Update app.py

app.py

@@ -85,7 +85,7 @@ def transcribe_audio(audio_file, chunk_length_s=30):
 
     # Resample the audio to 16000 Hz if it’s not already
     if sample_rate != 16000:
-        resampler = transforms.Resample(orig_freq=sample_rate, new_freq=16000)
+        resampler = torchaudio.transforms.Resample(orig_freq=sample_rate, new_freq=16000)
         waveform = resampler(waveform)
         sample_rate = 16000
 
@@ -105,11 +105,11 @@ def transcribe_audio(audio_file, chunk_length_s=30):
         if chunk_waveform.shape[0] > 1:
             chunk_waveform = torch.mean(chunk_waveform, dim=0, keepdim=True)
 
-        # Process the chunk
-        audio_input = processor(chunk_waveform.squeeze(0).numpy(), sampling_rate=sample_rate, return_tensors="pt")
+        # Process the chunk with the tokenizer
+        inputs = processor(chunk_waveform.squeeze(0).numpy(), sampling_rate=sample_rate, return_tensors="pt")
+        input_features = inputs.input_features
 
-        # Generate attention mask
-        input_features = audio_input.input_features
+        # Explicitly set the attention mask
         attention_mask = torch.ones(input_features.shape, dtype=torch.long)
 
         # ASR model inference on the chunk
@@ -136,7 +136,7 @@ def transcribe_audio(audio_file, chunk_length_s=30):
     # Format of the result
     result = (
         f"Time taken: {output_time:.2f} seconds\n"
-        f"Audio duration: {audio_duration / 60:.2f} minutes ({audio_duration:.2f} seconds)\n"
+        f"Audio duration: {audio_duration / 60:.2f minutes ({audio_duration:.2f} seconds)\n"
         f"Real-time Factor (RTF): {rtf:.2f}\n"
         f"Number of words: {len(text.split())}\n\n"
         "Real-time Factor (RTF) is a measure used to evaluate the speed of speech recognition systems. "
@@ -259,7 +259,7 @@ def save_to_pdf(text, summary):
 iface = gr.Blocks()
 
 with iface:
-    gr.HTML('<img src="https://huggingface.co/spaces/camparchimedes/transcription_app/resolve/main/banner_trans.png" alt="Banner Image" />')
+    gr.HTML('<img src="https://huggingface.co/spaces/camparchimedes/transcription_app/blob/main/lol.webp" alt="Banner Image" />')
     gr.Markdown("*Switch Work's JoJo-versjon som webapp for transkribering av lydfiler til norsk skrift. Språkmodell: NbAiLab/nb-whisper-large, Ekstra: oppsummering, pdf-download*")
 
     with gr.Tabs():