i need many recorded samples of word enunciated as 'go'; 'stop'; 'left' and 'right'. it is being difficult for me collecting samples; so i am requesting all forums members who are interested plz send your recording of these words in my mail given below:
Attributes of recording:
1) Software: wavesurfer or goldwave (i prefer wavesurfer)
2) Sampling frequency:16000 hz
3) Channel: mono or single channel
4) Duration: 1 sec (recording duration)
5) Format: .wav file
6) A single person should record: 'go'; 'stop'; 'left'; and 'right'.
7) Do the recording in a quite environment.
yes; i have collected many samples from my friends,.......,relatives;but i need huge samples around 200-300; that's why i am asking your humble assistance about this matter. i am doing a speech recognition project for home automation; the samples i able to collect around 88 yielding bad results.
My project supervisor emphasized on matter that neural network performance strongly depends on number of samples; i am getting good result for trained data; but not good for untrained data or testing data.
If u know any other reason plz suggest me; i am doomed stuck here.
Trained data are those which are used for training the neural network and and untrained data are those which are not used to train but use for testing or evaluating the performance of the NN network.