Description
This is a trivial Android application to apply screaming metal-like effects (even in real-time, headphones recommended), to voices recorded through the microphone. The app makes use of the ScreamNet neural network, converted into a tflite model.
Main limitations
- The model is quite worse than the original one due to the down-sampling algorithm of the recorded audio (to pass from 44100Hz to 22050Hz), that does not preserve fidelity with respect to the original signal. Currently, a simple linear interpolation is applied (with poor results). It has been proved that a "soxr_lq" strategy (as applied in the main Python scripts) would perfectly work; nonetheless, there is no such a library in Java to compute this operation.
- Real-time processing will inevitably occur with a 1-second delay (the network processes 22050 samples at a time), which can be distracting.
- The user will have to complete the listening of the results before being able to save them.