parkingcarma - getting local with speech
DESCRIPTION
Presentation for the SpeechTek Magazine conferenceTRANSCRIPT
Third levelFourth level
Fifth levelParkingCarma Helps with the Problems of:
• Congestion mitigation•Driver frustration• CO2 reduction
Speech
Distributed Application Topology
Third levelFourth level
Fifth level
Why Speech?
In-vehicle device penetration – new vehicles
Currently Low Percentage
Third levelFourth level
Fifth level
Mobile Applications & WAP• Apps require download •WAP experiences are not standard•Need data plan• Requires reading/navigating a device
while driving• “Hands-free” becoming increasingly
required
Third levelFourth level
Fifth level
87% Cell phone penetration rate in US
Source: CTIA
Third levelFourth level
Fifth level
Speech Application Overview
Third levelFourth level
Fifth level“Kate” Persona
• Inbound Speech Application
Third levelFourth level
Fifth level
Parking Facility Search
• Input location, or POI, or event• Return nearby locations that match user’s
preferences and are currently open• Return real-time parking space availability
at select facilities
Third levelFourth level
Fifth level
Parking Reservations
• Reserve parking at select parking facilities• Defaults to current day, but can change• Requires billing information to be entered if
not already a registered user
Third levelFourth level
Fifth level
Cell Phone Payment System
• Pay for parking with cell phone at select facilities• Enter in facility code• Call once for flat rate • Call to “check-in” & “check-out” at hourly
rate facilities
Third levelFourth level
Fifth levelSpeech Technology Platform
•Originally Microsoft Speech Server 2004•Migrating to OCS 2007
•Dialogic T-1 boards moving to SIP•Utilizing both Voice XML & Microsoft
Managed Workflows
Third levelFourth level
Fifth level
Speech Service Launched in December 2004
Third levelFourth level
Fifth levelBART – Bay Area Rapid Transit Application
• Advanced reservations and mobile payment•New ridership increased by 14% at station• Reduced circling by 5 minutes per car• Reduced VMT by 9 miles per user• Average 20 users per space (oversubscribed)
Third levelFourth level
Fifth levelSpecial Event Parking
• Advanced reservations & parking guidance•Offer multi and single day passes• Premium parking locations• Incentive of free movie tickets to use service
Third levelFourth level
Fifth level
One person had some very exasperating conversations with the voice recognition system, “Kate,” and has now mastered the strategy of “speaking softly and keying in…Kate does not like to be yelled at.”
Third levelFourth level
Fifth level
• 6% opt-out rate• 83% of calls in a 4 hour window Mon-Fri• 58% savings over equivalent call center reps
Third levelFourth level
Fifth level
Speech Application Challenges
Third levelFourth level
Fifth level
96% of all system users call from a cell phone
Ambient Noise in Decibels
= High false barge-in rate
Third levelFourth level
Fifth levelExtremely Large Grammars
• All street names in a city• All POI names in a city• All parking facility names in a city
Third levelFourth level
Fifth level
Large & Unstructured Grammars Lead to Incorrect recognitions
• Vanity license plates – “b”, “3”, “e”, etc.
• Similar POI and street names
High spike in peak volume due to daily weekday commuter use
Third levelFourth level
Fifth levelHow Hard Can it be?
Web Experience != VUI Experience
Third levelFourth level
Fifth level
Lessons Learned / Recommendations
Users will respond in ways you never expect
Monitor and adjust grammars accordingly
Third levelFourth level
Fifth levelSMS / Text Additional Information
•Directions to parking facility• Parking payment receipts
IVR System
Website
Use Consistent Credential Information
Separate web password and phone PIN == confused users
Third levelFourth level
Fifth levelCreate Dynamic Grammars
• Street names based upon city• Parking facility names based upon city• POI names based upon walking distance
radius of current location
Third levelFourth level
Fifth levelUse Patterns to Limit Expected Input• Custom license plate recognition• Each state has a pattern of alphanumeric• California = NAAANNN
• Patterns change over time• Increased non-vanity license plate
recognition by more than 25%
Third levelFourth level
Fifth levelResearch & Adjust Confirmation Needs
•More explicit confirmations• Before finalizing $$ transactions
• Less explicit confirmations• Experienced users
Third levelFourth level
Fifth levelUse Professional Voice Talent
•More professional• Pre-recorded “Vocabularies” save $$•Decrease development time
Third levelFourth level
Fifth levelImport Preferences from Web / CRM
•Default to “home” city•User’s favorite POI list• Preset filters for search
• Vehicle type• Price vs. distance
Third levelFourth level
Fifth levelMake Decisions for Users
• Based on “called-id” we select the city•Use Caller ID to identify users•Default to most popular search types• Always allows users to override automatic
selections
Other Lessons Learned / Recommendations
• Barge-in• Balance background noise with ability
for users to expedite calls•Use “recognition” rather than sound
barge-in and limit grammars
•Use Hosted Speech Vendor• If financially makes sense
Recorded Call
Third levelFourth level
Fifth level
Questions?
Christian McCarrickCTOParkingCarma, Inc.(650) 281.9864
http://www.parkingcarma.com