Update app.py

app.py

@@ -7,31 +7,24 @@ import soundfile as sf
 import librosa
 
 pipe1 = pipeline("automatic-speech-recognition", model="openai/whisper-base")
-#pipe2 = pipeline("summarization", model="facebook/bart-large-cnn")
-#pipe3 = DiffusionPipeline.from_pretrained("stable-diffusion-v1-5/stable-diffusion-v1-5")
-#pipe3.to("cuda" if torch.cuda.is_available() else "cpu")
 
 def audio_to_text(audio):
     # Check if audio is a file path or a tuple of (sample_rate, numpy_array)
     if isinstance(audio, tuple):
         sample_rate, audio_data = audio  # Unpack sample rate and data
     else:
-        # If it's a file path, load the audio
         audio_data, sample_rate = librosa.load(audio, sr=16000)  # Load and resample to 16kHz directly
     
     # Convert to mono if the audio has more than one channel
     if len(audio_data.shape) > 1:
-        audio_data = np.mean(audio_data, axis=1)  # Averaging channels to convert to mono
+        audio_data = np.mean(audio_data, axis=1)
+    
     
-    # Resample the audio to 16 kHz if it's not already at 16 kHz
-    if sample_rate != 16000:
-        audio_data = librosa.resample(audio_data, orig_sr=sample_rate, target_sr=16000)
     
-    # Convert to numpy array with float32 data type
     audio_array = np.array(audio_data).astype(np.float32)
     
     # Transcribe the audio input
-    transcription = pipe1(audio_array)
+    transcription = pipe1([audio_array])
     transcription_text = transcription['text']
     
     # Print and return the transcription text