The TeleMedia 900 Audiotel Case

TeleMedia is one of 2 companies in Egypt that provide AudioTel applications i.e. prepaid phone services and contexts that use 900 numbers based on closed publicity activity in the television.

The Situation

Based on Interactive Voice Response (IVR) systems used in Computer Telephony (CT), TeleMedia provides 900 numbers with special tariffs in cooperation with Egypt Telecom. Users have to follow recorded instructions and reply by pressing number keypads to select an answer or respond to instructions. Recording message isn’t practical and costs a lot with dynamic data and the use of DTMF instead of natural speech is tedious for the speakers in large menus.

The Challenge

The challenge was to deliver Arabic speech engines compatible with TeleMedia infrastructure, and able to automate the tedious and costly process of recording different prompts for various applications. Moreover the need to start a new human recording if the data is changed limits the scope of the application that the original system can handle. For example it won’t be able to provide dynamic listings of names or broadcasting of the latest news from the web.

On the other side, handling the caller requests in natural speech instead of DTMF is really challenging due to the variability of possible utterance that people can say in different ways even if the choices were limited. Moreover the probability of providing different speech than the expected and noise interference may limit the accuracy of any speech recognition system.

Finally this automation by Sakhr technologies should manage with the huge number of calls that TeleMedia receives specially in the peak hours of TV when publicity campaigns encourage people to call for a contest or service. The design should be flexible enough to toggle between different applications based on the online load and therefore route the service accordingly.

The Solution

We proposed a complete solution that solves all customers concerns and fulfill all the requirements including:

1) Automatic Speech Recognition (ASR): this engine was:

• Based on a complete SDK that enables the developer to build any context easily.
• Flexible enough to handle noise or additional utterance using keyword spotting.
• Able to handle concurrently up to 50 channels.
• Designed to manage continuous speech efficiently.

2) Text To Speech (TTS): Major dynamic recordings now can be substituted by the Arabic TTS that handles any un-Diacritized Arabic text efficiently, thanks to Sakhr’s automatic Diacritizer, in both intelligible and natural male or female voices.

3) Load Balancing: TeleMedia and Sakhr worked jointly to cover:

• Concurrent service covering online callers with optimized load distribution of resources
• Offline services generating TTS waves to be used by different requests latter.

The Results

The solution brought immense functionality and benefits on the two fronts: TTS used in dynamic prompt messages and users efficient ASR. The implementation of the complete solution satisfied TeleMedia requirements, with their actual heavy calls load with an accuracy of 95% in Automatic Speech Recognition and very high degree of intelligibility and naturalness in Text To Speech.