anyone fancy helping me with my dissertation (test samples needed)..

[5lab]

h'ok.. i should probably have realised i needed this sooner, but anyways..

i'm doing a dissertation (due in on thurs :S) which is trying to work out emotion in voice. for this i've implemented a database which stores voice samples which are labled up (with emotions) and then you stick in another voice sample and it works out which emotion it represents. unfortunately i've got a database which now needs to be filled with training data.. sooo...

i need your voice sample.. what i'm after is clips of voice, around 20 secs long (bit longer or bit shorter is no problem, but shorter than 10 secs or so might confuse it a bit..). I need a 'standard' sample (of you just talking in a normal voice, about something boring, like what you did at work on friday or something). then, in a separate sound file, i need a sample which is either angry, excited, or sad.. if you could do this for each of them, it'd be great, or if you've not got the time, a single one will be helpful too. the results will work best if the subject you are talking about makes you feel that way - so for instance, talk about an online company messing up your order for the angry one, some loss of some kind for the sad one.. that sort of thing. the words themselves aren't important, its how you're saying them..

the sound files need to be 8khz, 16 bit, mono audio. recording in more detail is perfectly fine, but if you record in less, the program might get confused again. to record some sound go to start-programs-accessories-entertainment-sound recorder, and to set the audio details go to file-properties-convert now. select 'pcm' in the 'format' box then in attributes select '8000 khz, 16bit, mono ............. 15kb/s' or the nearest equivilent.

if possible, keep the sound files as wav (not mp3) as this will help preserve some of the audio.

if you're still with me, and havent decided that you cba, email the audio (labled with the emotion on each file), to hughlunnon@yahoo.com - if you are feeling extra nice, include your age and sex, cos that'll go into the 'i tested 10 blokes aged 10-50' section..

erm.. so yeah, thats it. uber-thanks to anyone who does this for me. you can each have a free thingy of your choice from my bag-oh-pc-bits. send what you want and i'll see if i've got it (i've got a dead mobo, some fans, loadsa cables, some grills, dust filters, ya name it..)

should only take you 2 mins to do despite sounding complex. and you'll get a mention in my disso if you include yer name.

again, thanks for doing this if you do. if i've not been clear about something then please ask

Hugh

[yamangman]

Sounds interesting. Whilst everyone is away scrabbling for their mic's, can you relay anything on how you've acheived this?

[5lab]

sure. what do you want to know? basically the system i've implemented picks up on cues in the voice, such as trembling, word rate, differences in volume, the raising, lowing and ranges of pitch and so on.. whilst 'learning' it is given a series of voice samples that are already labled - so it can understand what an 'angry' voice typically sounds like, in comparison to the same speaker using a 'normal' voice. when trying to work out an emotion it looks at the new voice in comparison to that speakers voice, looks at the differences and works out the emotion based on what those differences meant when other speakers were speaking..

if that makes sense??

hopefully one day this sorta technology could be used in voice recognition systems on a phone line to see if it can understand when someone is getting angry/frustrated with the phone system and put them straight thru to a rep..

[Lee H]

Still need more samples ?

If so I can get one done over the next day or two if required

[Wyoming]

I'm working on some samples, they should be with you within the next hour.