from dataclasses import dataclass from typing import List, Tuple, Dict, Optional import os import json import httpx from openai import OpenAI import edge_tts import tempfile from pydub import AudioSegment import base64 from pathlib import Path import numpy as np @dataclass class ConversationConfig: max_words: int = 3000 prefix_url: str = "https://r.jina.ai/" model_name: str = "meta-llama/Meta-Llama-3-8B-Instruct" class URLToAudioConverter: def __init__(self, config: ConversationConfig, llm_api_key: str): self.config = config self.llm_client = OpenAI(api_key=llm_api_key, base_url="https://api.together.xyz/v1") self.llm_out = None def fetch_text(self, url: str) -> str: """Obtiene texto desde una URL""" if not url: raise ValueError("URL cannot be empty") full_url = f"{self.config.prefix_url}{url}" try: response = httpx.get(full_url, timeout=60.0) response.raise_for_status() return response.text except httpx.HTTPError as e: raise RuntimeError(f"Failed to fetch URL: {e}") def extract_conversation(self, text: str) -> Dict: """Convierte texto plano a estructura de conversación""" if not text: raise ValueError("Input text cannot be empty") try: prompt = ( f"{text}\nConvert this text into a podcast conversation between two hosts. " "Return ONLY JSON with this structure:\n" '{"conversation": [{"speaker": "Host1", "text": "..."}, {"speaker": "Host2", "text": "..."}]}' ) response = self.llm_client.chat.completions.create( messages=[{"role": "user", "content": prompt}], model=self.config.model_name, response_format={"type": "json_object"} ) json_str = response.choices[0].message.content.strip() return json.loads(json_str[json_str.find('{'):json_str.rfind('}')+1]) except Exception as e: raise RuntimeError(f"Failed to extract conversation: {str(e)}") async def text_to_speech(self, conversation_json: Dict, voice_1: str, voice_2: str) -> Tuple[List[str], str]: """Convierte JSON de conversación a archivos de audio""" output_dir = Path(self._create_output_directory()) filenames = [] try: for i, turn in enumerate(conversation_json["conversation"]): filename = output_dir / f"segment_{i}.mp3" voice = voice_1 if turn["speaker"] == "Host1" else voice_2 tmp_path = await self._generate_audio(turn["text"], voice) os.rename(tmp_path, filename) filenames.append(str(filename)) return filenames, str(output_dir) except Exception as e: raise RuntimeError(f"Text-to-speech failed: {e}") async def _generate_audio(self, text: str, voice: str) -> str: """Genera audio temporal con edge-tts""" if not text.strip(): raise ValueError("Text cannot be empty") communicate = edge_tts.Communicate( text, voice.split(" - ")[0], rate="+0%", pitch="+0Hz" ) with tempfile.NamedTemporaryFile(delete=False, suffix=".mp3") as tmp_file: await communicate.save(tmp_file.name) return tmp_file.name def _create_output_directory(self) -> str: """Crea directorio único para los archivos""" folder_name = base64.urlsafe_b64encode(os.urandom(8)).decode("utf-8") os.makedirs(folder_name, exist_ok=True) return folder_name def combine_audio_files(self, filenames: List[str]) -> AudioSegment: """Combina segmentos de audio""" if not filenames: raise ValueError("No audio files provided") combined = AudioSegment.empty() for filename in filenames: combined += AudioSegment.from_file(filename, format="mp3") return combined def _detect_silences(self, audio: AudioSegment, min_len: int = 500, thresh: int = -40) -> List[Tuple[int, int]]: """Detecta intervalos de silencio en el audio""" silent_ranges = [] start = None samples = np.array(audio.get_array_of_samples()) window_size = int(min_len * audio.frame_rate / 1000) for i in range(0, len(samples) - window_size, window_size): window = samples[i:i+window_size] if np.max(window) < thresh: if start is None: start = i else: if start is not None: silent_ranges.append((start, i)) start = None return silent_ranges def add_background_music_and_tags( self, speech_audio: AudioSegment, music_path: str, tags_paths: List[str] ) -> AudioSegment: """Mezcla música de fondo y tags inteligentemente""" # 1. Cargar y ajustar música music = AudioSegment.from_file(music_path).fade_out(2000) music = music - 25 # Reducir volumen # 2. Loop inteligente (solo si es necesario) if len(music) < len(speech_audio): loops = (len(speech_audio) // len(music)) + 1 music = music * loops music = music[:len(speech_audio)] # 3. Mezclar voz y música mixed = speech_audio.overlay(music, position=0) # 4. Insertar tags tag_intro = AudioSegment.from_file(tags_paths[0]) - 10 tag_transition = AudioSegment.from_file(tags_paths[1]) - 10 # Tag inicial final_audio = tag_intro + mixed # Tags en pausas (opcional) silences = self._detect_silences(speech_audio) for start, end in reversed(silences): if (end - start) > len(tag_transition): final_audio = final_audio.overlay( tag_transition, position=start + 100 # Pequeño margen ) return final_audio async def process_content( self, content: str, voice_1: str, voice_2: str, is_url: bool = False ) -> Tuple[str, str]: """Procesa contenido (URL o texto) a audio final""" try: # 1. Obtener texto estructurado if is_url: text = self.fetch_text(content) if len(words := text.split()) > self.config.max_words: text = " ".join(words[:self.config.max_words]) conversation = self.extract_conversation(text) else: conversation = self.extract_conversation(content) # 2. Generar audio audio_files, folder_name = await self.text_to_speech(conversation, voice_1, voice_2) combined = self.combine_audio_files(audio_files) # 3. Mezclar con música y tags final_audio = self.add_background_music_and_tags( combined, "musica.mp3", ["tag.mp3", "tag2.mp3"] ) # 4. Exportar output_path = os.path.join(folder_name, "podcast_final.mp3") final_audio.export(output_path, format="mp3") # 5. Limpieza for f in audio_files: os.remove(f) # Texto de conversación conversation_text = "\n".join( f"{turn['speaker']}: {turn['text']}" for turn in conversation["conversation"] ) return output_path, conversation_text except Exception as e: raise RuntimeError(f"Processing failed: {str(e)}")