Add pydub to cut the audio file to 10 seconds

app.py

@@ -9,6 +9,7 @@ import os
 
 import numpy as np
 import tensorflow.compat.v2 as tf
+from pydub import AudioSegment
 
 import functools
 import gin
@@ -35,6 +36,8 @@ SAMPLE_RATE = 16000
 SF2_PATH = 'SGM-v2.01-Sal-Guit-Bass-V1.3.sf2'
 
 def upload_audio(audio, sample_rate):
+  audio = AudioSegment.from_wav(audio)
+  audio = [:10000]
   return note_seq.audio_io.wav_data_to_samples_librosa(
     audio, sample_rate=sample_rate)
 
@@ -242,8 +245,7 @@ def inference(audio):
   with open(audio, 'rb') as fd:
       contents = fd.read()
   audio = upload_audio(contents,sample_rate=16000)
-  # Get only first 10 seconds of audio with free Space hosting
-  audio = audio[:160000]
+
   est_ns = inference_model(audio)
   
   note_seq.sequence_proto_to_midi_file(est_ns, './transcribed.mid')

requirements.txt

@@ -7,4 +7,5 @@ jax[cpu]==0.3.15 -f https://storage.googleapis.com/jax-releases/jax_releases.htm
 # pin CLU for python 3.7 compatibility
 clu==0.0.7
 # pin Orbax to use Checkpointer
-orbax==0.0.2
+orbax==0.0.2
+pydub