| Abstract [eng] |
VoIP (voice over IP protocol) is quickly growing and expanding technology. Calls and other services are much cheaper than traditional ones. Nevertheless it meets some quality problems such as packet delay and packet loss. The object of this paper is to explore voice transmitted by IP channels quality and to conclude about if that voice could be used in voice recognition systems. To complete this object, there are 4 main tasks: • Find out the technologies and working techniques of the VoIP. • Determine tools for quality analysis of the transmitted voice • Build an algorithm which could help to extract useful data from voice signal. • Summarize the results and conclude about ability to use transmitted voice in voice recognition systems. There was made 150 transmissions („Skype“ system was used) of the voice during the experiment and all of them was recorded. Fast Fourier transform was used to transform those recording to frequency scale. The quality loss was analyzed; original voice record specter was compared with transmitted voice record specter. The transmissions were divided into two parts. Transmissions was made in 2008 February (1st part) and 2008 May (2nd part). In the first part the results are almost the same, quality variations are stable (specter difference average is 1,44), all values are between 1,44 and 1,55 (specter difference average value). It was noticed that transmitted voice signal has large quality variations in the second part of the experiment. The quality loss is not stable (specter difference average values are between 0,36 and 1,45) and is very different in the second part. Nevertheless the average value of all specter difference of the 2nd part was 1,09. This means that quality of the transmitted voice got better. If we want to use voice transmitted by IP channels in voice recognition systems, we should stabilize signal quality loss variations or create model witch could recognize those variations and adjust voice signal for voice recognition. |
