import gradio as gr
import numpy as np
import torch
from datasets import load_dataset

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)

languages = ["id"]

def translate(audio, target_language):
    outputs = asr_pipe(audio, max_new_tokens=256, generate_kwargs={"task": "translate", "language": target_language})
    return outputs["text"]

def synthesise(text, target_language):
    inputs = processor(text=text, return_tensors="pt")
    speech = model.generate_speech(inputs["input_ids"].to(device), speaker_embeddings.to(device), vocoder=vocoder)
    return speech.cpu()

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

title = "Multilingual Cascaded STST"
description = """
Demo for cascaded speech-to-speech translation (STST), mapping from source speech in any language to target speech in another language. Demo uses OpenAI's [Whisper Base](https://huggingface.co/openai/whisper-base) model for speech translation, and Microsoft's
[SpeechT5 TTS](https://huggingface.co/microsoft/speecht5_tts) model for text-to-speech:
![Cascaded STST](https://huggingface.co/datasets/huggingface-course/audio-course-images/resolve/main/s2st_cascaded.png "Diagram of cascaded speech to speech translation")
"""

demo = gr.Blocks()

def create_interface(source):
    return gr.Interface(
        fn=speech_to_speech_translation,
        inputs=[gr.Audio(source=source, type="filepath"), gr.Dropdown(choices=languages, label="Target Language")],
        outputs=gr.Audio(label="Generated Speech", type="numpy"),
        title=title,
        description=description,
    )

mic_translate = create_interface("microphone")
file_translate = create_interface("upload")

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

demo.launch()