Audio analysis with Newtest
The Newtest Robot / NTBR 7.6 CU7 offers a new primitive for sound recording: StartAudioCapture. This function enables you to perform the following actions:
- audio capture and transmission of the file to NMC supervision in the form of a diagnostic (generic sending function for diagnostics)
- reprocessing of the recorded signal : deletion of blanks, re-sampling (8 kHz, etc.)
- detection of silences, for instance when monitoring streaming or video transmissions
In addition, on Robots that have a Newtest voice license, MOS PESQ analysis can then be used to assess the sound quality of a capture. This function is useful, for example, in evaluating the quality of voice calls on ‘softphones’ like Lync, Skype, or other unified communications client.
As a reminder, MOS PESQ is calculated by comparing a signal captured during a conversation (Test) to a file recorded beforehand, the ‘reference’ file. This algorithm is based on a model of the human ear in order to make an estimate of the audible components. Next, it compares these components to obtain the audible difference, which is submitted to a model comparable to that of the human brain, called ‘cognitive model’.
A full picture of quality can then be generated. The process is schematized below:
The advantage or this process is that it has a simple, objective scale for sound quality. The table below sets forth the scale as defined by the ITU:
The PESQ method makes it possible to retrieve the following secondary speech indicators as well:
- the noise index corresponds to the quantity of additional data (in frequency) when the degraded file presents an offset
- the omission index corresponds to the quantity of missing data when there is an offset with respect to the reference file
- the offset index corresponds to the delay between utterances
The Newtest Voice offer has long supported telephone calls, providing MOS and associated supplementary metrics (noise, etc.) for Robots with analog or digital boards (ISDN) and a connection to telephone lines for automated calls.
With the Newtest Robot version 7.6 CU7, these ‘voice’ monitoring functionalities are now possible without a dedicated voiceboard, and allow for ‘softphone’ use (or unified communications client) like Lync, Skype, Jabber, and others, connected to an IP network.
However, the conditions for recording via sound cards are different from those specified by the ITU PESQ standard for telephone lines. Therefore the values cannot be compared with landline telephone MOS values, but rather provide an indicator that can be used to follow variations over time or compare quality at different locations.
As usual in the Newtest portal, real-time supervision views show the performances of call setup and quality. The graphical view in NMC is shown below for the example of a Skype call scenario: