubt that the dramatic improvement of noise reduction headphones will make it the best partner for artificial intelligence engines.


Comparison of Original Voice Files:

Noise Cancelled Headset Recording Sound

Normal Headset Recording Sound

Speech Waveform Comparison

Sound waveform of internationally renowned manufacturers noise reduction headset :

Sound waveform of noise reduction headset:

When both parties speak at the same time, the seat noise reduction headset almost completely eliminates the neibouring  seat voice and provides pure voice data to the voice recognition engine. When the seat is not speaking, the headphone eliminates the neibouring seat’s voice and avoids it from being recognized by the voice recognition engine.


Sound waveforms processed by VoiceCyber recording software : 

VoiceCyber​ also uses software which can completely eliminate the voices of neibouring seats, preventing uncorrelated voices from affecting the recognition rate of artificial intelligence engine and improves the overall recognition rate. The contact center can ask the agents to repeat the conversation to achieve best recognition effect.

Noise Cancelling Headset + VoiceCyber ​Recording System + Agent Speech Repetition is the best choice for artificial intelligence application in contact center.