import os
import logging
logging.getLogger('matplotlib').setLevel(logging.WARNING)
logging.basicConfig(level=logging.DEBUG,
                    format='%(asctime)s %(levelname)s %(message)s')

import torch
import numpy as np
import random
import librosa

from cosyvoice.utils.file_utils import load_wav

from cosyvoice.cli.cosyvoice import CosyVoice

cosyvoice= CosyVoice('FunAudioLLM/CosyVoice-300M')
cosyvoice_sft= CosyVoice('FunAudioLLM/CosyVoice-300M-SFT')
cosyvoice_instruct= CosyVoice('FunAudioLLM/CosyVoice-300M-Instruct')

example_tts_text = [
    "Every step we take is part of our strategy; everything you see, including the conversation I am having with you at this moment, every action I take, every word I speak, has a profound meaning.",
    "That comedian is really talented; as soon as he opens his mouth, he makes the whole audience burst into laughter.",
    "The prank he played made everyone unable to help but laugh."
]

example_prompt_text = ["我是通义实验室语音团队全新推出的生成式语音大模型，提供舒适自然的语音合成能力。",
                        "I am a newly launched generative speech large model by the Qwen Voice Team of the Tongyi Laboratory, offering comfortable and natural text-to-speech synthesis capabilities."]

prompt_sr, target_sr = 16000, 22050
default_data = np.zeros(target_sr)
def set_all_random_seed(seed):
    random.seed(seed)
    np.random.seed(seed)
    torch.manual_seed(seed)
    torch.cuda.manual_seed_all(seed)

max_val = 0.8
def postprocess(speech, top_db=60, hop_length=220, win_length=440):
    speech, _ = librosa.effects.trim(
        speech, top_db=top_db,
        frame_length=win_length,
        hop_length=hop_length
    )
    if speech.abs().max() > max_val:
        speech = speech / speech.abs().max() * max_val
    speech = torch.concat([speech, torch.zeros(1, int(target_sr * 0.2))], dim=1)
    return speech

def use_instruct(text):
    for symbol in ['<endofprompt>', '<laughter>', '</laughter>', '<strong>', '</strong>', '[laughter]', '[breath]']:
        if symbol in text:
            return True
    return False