k2kobayashi / shifter Goto Github PK

In the current ver., there is little information about the CLASS and function. It should be written to easily understand the perspective in the scripts.

Current implement, Shifter takes around 10s for a speech waveform. It should be shortened. In this issue, we should find where the code takes a long time and improve it.

I got a corruption when using function: f0trans.transform(....)

mycode is like below:

def changePitch(pitch_r, sample_data, sample_rate=16000):
    # F0 transoformation based on WSOLA and resampling
    f0trans = Shifter(sample_rate, pitch_r, frame_ms=20, shift_ms=10)
    transformed = f0trans.transform(sample_data)
    return transformed

--size of sample_data is 416600,pitch_r is 0.9

Error message:
Traceback (most recent call last):
File "audio_process.py", line 269, in
data = audio_process.changePitch(pitch_ratio, data, sample_rate)
File "audio_process.py", line 230, in changePitch
transformed = f0trans.transform(sample_data)
File "audio_process.py", line 59, in transform
wsolaed = self.duration_modification(data)
File "audio_process.py", line 112, in duration_modification
wsolaed[outp:outp + self.sl] = spdata + epdata
ValueError: could not broadcast input array from shape (160) into shape (60)

In order to use Shifter as a command line tool, it is necessary to implement an installation script for Linux and Mac environment. In this issue, we will modify the setup.py and requirements.txt to achieve it.