Update app.py

app.py

@@ -16,7 +16,9 @@
 import spaces
 import gradio as gr
 from PIL import Image
-from pydub import AudioSegment
+#from pydub import AudioSegment
+from scipy.io import wavfile
+
 import os
 import re
 import time
@@ -32,7 +34,7 @@ from gpuinfo import GPUInfo
 #import csv
 import numpy as np
 import torch
-import torchaudio
+#import torchaudio
 import torchaudio.transforms as transforms
 
 from transformers import pipeline, AutoModel
@@ -67,22 +69,19 @@ def convert_to_wav(filepath):
 pipe = pipeline("automatic-speech-recognition", model="NbAiLab/nb-whisper-large", chunk_length_s=30, generate_kwargs={'task': 'transcribe', 'language': 'no'})
 
 @spaces.GPU()
-def transcribe_audio(audio_file, batch_size=16, sample_rate =16000):
-    if isinstance(audio_file, tuple):
-        audio_file = audio_file[0] # assumes first element contains the file path;
+def transcribe_audio(audio_file, batch_size=16): # sample_rate=16000
+    sample_rate, samples = wavfile.read(audio_file)
+    waveform, sample_rate = torchaudio.load(audio_file) # avoids TypeError here?
         
-    waveform, sample_rate = torchaudio.load(audio_file) # to avoid TypeError here
-
-    if waveform.ndim > 1:
-        waveform = waveform[0, :]
-
-    waveform = waveform.numpy()
+    # --convert to mono
+    if len(samples.shape) > 1:
+        samples = samples[:, 0]
 
     start_time = time.time()
 
     # --pipe it
     with torch.no_grad():
-        outputs = pipe(waveform, sampling_rate=sample_rate, batch_size=batch_size, return_timestamps=False)
+        outputs = pipe(samples, sampling_rate=sample_rate, batch_size=batch_size, return_timestamps=False)
 
     end_time = time.time()
     
@@ -110,7 +109,6 @@ def transcribe_audio(audio_file, batch_size=16, sample_rate =16000):
     return text.strip(), system_info
 
 
-
 #              ------------summary section------------