TFLite Android

[zeynepgulhanuslu]

Hi,
I want to use tflite model in android project. When I load model to android studio it generates a code like below:

`
val model = Dtln.newInstance(context)

// Creates inputs for reference.
val inputFeature0 = TensorBuffer.createFixedSize(intArrayOf(1, 1, 512), DataType.FLOAT32)
inputFeature0.loadBuffer(byteBuffer)
val inputFeature1 = TensorBuffer.createFixedSize(intArrayOf(1, 2, 128, 2), DataType.FLOAT32)
inputFeature1.loadBuffer(byteBuffer)

// Runs model inference and gets result.
val outputs = model.process(inputFeature0, inputFeature1)
val outputFeature0 = outputs.outputFeature0AsTensorBuffer
val outputFeature1 = outputs.outputFeature1AsTensorBuffer

// Releases model resources if no longer used.
model.close()
`

My question is what is the inputFeature0 and inputFeature1 in this code? Should I read wav file as byte array than reshape it? Or Should I create feature vector of wav file? Can you help me with this?

Thanks

[ShivamSrivastavaAkaike]

Hi ,
I also want to use this model on android , if you have used it on android then could you help?

[zeynepgulhanuslu]

Hi,

I used it but did not have a chance to use it as streaming. It only runs as a batch. If you want, I can send the code but it's a little messy since I didn't use it and didn't get the time to finish it properly. If you give me your mail, I can send it to you as a reference.

[ShivamSrivastavaAkaike]

yeah sure ,it will help.
Thanks in advance
here is my email id :- [email protected]

[zeynepgulhanuslu]

Okay I will send it to you, If you want you can delete the comment. I hope it helps, have a nice day.

[mcig]

Hello all 👋, I have also been trying to make this model work on android devices with kotlin. And unfortunately, same like you @zeynepgulhanuslu i got stuck while running the model inference from a recorded batch and with real time data. I have tried to reimplement the real_time_processing_tf_lite.py in kotlin but it got way too complicated to perform array operations and fft calculations without numpy.

Have you open sourced or uploaded your version anywhere? Honestly, it would help me a lot at this point 😅.

[zeynepgulhanuslu]

Hello Mustafa I didn't complete the code either. But I created a tflite model for fft calculations. You can find more information with searching "create a TFLite model from one or more concrete functions in TensorFlow." . This way you can do numpy operations in a tflite model. I hope this help.