My project code

The apk is usable. The classes won’t execute because of confidentiallity with Tunify.

http://www.2shared.com/file/UWWs2HwD/code_activ_music_droid.html

 

Temp thesis report

A temporarily report of my thesis. Still have to write about the final evaluation :). Maybe I will introduce one more evaluation with another accelerometer formula…

http://www.2shared.com/document/Dp0pBINM/masterproef.html

poster

my poster can be found here : http://www.2shared.com/photo/gaYW60Fo/poster_template.html

2nd draft scientific paper

My scientific paper can be found here : http://www.2shared.com/document/qn5ADVm3/goede_paper.html

second presentation

the slides can be found here : http://www.2shared.com/file/bInSuiXe/thesispresentatie.html ( dutch)

What have I been up to?

Over the last 2 weeks I have improved my detection algorithm. In a first step I have implemented that the last activity is instantly started so the user can hear music instantly instead of waiting for the algorithm to detect the activity. After the first song the algorithm starts detection. Gps detection can be done while playing the first song but the audio sensor data can only be retrieved between 2 songs(otherwise the music can interfere with the data or I have to assume the user is wearingheadsets). Than 3 songs starts playing and after 3 songs the algorithm rechecks the activity to see if it’s still the same or not.  All this is implemented except the gps listening between songs but this can normally be done in a thread. This new algorithm will surely improve the correctness but I have to do some usertests to see if it doesn’t interfere with the user interaction.(maybe the waiting time after every 3 songs isn’t worth the improvement of correctness) The next thing I did was rewrite my paper for the conference in april. I sended in a second draft to my promotor yesterday. Last week I also have been working on my report. I hadn’t wrote anything so it was time to start. At the moment I’ve already written quiet a bit  but I still have a lot of work. This week I am in Paris with a very slow internet connection and I’m in a very intensive course so can’t do much work but I will make up for this next week. I hope I will be able to finish my 2ndthesisreport and presentation :). In the easter vacation I hope to finalize my detection algorithm(implement the gpslistener on a thread) and prepare my application for a total evaluation. I will also continue writing because I have been neglecting this part of my thesis for quiet a bit 🙂

 

 

Detection algorithm tests

Problems:

–          No transitions possible, once an activity is detected it will stay this way forever except if the user manually scans again.  ==> Solve by redetect an activity after x songs automatically

–          Sporting on a home trainer/in the gym is not detected.(while not moving) (no gps signal available automatically means that the application thinks you’re studying/relaxing inside) ==> Solve by using accelerometer (can be a lot of work/research to get this working)

–          Relaxing sometimes gets detected by studying when you’re in a very quiet room. No idea how to solve this yet. Maybe use time as an extra indicator. After 8 pm the student is probably relaxing and not studying anymore so I can use this as an extra indicator.[assumptions… ]

–          Studying sometimes gets detected by relaxing when the cellphone is placed down very loud. This is because I  make use of an upper sound bound in my detection algorithm. I can remove this and only work with the averaged values to detect sound better.

–          In general I can also lengthen the sound detection interval a bit to increase accuracy in detection and better average out these ‘peak values’.

-Still to test: Transporting. Movement speed seems like a good indicator + my test data shows that this is a good indicator but not tested yet 🙂

These are the problems I will take with me in this second iteration. I’m not sure if I should solve them all (can take a while …) or just take the most important one and solve these first before I start working on the ‘less important’ ones.

Changes I made(so far) after 1st digital usertest

• In CW building and sometimes during usertests the gps sensor wouldn’t find a signal. I’ve added a TIMEOUT so that after 15 seconds without a signal , the application assumes that the user is inside a building.(must still research timeout time + assumption) 
• During usertests 8/10 users started out by clicking the SCANbutton at the start. Because no genres are labelled then, the application showed a toast telling the user to label some genres to the activity. I solved this by automatically linking every genre to each activity when the users uses the application for the first time.
• The logout buttons I had let the application crash sometimes because of an unstable activity stack. I solved this problem and users are now able to use these logout buttons correctly.
• The total songs number in statistics menu was calculated incorrectly. I fixed this and made also some other minor changes to the statistics.(all songs button wasn’t visible in horizontal orientation etc)
• I gave my application a name (Activ Music Droid) …. 🙂
• … the rest of this weekend I will spend researching my usertests to find some more core problems that need fixing

The SUS score reveiled 3 problems that my application has at the moment. 5/10 people don’t see themself using my app frequently.(can be a implementation problem or they don’t like the concept)4/10 thinks that my application is a bit too complex. I tried to solve this a bit by labeling all genres to each activity at the start so the users can immediately hear music instead of searching through the application to find the labeling mechanism. The last problem was that the various functions weren’t well integrated. My statistics were wrong, the gps functionality could go in an infinite loop for waiting for a signal,maybe the playlist seems a bit odd… Most of these problems are solved but I will definitetly look out to see if these results change positive or not with the changes I made.

The last thing that I will implement for this second iteration is a mechanism that immediately starts playing music based on the previous activity ,after or during(if possible) a song the algorithm will correct the activity….

new detection algorithm idea:

 

1: predict activity at start by using the previous used activity , at first use of application just scan like it happens now. After first song (or during if possible without songlag) recheck activity with audio and gps data and correct if needed.

2: If the activity is relaxing or studying , scan after every song for 3 à 5 seconds the audio to change the activity if needed….(while buffering the next song) This must not be done while sporting or travelling because these don’t rely on audio. Scan after every 5 songs the gps to see if the student is not sporting/traveling (if the sound is still in studyingmode we don’t need to do that because sporting and traveling gives always more sound … (should test this but pretty sure))

3:  If the activity is traveling/sporting recheck this with the gps every 5 songs and correct if needed.

first iteration digital prototype usertests

Test set: 10 test persons

Profile: 7 male   3 female

Experience: 1 of the female already had some smartphone experience

5 of the men already had some smartphone experience

For my user tests I tried to find a good mix between experienced people and people with no experience at all. Both groups should be able to work with this application and understand the functionality.

Age	# persons
19	1
20	1
21	3
22	3
23	1
48	1

The graph below shows the frequency of listening to music in which 1 means close to never and 5 means daily. I tested only people that listened at least weekly to music.

The graph below shows the favorite music player of my test persons.

1 : Windows media player

2: Itunes

3: Spotify

4: other(Youtube/winamp…)

For testing purposes I made use of the think-aloud protocol. I didn’t made use of use cases. I just let the users play with my application after a short introduction and meanwhile checked if they were able to figure out all possible use cases by themselves. I made use of my first digital prototype of this application that can be found on this blog. Some test persons needed a bit more time than  others but in the end they were all able to figure out the total functionality of the application after 3 minutes. After the testing round I asked my testpersons the folowing questions.

note: I specifically tested on the User Interface/program flow etc. Not the detection correctness. This will be tested when I updated my algorithm a bit 🙂

Questions asked:(with responses summarized)

After each question is a number containing the total number of people that +/- gave this answer or said something similar. Because the questions were open there were people that said multiple things ,that’s why not all answers on a question add up to 10/10 but can exceed this.

1. Did you encounter any bugs in the application? Which one?

• Crash when next/previous song button is clicked very fast.(7/10)
• You can only navigate through songs with the next/previous button. Would be easier to navigate by clicking on a song in the list.(1/10)
• You can’t close the application(5/10) only the people with smartphone experience noticed this.
• Sometimes crash when the orientation of the phone changes(restarts activity)(5/10)

1. Could you easily navigate through the application with the tab bar menu? Any other advice on navigation?

• The tab bar navigation works good(9/10)
• Instead of clicking on the tabs use a slide function to slide through the tabs(1/10)

1. Was there too much user interaction needed to start hearing music? (Application doesn’t start scanning automatically/The user has to label each activity at the beginning )

• It would be better if the music would start playing immediately (last activity and then correct activity after/during playing of song)(6/10)
• Labeling works good and intuitive(4/10)
• Omit genre labeling and just work with standard assumptions (studying: quiet/slow music) (sporting: high bpm music) even less user interaction needed(Aristo)

1. Was the waiting time for the gps signal disturbing?

• Not that much. You can just start the application and put your smart phone away. The music starts playing eventually and you don’t need to check it the whole time.(6/10)
• Work with an upper limit of time to wait for the gps to find a signal(1/10)
• Just scan while music is playing … not disturbing at all then.(4/10)

1. Would it be good to make an initial “activity guess” so the application can start playing music instantly without the sensor data waiting time? For example, automatically start the activity the user had last time he opened the application.

• Yes this delays waiting time for the sensor data but since music is playing it isn’t that disturbing. You can implement that the last activity detected starts automatically and make sure that a labeled song is first in the playlist (this way the user will surely like the song , even if it doesn’t belong to the real activity at the moment)(7/10)
• No, it breaks the whole purpose of your application of offering the right music at the right time.(3/10)

1. Were there enough genre choices for the music playlists?

• Yes, but it is always nice to add more(9/10)
• A bit too restrictive (work with bpm etc)(1/10)
• Don’t make use of genrelabeling but hardcode this so the user doesn’t has to do this interaction with the application (Aristo)

1.  Is it necessary to show the whole list of songs that are in the playlist or can it be more like an IPod shuffle where you don’t need to know the order of the songs.

• It is not necessary but it isn’t disturbing either.(3/10)
• Helpful when you’re hearing a song and want to label it instantly(5/10)
• Helpful if you make navigation through songs more easier(click on them to play or something like that)(2/10)

1. Would you use this application to listen to music? Why/why not?

• Yes, it is fun to discover new music.(6/10)
• Yes, not much user interaction needed to create a playlist and hear music you like.(1/10)
• Yes but not as a standard music player. More for something when your laptop is not around.(3/10)

1. Which problem/update should I definitely handle in the next iteration?

• Disable genre labeling(1/10)
• Start playing music immediately (guess activity at beginning)(4/10)
• More specific labeling (than just genres)(1/10)
• Better navigation through the songs in a playlist(2/10)
• Integrate activity-detection and playing music better (detect while playing music etc)(4/10)

I ended the testing by presenting my testusers a SUS-test. The results were pretty decent.

scientific paper(first draft)

The first draft of my scientific paper can be found here: wetenschappelijk . There is still no section about usetests that I will definitly include in my final paper. There are also some figure problems that I will have to discuss with my advisor next week. 🙂