1

I want to resample a recording in 32000 KHz to 16000 KHz. I have done this with the code below. But the output audio is somewhat messed up.

You can find the original audio as well output in the following folder https://drive.google.com/drive/folders/1vr-ib8zvZagH_QeE4JSUtAUpp3EG75va

Any ideas what I am doing wrong?

import os
import librosa 
import soundfile as sf 

folder_name = "trial_sess"
os.chdir(os.path.join("process",folder_name))

for file in os.listdir():
    if file.endswith(".m4a") or file.endswith(".mp4") or file.endswith(".mp3"):
        nm,ext = file.split(".")
        sr = librosa.get_samplerate(file)
        y, sr = librosa.load(file, sr = sr)
        sf.write(os.path.join(zoom_loc,"sessions",folder_name,"output_resampled_audio" + "." + "wav"), data = y, samplerate=16000)
tripleee
  • 175,061
  • 34
  • 275
  • 318
imantha
  • 2,676
  • 4
  • 23
  • 46

2 Answers2

1

There is no resampling performed in your code, and the incorrect samplerate is passed to write. librosa.load will resample on-demand if the sr argument is different from that of the original file. So the code should be something like:

target_sr = 16000
y, sr = librosa.load(file, sr=target_sr)
assert sr == target_sr # check that librosa did resample
out_path = os.path.join(zoom_loc,"sessions",folder_name,"output_resampled_audio" + "." + "wav")
sf.write(out_path, data = y, samplerate=target_sr)
Long
  • 1,482
  • 21
  • 33
Jon Nordby
  • 5,494
  • 1
  • 21
  • 50
0

When you use librosa to load the file, make sure to add sr=None:

y, sr = librosa.load(file_path,sr=None)

This will tell librosa to not resample on load.

Long
  • 1,482
  • 21
  • 33