Update utils_hf.py

utils_hf.py

@@ -19,6 +19,7 @@ load_dotenv()
 hf_token = os.environ.get("HF_DIARIZATION_TOKEN")
 api_key = os.environ.get("OPENAI_API_KEY")
 
+client = Client(api_key=api_key)
 
 device = torch.device("cuda" if torch.cuda.is_available() else "cpu")
 st.info(f"Usando dispositivo: {device}")
@@ -33,7 +34,7 @@ if torch.cuda.is_available():
 
 sumamry_pipeline = pipeline(
     "text-generation",
-    model="gpt-3.5-turbo",
+    model="gpt2",
     device=device,
 )
 
@@ -149,6 +150,20 @@ def transcribe_audio_whisper_lib(audio_path):
         st.stop()
 
 
+def transcribe_audio_openai(audio_path):
+    try:
+        with open(audio_path, "rb") as audio_file:
+            transcript = client.audio.transcriptions.create(
+                model="whisper-1",
+                file=audio_file,
+                response_format="json"
+            )
+        return transcript.text
+    except Exception as e:
+        st.error(f"Error al transcribir el audio: {e}")
+        st.stop()
+
+
 def download_youtube_audio(url, progress_bar):
     # No incluyas la extensión '.wav' en el 'outtmpl'
     output_path = os.path.join('temp_audio')
@@ -227,3 +242,47 @@ class StreamlitProgressHook(ProgressHook):
             progress_message = f"{step_name:<20} ━ Progress data unavailable"
         self.progress_text.text(progress_message)
 
+# def diarize_in_segments(audio_path, segment_duration=300):
+#     try:
+#         audio_duration = get_audio_duration(audio_path)
+#         diarization_segments = []
+#         diarization_text = ""
+#         speakers = set()
+#
+#         # Procesar cada segmento por separado
+#         for start_time in range(0, int(audio_duration), segment_duration):
+#             end_time = min(start_time + segment_duration, audio_duration)
+#             st.info(f"Procesando segmento desde {start_time} hasta {end_time} segundos...")
+#
+#             with StreamlitProgressHook() as hook:
+#                 diarization = pyannote_pipeline({
+#                     'audio': audio_path,
+#                     'start': start_time,
+#                     'end': end_time
+#                 }, hook=hook)
+#
+#             for segment, _, speaker in diarization.itertracks(yield_label=True):
+#                 diarization_segments.append((segment.start, segment.end, speaker))
+#                 diarization_text += f"{segment.start:.2f} - {segment.end:.2f}: {speaker}\n"
+#                 speakers.add(speaker)
+#
+#         return diarization_segments, diarization_text, speakers
+#     except Exception as e:
+#         st.error(f"Error al realizar la diarización de un segmento: {e}")
+#         return None, None, 0
+
+# def transcribe_audio_whisper_transformers(audio_path):
+#     try:
+#         # Leer el archivo de audio y convertirlo en un numpy ndarray
+#         audio_data, _ = sf.read(audio_path)
+#
+#         # Convertir a un solo canal si es necesario
+#         if len(audio_data.shape) > 1:
+#             audio_data = np.mean(audio_data, axis=1)
+#
+#         # Transcribir el audio usando el modelo Whisper
+#         transcript = whisper_pipeline(audio_data, return_timestamps=True)
+#         return transcript['text']
+#     except Exception as e:
+#         st.error(f"Error al transcribir el audio: {e}")
+#         st.stop()