Update utils_hf.py

utils_hf.py

@@ -1,23 +1,26 @@
 # import numpy as np
 # import soundfile as sf
 # import torchaudio
+# from transformers import pipeline
 
 import os
 import yt_dlp
 import streamlit as st
 import torch
 import whisper
+import requests
 from pyannote.audio import Pipeline
 from pyannote.audio.pipelines.utils.hook import ProgressHook
 from moviepy.editor import AudioFileClip
-from transformers import pipeline
 from openai import Client
 from dotenv import load_dotenv
 
 load_dotenv()
 
 hf_token = os.environ.get("HF_DIARIZATION_TOKEN")
-api_key = os.environ.get("OPENAI_API_KEY")
+oapi_key = os.environ.get("OPENAI_API_KEY")
+
+client = Client(api_key=oapi_key)
 
 device = torch.device("cuda" if torch.cuda.is_available() else "cpu")
 st.info(f"Usando dispositivo: {device}")
@@ -30,12 +33,6 @@ pyannote_pipeline = Pipeline.from_pretrained(
 if torch.cuda.is_available():
     pyannote_pipeline.to(torch.device("cuda"))
 
-sumamry_pipeline = pipeline(
-    "text-generation",
-    model="gpt2",
-    device=device,
-)
-
 whisper_model = whisper.load_model("turbo")
 
 
@@ -84,18 +81,33 @@ def process_transcripts(diarization_text, speakers):
 
 
 def summarize_speaker_transcripts(speaker_transcripts):
-    summaries = {}
+    url = "https://api.gemini.com/v1/generateContent"
+    api_key = os.environ.get("GEMINI_API_KEY")
+
     for speaker, transcripts in speaker_transcripts.items():
         full_text = ' '.join(transcripts)
         max_tokens = 1024  # Ajusta según el modelo y tus necesidades
         if len(full_text) > max_tokens:
             full_text = full_text[:max_tokens]
 
-        prompt = f"Por favor, resume el siguiente texto:\n\n{full_text}"
+        data = {
+            "prompt": f"Por favor, resume el siguiente texto:\n\n{full_text}",
+            "model": "gemini-1.5-pro",
+            "max_tokens": 150
+        }
 
-        summary = sumamry_pipeline(prompt, max_length=150, min_length=30, do_sample=False)[0]['generated_text']
-        summaries[speaker] = summary
-    return summaries
+        headers = {
+            "Authorization": f"Bearer {api_key}",
+            "Content-Type": "application/json"
+        }
+
+        response = requests.post(url, json=data, headers=headers)
+
+        if response.status_code == 200:
+            result = response.json()
+            return result["text"]
+        else:
+            st.error(f"{response.status_code}, {response.text}")
 
 
 def diarize_full_audio(audio_path):
@@ -142,8 +154,7 @@ def extract_audio_segment(input_audio_path, output_audio_path, start_time, end_t
 
 def transcribe_audio_whisper_lib(audio_path):
     try:
-        with open(audio_path, "rb") as audio_file:
-            transcript = whisper_model.transcribe(audio_path)
+        transcript = whisper_model.transcribe(audio_path)
         return transcript["text"]
     except Exception as e:
         st.error(f"Error al transcribir el audio: {e}")
@@ -241,3 +252,69 @@ class StreamlitProgressHook(ProgressHook):
         else:
             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()
+
+
+# sumamry_pipeline = pipeline(
+#     "text-generation",
+#     model="gpt2",
+#     device=device,
+# )
+
+# def summarize_speaker_transcripts_gpt2(speaker_transcripts):
+#     summaries = {}
+#     for speaker, transcripts in speaker_transcripts.items():
+#         full_text = ' '.join(transcripts)
+#         max_tokens = 1024  # Ajusta según el modelo y tus necesidades
+#         if len(full_text) > max_tokens:
+#             full_text = full_text[:max_tokens]
+#
+#         prompt = f"Por favor, resume el siguiente texto:\n\n{full_text}"
+#
+#         summary = sumamry_pipeline(prompt, max_length=150, min_length=30, do_sample=False)[0]['generated_text']
+#         summaries[speaker] = summary
+#     return summaries