In-Vehicle Speech and Noise Corpora - Digital Signal Processing for In-Vehicle Systems and Safety - page 142

compiling speech variability for an in-car task along with the environment
variability encountered for different task scenarios. This corpus consists of

2 phases:

• Phase I: Speech & speaker data collection
• Phase II: Acoustic noise data collection (CU-Move Noise)

9.3.1 Phase I: Speech and Speaker Data Collection

The speech and speaker data collection is divided in two sections. First (part 1) is
structured text where the user is prompted to utter text and numbers similar to what
is observed in a command and control application. The second section (part 2) is a
dialog system scenario with a real person on the other end.

9.3.1.1 Part 1: Structured Text Prompts

The driver performs a fixed route that includes a combination of several driving
conditions (city, highway, traffic noise, etc.). For each speaker, prompts were given
for specific tasks listed below from a laptop display situated around the glove
compartment of the vehicle. This portion is 30 min long. There are four subsections
that include:

• Navigation direction phrases section: a collection of phrases which are deter-
mined to be useful for in-vehicle navigation interaction (prompts fixed for all
speakers)
• Digits prompts section: strings of digits for the speaker to say (prompts
randomized)
• Streets/address/route locations section: street names or locations within the city;
some street names will be spelled, some just spoken (prompts randomized)
• Sentences - general phonetically balanced sentences section: collection of phonet-
ically balanced sentences for the speaker to produce (prompts randomized)

9.3.1.2 Part 2: Dialog Wizard of Oz Collection

Here, the user calls a human “wizard” (WOZ) who guides the subject through
various routes determined for that city. More than 100 route scenarios particular to
each city were generated so that users would be traveling to locations of interest for
that city. The human WOZ had access to a list of establishments for that city where
subjects would request route information (e.g., “How do I get to the closest police
station?”, “How do I get to the Hello Deli?”). The user would call in with a modified
cell phone in the car, which allows for data collection using one of the digital
channels from the recorder.