Update app.py

app.py

@@ -6,15 +6,15 @@ import re
 from pydub import AudioSegment
 from pydub.generators import Sine
 import io
-import torch
+# import torch
 
-device = "cuda:0" if torch.cuda.is_available() else "cpu"
-model_id = "openai/whisper-large-v3"
+# device = "cuda:0" if torch.cuda.is_available() else "cpu"
+model_id = "openai/whisper-whisper-large-v3"
 
 model = AutoModelForSpeechSeq2Seq.from_pretrained(
     model_id, low_cpu_mem_usage=True, use_safetensors=True
 )
-model.to(device)
+# model.to(device)
 
 processor = AutoProcessor.from_pretrained(model_id)
 
@@ -26,7 +26,7 @@ pipe = pipeline(
     max_new_tokens=128,
     chunk_length_s=30,
     batch_size=8,
-    device=device,
+    # device=device,
 )
 
 
@@ -55,7 +55,6 @@ def classifier(word_list_with_timestamp, language):
             if word.lower() in list_to_search:
                 foul_words.append(word)
                 negative_timestamps.append(item['timestamp'])
-                break
     else:
         list_to_search = list(arabic_bad_Words["words"])
         for item in word_list_with_timestamp:
@@ -63,8 +62,7 @@ def classifier(word_list_with_timestamp, language):
             for word_in_list in list_to_search:
                 if word_in_list == word:
                     foul_words.append(word)
-                    negative_timestamps.append(item['timestamp'])
-                    break   
+                    negative_timestamps.append(item['timestamp'])   
         
     return [foul_words, negative_timestamps]
 
@@ -76,8 +74,8 @@ def generate_bleep(duration_ms, frequency=1000):
 def mute_audio_range(audio_filepath, ranges, bleep_frequency=800):
     audio = AudioSegment.from_file(audio_filepath)
     for range in ranges:
-        start_time = range[0] - 0.1
-        end_time = range[-1] + 0.1
+        start_time = range[0]
+        end_time = range[-1]
         start_ms = start_time * 1000  # pydub works with milliseconds
         end_ms = end_time * 1000
         duration_ms = end_ms - start_ms
@@ -90,6 +88,9 @@ def mute_audio_range(audio_filepath, ranges, bleep_frequency=800):
 
     return audio
 
+def resample_audio(audio_segment, target_sample_rate=16000):
+    return audio_segment.set_frame_rate(target_sample_rate).set_channels(1).set_sample_width(2)
+
 def format_output_to_list(data):
     formatted_list = "\n".join([f"{item['timestamp'][0]}s - {item['timestamp'][1]}s \t : {item['text']}" for item in data])
     return formatted_list
@@ -113,6 +114,10 @@ def transcribe(input_audio, audio_language, task, timestamp_type):
 
 
     audio_output = mute_audio_range(input_audio, negative_timestamps)
+
+    # Resample the output audio to 16kHz
+    audio_output = resample_audio(audio_output, 16000)
+
     # Save the output audio to a BytesIO object
     output_buffer = io.BytesIO()
     audio_output.export(output_buffer, format="wav")
@@ -122,16 +127,26 @@ def transcribe(input_audio, audio_language, task, timestamp_type):
     sample_rate = audio_output.frame_rate
     audio_data = np.frombuffer(output_buffer.read(), dtype=np.int16)
 
-
     return  [text, timestamps, foul_words, (sample_rate, audio_data)]
 
 examples = [
-        ["arabic_english_audios/audios/arabic_audio_1.mp3", 'Arabic', 'transcribe', 'word'],
-        ["arabic_english_audios/audios/arabic_audio_2.flac", 'Arabic', 'transcribe', 'word'],
-        ["arabic_english_audios/audios/arabic_audio_3.flac", 'Arabic', 'transcribe', 'word'],
-        ["arabic_english_audios/audios/arabic_audio_31.mp3", 'Arabic', 'transcribe', 'word'],
-        ["arabic_english_audios/audios/arabic_audio_32.mp3", 'Arabic', 'transcribe', 'word'],
+        ["arabic_english_audios/audios/arabic_audio_11.mp3", 'Arabic', 'transcribe', 'word'],
+        ["arabic_english_audios/audios/arabic_audio_12.mp3", 'Arabic', 'transcribe', 'word'],
+        ["arabic_english_audios/audios/arabic_audio_13.mp3", 'Arabic', 'transcribe', 'word'],
     
+        ["arabic_english_audios/audios/english_audio_18.mp3", 'English', 'transcribe', 'word'],
+        ["arabic_english_audios/audios/english_audio_19.mp3", 'English', 'transcribe', 'word'],
+        ["arabic_english_audios/audios/english_audio_20.mp3", 'English', 'transcribe', 'word'],
+        ["arabic_english_audios/audios/english_audio_21.mp3", 'English', 'transcribe', 'word'],
+        ["arabic_english_audios/audios/english_audio_22.mp3", 'English', 'transcribe', 'word'],
+        ["arabic_english_audios/audios/english_audio_23.mp3", 'English', 'transcribe', 'word'],
+        ["arabic_english_audios/audios/english_audio_24.mp3", 'English', 'transcribe', 'word'],
+        ["arabic_english_audios/audios/english_audio_25.mp3", 'English', 'transcribe', 'word'],
+        ["arabic_english_audios/audios/english_audio_26.mp3", 'English', 'transcribe', 'word'],
+        ["arabic_english_audios/audios/english_audio_27.mp3", 'English', 'transcribe', 'word'],
+        ["arabic_english_audios/audios/english_audio_28.mp3", 'English', 'transcribe', 'word'],
+        ["arabic_english_audios/audios/english_audio_29.mp3", 'English', 'transcribe', 'word'],
+        ["arabic_english_audios/audios/english_audio_30.mp3", 'English', 'transcribe', 'word'],
         ["arabic_english_audios/audios/english_audio_31.mp3", 'English', 'transcribe', 'word'],
         ["arabic_english_audios/audios/english_audio_32.mp3", 'English', 'transcribe', 'word'],
         ["arabic_english_audios/audios/english_audio_33.mp3", 'English', 'transcribe', 'word'],
@@ -147,9 +162,6 @@ examples = [
         ["arabic_english_audios/audios/english_audio_43.mp3", 'English', 'transcribe', 'word'],
         ["arabic_english_audios/audios/english_audio_44.mp3", 'English', 'transcribe', 'word'],
         ["arabic_english_audios/audios/english_audio_45.mp3", 'English', 'transcribe', 'word'],
-        ["arabic_english_audios/audios/english_audio_46.mp3", 'English', 'transcribe', 'word'],
-        ["arabic_english_audios/audios/english_audio_48.mp3", 'English', 'transcribe', 'word'],
-        ["arabic_english_audios/audios/english_audio_49.mp3", 'English', 'transcribe', 'word'],
     ]
 
 with gr.Blocks(theme=gr.themes.Default()) as demo:
@@ -164,14 +176,14 @@ with gr.Blocks(theme=gr.themes.Default()) as demo:
             with gr.Row():
                 clear_button = gr.ClearButton(value="Clear")
                 submit_button = gr.Button("Submit", variant="primary", )
-            
+
         with gr.Column():
             transcript_output = gr.Text(label="Transcript")
             timestamp_output = gr.Text(label="Timestamps")
             foul_words = gr.Text(label="Foul Words")
             output_audio = gr.Audio(label="Output Audio", type="numpy")
     
-    examples = gr.Examples(examples, inputs=[audio_input, audio_language, task, timestamp_type], outputs=[transcript_output, timestamp_output,  foul_words, output_audio], fn=transcribe, examples_per_page=20,  cache_examples=False)
+    examples = gr.Examples(examples, inputs=[audio_input, audio_language, task, timestamp_type], outputs=[transcript_output, timestamp_output,  foul_words, output_audio], fn=transcribe, examples_per_page=50,  cache_examples=False)
 
     submit_button.click(fn=transcribe, inputs=[audio_input, audio_language, task, timestamp_type], outputs=[transcript_output, timestamp_output, foul_words, output_audio])
     clear_button.add([audio_input, audio_language, task, timestamp_type, transcript_output, timestamp_output, foul_words, output_audio])