Speech Services

LPaaS

LPAAS (Language Processing as a Service) offers you services which allow you to continuously monitor and control the performance of the speech recognition implemented in voice services (voice portals, qualifications in call centers, etc.) It's a process for monitoring quality of service, based on an industrial approach, implemented by OnMobile's Professional Services teams.

Using LPaaS ensures that you have a high quality voice service which will stay optimal over time (on average for 2 years). In particular it covers a statistical commitment on the rate of successful transactions, for the dialogues based on speech recognition.

LPaaS takes into account not only the evolution of the service, but also the behavioral evolutions of the callers; therefore it ensures that your service stays optimal over time.

Although it is based on an industrial approach, the implementation of LPAAS is extremely flexible and can be adapted to the specific requirements of each voice service, such as, for example, the frequency of realization of each analysis cycle, the criteria for improving the quality of service, or metrics definition.

Detailed description

The LPAAS process is based on a series of improvement cycles; each cycle is made up of five key steps:

Live Service data collection

These processes feed a dedicated platform which deals with language development on one hand, and voice services optimization on the other hand. This platform is used by our language teams worldwide

Manual Transcription

Over and above the transcriptions of the words and phrases pronounced, our experts also integrate noise markers, such as for surrounding noises, background speech or an echo which can be heard on the live services. Thus, LPaaS allow you to optimize your services in the manner the best adapted to speech recognition in the context of your service

Analysis & Testing

In this phase, the users' behavior is analyzed at each stage of the dialogue and the main areas for improvement which need to be implemented are clearly identified. This work leads to a global improvement plan for the Quality of Service which could include:

• Ergonomic improvements
• Improvements to the audio guides
• Detection of new reasons of requests by users
• Adapting the speech recognition to the users' utterances noticed on the service
• Considering new content

Usage adaptation & Acoustic Tuning

Thanks to this phase, we can guarantee a 30% improvement in the error rate observed on a service. The LPaaS service optimization tools will allow you to measure and quantify the benefits of each optimization thanks to batch tests in conditions identical to those of the live service. This approach guarantees the individual quality and the non-regression of the different grammars and models which are optimized and put into production; this is the case at the end of each LPaaS optimization cycle

Reports, Statistics & Suggestions

Each report precisely describes the distribution of calls on the service, the different areas of improvement which have been identified, and recommendations for developments which need to be planned for the service as well as the statistical performance levels reached before and after the optimization. Live call tapping of your services will give you precious indicators and feedback from the field, which you cannot obtain from automated statistics:

• Detailed diagram showing the distribution of calls
• The sound environment of callers
• Callers' stress levels
• Callers' reactions to the ergonomics of the service
• Detection of requests by callers which out of the usual scope of the service

References