In the first stage of this cycle, you simultaneously develop and test your grammars independently from the use of your voice service.

In the second stage of the cycle, you improve recognition performance via phonetic and acoustic optimisation.

So the teliSpeech Quality of Service Management tools can be the basis of the definition of a commitment between our partners and their clients, since the tools ensure that they can master the work to achieve given performance targets.

The quality of a voice service can be measured at each stage of its life cycle :

• development produces the source code of the grammars, the words that need to be recognised, on the basis of the dialogue specifications
• testing validates the grammars one by one and then integrates them together outside the target platform
• data collection is essential to define non-regression tests and obtain pertinent results during the performance optimisation stage.
• evaluation of recognition rate must be carried out through preliminary tests until the follow up during production; this is the cornerstone of the monitoring metrics and the Quality of Service Management.
• phonetic optimisation improves the precision of the phonetic definitions. You can create your own lexicons composed of words with particular pronunciations.
• acoustic optimisation improves the acoustic representation of the words in a grammar on the basis of collected speech data. This is the ultimate step in optimising performance with teliSpeech.

• load balancing steers decoding queries by default to the least used server, thereby reducing the required server resources
• hot plugging allows the system to automatically recognise a new server the moment it is put into use. The absence of service interruption is a guarantee of scalability.
• high availability, in case a server breaks down, this ensures that application queries are automatically sent through to the available servers without interrupting the service or losing any queries.
• cache management speeds up access to data, thereby improving the response time of teliSpeech.

You can integrate teliSpeech on a VoiceXML or IVR platform by choosing the most appropriate mode for your use: either the Native Programming Interface or the standard MRCP protocol (Media Resource Control Protocol).

• native programming interface:
  
  
  when using this mode, a telephony platform implements teliSpeech functions through one single client programming interface, thereby offering a simple and fast integration procedure. This interface can be seen as a distant access to teliSpeech resources, the compiling and decoding servers.
• MRCP protocol:
  
  
  The teliSpeech MRCP server offers a standard interface for any partner platform supplier who has a generic MRCP client. With this mode, teliSpeech becomes a black box, with transparent evolutions, which is accessible through an IP network. In its present definition, MRCP does not cover certain advanced recognition functions. If none of these mechanisms are necessary for the deployed voice services, the MRCP interface stands out as the most appropriate integration mode, by drastically reducing the validation time and support effort.

A flexible administration environment!

The ease of use and the possibility of live monitoring of teliSpeech in a network infrastructure are important success factors. TeliSpeech allows for a more flexible and continuous integration of voice services in your strategy through SNMP-based management and close monitoring of used services.

• SNMP-based use monitoring:
  
  A SNMP - based monitoring programme supervises the status of all the teliSpeech processes used, with each teliSpeech server on the network being able to send a SNMP alarm.
  
• a precise analysis of your voice services used :
  
  TeliSpeech implements log management, a journal of activity which logs every operation performed by teliSpeech for a voice service. With these logs it is possible to analyse every teliSpeech process carried out on a distributed environment. It is even possible to analyse all the interactions that occur during one single call. A single configuration file sets the parameters for the different levels of logging/analysis and defines the vocal service surveillance strategy.