Update app.py

app.py

@@ -17,7 +17,7 @@ import spaces
 import gradio as gr
 from PIL import Image
 #from pydub import AudioSegment
-from scipy.io import wavfile
+#from scipy.io import wavfile
 
 import os
 import re
@@ -26,6 +26,7 @@ import warnings
 #import datetime
 import subprocess
 from pathlib import Path
+import tempfile
 from fpdf import FPDF
 
 import psutil
@@ -69,23 +70,26 @@ 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(filepath, batch_size=16, sample_rate=16000): 
-     
-    waveform, sample_rate = torchaudio.load(filepath) 
+def transcribe_audio(audio_file, batch_size=16, sample_rate=16000):
+
+    with tempfile.TemporaryDirectory() as tmpdirname:
+        temp_path = Path(tmpdirname) / "audio_file"
+        
+        with open(temp_path, "wb") as f:
+            f.write(audio_file.read()) 
+        
+    waveform, sample_rate = torchaudio.load(str(temp_path))
+        
+    samples = waveform.numpy()
 
-    # --convert to mono
-    if waveform.ndim > 1:
-        waveform = waveform[0, :]
-
-    # --waveform to ndnumpy array
-    waveform = waveform.numpy()
     
-    start_time = time.time()
-
+    if samples.ndim > 1:
+        samples = samples[0, :]
+        
 
     # --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()