conver.py

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

@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:
        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:
        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]:
        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:
        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:
        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:
        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 add_background_music_and_tags(
        self,
        speech_audio: AudioSegment,
        music_path: str,
        tags_paths: List[str]
    ) -> AudioSegment:
        music = AudioSegment.from_file(music_path).fade_out(2000) - 25
        if len(music) < len(speech_audio):
            music = music * ((len(speech_audio) // len(music)) + 1)
        music = music[:len(speech_audio)]
        mixed = speech_audio.overlay(music)
        
        tag_intro = AudioSegment.from_file(tags_paths[0]) - 10
        tag_trans = AudioSegment.from_file(tags_paths[1]) - 10
        final_audio = tag_intro + mixed
        
        silent_ranges = []
        for i in range(0, len(speech_audio) - 500, 100):
            chunk = speech_audio[i:i+500]
            if chunk.dBFS < -40:
                silent_ranges.append((i, i + 500))
        
        for start, end in reversed(silent_ranges):
            if (end - start) >= len(tag_trans):
                final_audio = final_audio.overlay(tag_trans, position=start + 50)
        
        return final_audio

    async def url_to_audio(self, url: str, voice_1: str, voice_2: str) -> Tuple[str, str]:
        text = self.fetch_text(url)
        if len(words := text.split()) > self.config.max_words:
            text = " ".join(words[:self.config.max_words])
        conversation = self.extract_conversation(text)
        return await self._process_to_audio(conversation, voice_1, voice_2)

    async def text_to_audio(self, text: str, voice_1: str, voice_2: str) -> Tuple[str, str]:
        conversation = self.extract_conversation(text)
        return await self._process_to_audio(conversation, voice_1, voice_2)

    async def raw_text_to_audio(self, text: str, voice_1: str, voice_2: str) -> Tuple[str, str]:
        conversation = {"conversation": [{"speaker": "Narrator", "text": text}]}
        return await self._process_to_audio(conversation, voice_1, voice_2)

    async def _process_to_audio(
        self,
        conversation: Dict,
        voice_1: str,
        voice_2: str
    ) -> Tuple[str, str]:
        audio_files, folder_name = await self.text_to_speech(conversation, voice_1, voice_2)
        combined = self.combine_audio_files(audio_files)
        final_audio = self.add_background_music_and_tags(
            combined,
            "musica.mp3",
            ["tag.mp3", "tag2.mp3"]
        )
        output_path = os.path.join(folder_name, "output.mp3")
        final_audio.export(output_path, format="mp3")
        
        for f in audio_files:
            os.remove(f)
            
        text_output = "\n".join(
            f"{turn['speaker']}: {turn['text']}" 
            for turn in conversation["conversation"]
        )
        return output_path, text_output