TeleBanter is a mobile conversation game where anonymous callers are randomly connected together to share in a conversation and then rated based on "good" conversation characteristics. Access to the Internet and any mobile phone are the only things needed to play. Interested participants are connected based on time availability to have a conversation and shared topic interests. The conversation will be monitored for analysis of "good" conversation characteristics and then given a score based on both participants. Both the caller and receiver will also have an opportunity to rate each other's conversing skills. Afterwards, the participants are encouraged to visit the TeleBanter website to see their recent conversation performance and the scores of the top conversationalists.

Background
Over the centuries, we have developed complex grammar to languages, but “good” conversations are not bound by the proper use of syntax and semantics. “Good” conversation are not exclusive to economic class, education level, or different cultures. They are shared by everyone. The early stages of research started at oddly enough, grocery stores. Everyone has to eat and living in Manhattan the grocery check out lines can be a nightmare. Some of my fondest conversations started in these long lines. There are two things I shared in common with everyone in the checkout line: 1.) We are buying food. 2.) We do not like waiting in line. Why not pass the time waiting and share in an interesting conversation.

It is quite common to walk the streets of New York City and overhear several languages representing it's cultural diversity. I live in the primarily Hispanic neighborhood of “Alphabet City” and have the opportunity to listen in on conversations in a variety of locations but typically at the laundromat, grocery store and the park. These conversations are taking place everywhere. Since I do not speak Spanish, the only thing I am able to infer in the conversations are the vocal tones made up of pitch, amplitude and rhythm between the two speakers.

My interest in this project is driven by my curiosity as to whether “good” conversations share similar characteristics, though it may seem to be subjective at first glance. Listening and trying to understand a foreign language spoken in a conversation was valuable research since context is ignored, rather not understood.

“Good” Conversation Characteristics
I was invited to share an evening out with my friends and it turned out the other invited guests were visiting from Austria. After placing our orders, everyone was eager to catch up, and they all started to converse in German. I listened carefully and noticed that the most exciting parts of the conversation were during the high frequency of turn-taking between the speakers. These were peaks of interest, where all the people were actively involved in the conversation. Each speaker contributes to the conversation by sharing their own insight built on from previous speakers. This is the first “good” conversation characteristic TeleBanter will be analyzing for.

The second characteristic I am studying is the well balanced turn-taking between callers. When one speaker's amplitude sustains for a period of time then drops and the second speaker's amplitude rises and sustains for a period of time constitutes a "turn." Since all conversations are equal length of five minutes, the amount of “turns” taken can be compared to other recorded conversations.

The greater portion of research was to break down the characteristics of “good” conversation so that the software would be able to perform the vocal analysis and score a given conversation. Great detail of conversations can be found using only the vocal tones (pitch and amplitude) produced by both speakers. Timing and rhythm uses the silence. All “good” conversation share a well-balanced exchange of words between two people. The total amount of times of each “turn” divided by the fixed conversation length of five minutes is how the computer would be able to score a given conversation and be compared to other conversations. The second characteristic will analyze the dynamic frequency (pitch) range of the speaker. More points will be awarded to the speakers that have a greater frequency range. This characteristic demonstrates more vocal variation, therefore likely signify a more interesting vocal pattern for the listener. This will computed by analyzing the variation in intonation in spoken words. Attentive listening is necessary to foster an interesting dialog. The last characteristic measures utterances made by the listener, which measures attentiveness and reassurance to the speaker that someone is listening.

Audience
Currently this game will only be available in English, so anyone who is able to speak English will be able to participate. The most interested callers I anticipate using TeleBanter are people who want to sharpen their conversing skills, talking to random people, and foreign students (ESL) working on their English.

User Scenario
The current version of TeleBanter is made up of a website and an Asterisk phone application. New users will need to create a new account online and submit a mobile phone number, time availability and topics of interest. No detailed information is given about the scoring to the participants. Once all the information is in the database, the user is given a number to call to initiate a random conversation. When the user calls, the Asterisk phone application will check for a match in time availability with users in the database. Once there is a match, TeleBanter will call the random caller in the database to be connected with the caller. The phone call will be recorded for vocal analysis and scoring of the conversation. Pitch (frequency) and volume (amplitude) are the two vocal characteristics that will be analyzed during the conversation. Since both caller and receiver are on separate phone lines, the audio I record for analysis are separated. This offers the possibility of a more accurate speech analysis when the caller and receiver happen to speak at the same time. Once the caller and receiver finish their phone call, they can return to the TeleBanter website to view their score to their conversation and a visual map visualizing their conversation. Since all conversations are five minutes in length, all conversations can be compared to each other.

Implementation
TeleBanter is made up of a website with a database and an Asterisk-based phone application. Phone numbers and time availabilities will be stored for callbacks and the phone application will record and analyzed for good conversation characteristics. The website will list personal scores and the top conversationalist's scores. Max/MSP patch is used to analysis each person in the conversation for pitch and amplitude over the course of time. The numbers for the patch will be used to visually map the data into a Processing application.

Conclusion
TeleBanter's current form analyzes conversation characteristics based on vocal tones characteristics: average pitch and amplitude. The human voice (in particular) is made up of a range of frequencies, also known as harmonics. In the next iteration of TeleBanter, I intend to use the Fast Fourier Transform (FFT) analysis to acquire the whole frequency spectrum at any given time. Another interesting avenue to explore is contextual analysis. Conversations can be dissected by the topics and the words that make up the dialog. Once speech-to-text technology has improved, contextual analysis would be incorporated into the scoring. Plans to develop a social community such as forums are considered to possibly reconnect with past anonymous conversations and share in conversation tips.