Speech recognition with the Deutsche Bahn

1

• Speech recognition with the Deutsche Bahn

Spotlight colloquium byMatias MusmacherDB
Dialog Telefonservice GmbHSolmsstraße 71 60486
Frankfurt Telefon 0 69 - 478 96 750Fax 0 69 -
478 96 709 London, March 2001
2
Agenda
Goals for the project
Concept
Introduction of the system
Usability trials
Lessons learned
3
Goals for the project

• The customers were dissatisfied with the Deutsche
  Bahn.

Fernmündliche Reiseauskunft
Ansagedienste

• about 1,7 million calls per year
• local number 011581-011587
• service for the 60 major cities in Germany

• about 25 million calls per year
• local number with access code19 4 19
• 72 telephony rooms
• about 1200 employees

• Introduction of an IVR-System

• Integration in 12 Call Centres in 1999

4
Goals for the project

• The recorded messages had to be improved in three
  main areas.

5
Goals for the project

• Deutsche Bahn has six major goals for the
  project.

• Create a general acceptance for a speech
  recognition system
• Give the customers the correct timetable
  information
• Choose a system for better service
• Introduce a reliable system
• Keep it simple for the IT-department
• no sunk costs

6
Goals for the project

• The Deutsche Bahn made a decision for a query
  based system.

Methods of interaction
Costs
Freies Sprach- dialogsystem
Freie Abfrage Reisedaten
Menüabfrage Reisedaten
Timetable depending on time of call
Chosen method
Status Quo
Customer value complexity
7
Concept

• We took a conservative approach within the
  project.

8
Concept

• From the marketing point of view the IVR is today
  positioned as a stand-alone service

reasons

• Clear position as a substitute for the
  Ansagedienste
• Launch as completely new service
• Introduction with only a small negative image
  transfer in case of failure
• Integration with the call centre service is
  possible in the future

9
Concept

• The design phase was about implementing the
  dialogues for the prompts.

• Start with an explanation of the system
• Dialogues structure similar to the web
• One word utterances by the customer
• Help function in case of misrecognition
• Call transfer to a call-centre agent in case of
  three misrecognitions in a row

Process

• Strict call for action
• Whole sentences
• Text based on customer language

text
10
Concept

• We then developed a request for proposal.

11
Introduction of the system

• Technical architecture

The system consists of 4 crucial
parts 1. Dialogue platform 2. Speech recognis
er 3. Online database query to the mainframe
4. ACD of the call centre
Call Centre
IVR-System
Dialogue
caller
Phonem based recognition
01805-996622
Train connection
Erkanntes Wort
Recogniser
Platform
Train connection
query
DB-Mainframe
12
Introduction of the system

• The project consisted of two teams, one from the
  business side and one technical.

• decision about the number and names of included
  stations
• decision about possible utterances and phrases
  (London Liverpool Street vs Liverpool Street
  London)
• decision about foreign stations
• Recording of the station names by employees from
  all over Germany
• Connection to the mainframe

DB-insider
13
Introduction of the system

• The project consisted of two teams, one from the
  business side and one technical.

• Implementation of the dialogue by the vendor
• programming and training of the recogniser
• connection to the ACD via DPNSS
• Recording of the prompts

partners
14
Introduction of the system

• Example dialogue

Elements
Prompts
Caller utterances

Herzlich Willkommen beim ReiseDialogSystem der
Deutschen Bahn! Wenn Sie schon mit dem System
vertraut sind, sagen Sie weiter.
Weiter.
shake hands
Bitte nennen Sie Ihren Abfahrtsbahnhof. Bitte
nennen Sie Ihren Zielbahnhof. Sie möchten also
von Bremen Hauptbahnhof nach Hannover
Hauptbahnhof fahren?
Bremen Hauptbahnhof. Hannover
Hauptbahnhof. Ja.

Station prompt

An welchem Tag möchten Sie abfahren? Um welche
Uhrzeit möchten Sie abfahren? Sie möchten also
am 20. März um 0700 Uhr abfahren?
20.03. 700 Uhr Ja.
Date and time prompt
Es gibt die folgende Verbindung Am 20. März mit
Inter Regio 2581 um 0709 Uhr von Bremen HBF,
Ankunft in Hannover HBF um 0814 Uhr. Möchten Sie
früher, später oder wiederholen?

Weiter.
Train connection

Sie können nun einen der folgende Punkte
auswählen Rückfahrt suchen, Zeit eingeben,
Verbindung suchen, Ende. Vielen Dank für
Ihren Anruf. Wir wünschen Ihnen eine gute Reise.
Auf Wiederhören.

Ende.
termination
15
Usability trials

• The system was tested by an external market
  research company.

A. Studiotest with 200 persons (group of future
customers) in Munich and Cologne
B. discussion in two rows (Frankfurt)
C. test of the speech recogniser
16
Usability trials

• Findings

17
Usability trials

• The people were asked about their satisfaction
  and the ease of use.

18
Lessons learned

• The introduction of the system led to mixed
  reactions within the public, the customers
  reacted calmly.

• high percentage of frequent users ( about 15)
• 2000 callers per day without any communication
  campaign
• call transfers rate at 2 percent
• harsh reaction from the yellow press
• tests from the media show criticism in both ways
  ( good system vs. stupid machine)

19
Lessons learned

• Because of the mixed reactions we decided to
  improve the system and put it up to a new
  usability trial within spotlight.

• Integration of all 7800 German stations
• Optimisation of the dialogue
• Create a general acceptance for speech based
  recognition in Germany
• Look into dialogue of other players in the
  industry

20
Lessons learned

• The usability trial in November 2000 led to
  interesting results forthe Deutsche Bahn.

• System is seen as an alternative for direct
  train connection, the length of the train
  connection prompt is the key influence for the
  efficiency
• system is too complex for people with a low level
  of education
• 80 of the calls were successful
• a huge majority of people prefer the call-centre
  service to the IVR system
• There were less problems with the date and time
  prompt
• people were missing information about prices and
  platforms

21
Lessons learned

• Deutsche Bahn will take a conservative approach
  to develop the next services.

Integration with call centres
time
22
Questions
?