Update app.py

app.py

@@ -54,102 +54,49 @@ CACHE_EXAMPLES = torch.device('cuda') and os.getenv("CACHE_EXAMPLES", "0") == "1
 device = torch.device('cuda')
 #device = torch.device("cuda:0" if torch.cuda.is_available() else "cpu")
 
-file = file_upload
-
-def chunk_audio(file, chunk_length_ms=30000, overlap_length_ms=5000):
-    # -- pydub
-    audio = AudioSegment.from_file(file)
-    
-    # -- create chunks with overlap
-    chunks = []
-    for i in range(0, len(audio), chunk_length_ms - overlap_length_ms):
-        start = max(0, i)
-        end = min(len(audio), i + chunk_length_ms)
-        chunks.append(audio[start:end])
-    
-    return chunks
-
-def transcribe(file_upload, progress=gr.Progress(track_tqdm=True)): 
-    start_time = time.time()
-
-    # Load the speech recognition model
-    with torch.no_grad():
-        pipe = pipeline("automatic-speech-recognition", model="NbAiLab/nb-whisper-large", device=device)
-    
-    # -- chunking
-    chunks = chunk_audio(file, chunk_length_ms=30000, overlap_length_ms=5000)
-    
-    full_transcription = []
-    for chunk in chunks:
-        # -- convert to temporary file-like object
-        temp_audio = chunk.export(format="wav")
-        
-        # -- transcribe chunk
-        text = pipe(temp_audio)["text"]
-        full_transcription.append(text)
-    
-    # -- join
-    full_text = " ".join(full_transcription)
-
-    # -- timimg, word count
-    end_time = time.time()
-    output_time = end_time - start_time
-    word_count = len(full_text.split())
-
-    # -- metrics
-    memory = psutil.virtual_memory()
-    cpu_usage = psutil.cpu_percent(interval=1)
-
-    # --system info string
-    system_info = f"""
-    Processing time: {output_time:.2f} seconds.
-    Number of words: {word_count}
-    """
-
-    return full_text, system_info
-    
 #@spaces.GPU
-#def transcribe(file_upload, progress=gr.Progress(track_tqdm=True)): # microphone
+def transcribe(file_upload, progress=gr.Progress(track_tqdm=True)): # microphone
     
-    #file = file_upload # microphone if microphone is not None else 
-    #start_time = time.time()
+    file = file_upload # microphone if microphone is not None else 
+    start_time = time.time()
     
     #--------------____________________________________________--------------"
 
-    #with torch.no_grad():
-        #pipe = pipeline("automatic-speech-recognition", model="NbAiLab/nb-whisper-large", device=device)
+    with torch.no_grad():
+        pipe = pipeline("automatic-speech-recognition", model="NbAiLab/nb-whisper-large", device=device)
 
-    #text = pipe(file)["text"] 
+    text = pipe(file)["text"] 
 
     #--------------____________________________________________--------------"
     
-    #end_time = time.time()
-    #output_time = end_time - start_time
+    end_time = time.time()
+    output_time = end_time - start_time
     
     # --Word count
-    #word_count = len(text.split())
+    word_count = len(text.split())
 
     # --Memory metrics
-    #memory = psutil.virtual_memory()
+    memory = psutil.virtual_memory()
     
     # --CPU metric
-    #cpu_usage = psutil.cpu_percent(interval=1)
+    cpu_usage = psutil.cpu_percent(interval=1)
 
     # --GPU metric
-    #gpu_utilization, gpu_memory = GPUInfo.gpu_usage()
+    gpu_utilization, gpu_memory = GPUInfo.gpu_usage()
     
     # --system info string
-    #system_info = f"""
-    #Processing time: {output_time:.2f} seconds.
-    #Number of words: {word_count}
-    #GPU Memory: {gpu_memory}
+    system_info = f"""
+    Processing time: {output_time:.2f} seconds.
+    Number of words: {word_count}
+    GPU Memory: {gpu_memory}"""
  
     #--------------____________________________________________--------------"
+    
     #CPU Usage: {cpu_usage}%
-    #Memory used: {memory.percent}%
-    #GPU Utilization: {gpu_utilization}%
-
-    #return text, system_info
+    Memory used: {memory.percent}%
+    GPU Utilization: {gpu_utilization}%
+    
+    return text, system_info
 
 
 ###############################################################################