Audio Scene Classification- FP-AI-SENSING1- WITH 6 NO. OF AUDIO FILES

[ShivaRagu]

Hi,

I am trying to test the ASC with 6 classes and 6 audio files. Am modifying the same python script from the function pack and trying to create the Tensorflowlite model. Is that possible to add more classes or only we can use 3 classes?

If we can use maximum 3 classes, but am trying to convert 6 Audio files into model file, then what would be the option to procedure?

Thanks & Regards,

Shivashankari

[niccolò]

Hi shankari @ShivaRagu​ ,

if you want to add 3 classes I would make a new model from scratch.

the ASC was trained with a very big dataset and if your dataset for the new classes is not comparable there can be problems.

Niccolò

[ShivaRagu]

Hi,

Thanks for the reply.

I can see you have used 30 sec audio files (bus, home and park) for model creation, but also mentioned we need to use complete dataset from "TUT ACOUSTIC SCENES 2016" for actual application which has more than 20gb data. Can you tell me how to compress the data into 30sec. How you got that 30 sec audio which u have used to create the model for ASC?

If I want to try ASC for speech recognition, then what is the minimum dataset I need to use?

Regards,

shankari

[niccolò]

Hi shankari @ShivaRagu​ ,

the 30sec audio files are for testing the network, they are not the ones used to train it.

I developed a network for sound recognition using something like 15G of audio data.

for speech recognition I don't really know if there is a defines limit, I would suggest to search for online research paper on the subject for that.

Niccolò