import gradio as gr
import numpy as np
import torch
from datasets import load_dataset
import librosa  # librosa 라이브러리를 임포트합니다.

from transformers import SpeechT5ForTextToSpeech, SpeechT5HifiGan, SpeechT5Processor, pipeline

device = "cuda:0" if torch.cuda.is_available() else "cpu"

# Load speech translation checkpoint
asr_pipe = pipeline("automatic-speech-recognition", model="openai/whisper-base", device=device)

# Load text-to-speech checkpoint and speaker embeddings
processor = SpeechT5Processor.from_pretrained("microsoft/speecht5_tts")
model = SpeechT5ForTextToSpeech.from_pretrained("microsoft/speecht5_tts").to(device)
vocoder = SpeechT5HifiGan.from_pretrained("microsoft/speecht5_hifigan").to(device)

embeddings_dataset = load_dataset("Matthijs/cmu-arctic-xvectors", split="validation")
speaker_embeddings = torch.tensor(embeddings_dataset[7306]["xvector"]).unsqueeze(0)

def translate(audio):
    audio_length = librosa.get_duration(filename=audio)
    if audio_length > 30:
        # 오디오를 30초로 자르기 (이 부분은 구현에 따라 달라질 수 있습니다.)
        audio = audio[:int(30 * 16000)]  # 가정: 오디오 샘플링 레이트가 16000Hz
    outputs = asr_pipe(audio, max_new_tokens=256, generate_kwargs={"task": "translate", "return_timestamps": False})
    return outputs["text"]

def synthesise(text):
    inputs = processor(text=text, return_tensors="pt")
    max_length = 600  # 모델이 처리할 수 있는 최대 길이
    if inputs["input_ids"].size(1) > max_length:
        inputs["input_ids"] = inputs["input_ids"][:, :max_length]
    speech = model.generate_speech(inputs["input_ids"].to(device), speaker_embeddings.to(device), vocoder=vocoder)
    return speech.cpu()

def speech_to_speech_translation(audio):
    translated_text = translate(audio)
    synthesised_speech = synthesise(translated_text)
    synthesised_speech = (synthesised_speech.numpy() * 32767).astype(np.int16)
    return 16000, synthesised_speech

css = """
footer {
    visibility: hidden;
}
"""

demo = gr.Blocks(theme="Nymbo/Nymbo_Theme", css=css)

mic_translate = gr.Interface(
    fn=speech_to_speech_translation,
    inputs=gr.Audio(sources=["microphone"], type="filepath"),
    outputs=gr.Audio(label="Generated Speech", type="numpy"),
)

file_translate = gr.Interface(
    fn=speech_to_speech_translation,
    inputs=gr.Audio(sources=["upload"], type="filepath"),
    outputs=gr.Audio(label="Generated Speech", type="numpy"),
    examples=[["./example.wav"]],
)

with demo:
    gr.TabbedInterface([mic_translate, file_translate], ["Microphone", "Audio File"])

demo.launch()