from transformers import pipeline
import gradio as gr
from pyctcdecode import BeamSearchDecoderCTC
import os
import torch
import torch.nn as nn
import torch.nn.functional as F
import torchaudio
from transformers import AutoConfig, AutoModel, Wav2Vec2FeatureExtractor
import librosa
import numpy as np
import subprocess


def resample(path, sampling_rate):
    speech_array, _sampling_rate = torchaudio.load(path)
    resampler = torchaudio.transforms.Resample(_sampling_rate)
    speech = resampler(speech_array).squeeze().numpy()
    return speech


def predict(path, sampling_rate=SR):
    speech = resample(path, sampling_rate)
    inputs = feature_extractor(speech, sampling_rate=sampling_rate, return_tensors="pt", padding=True)
    inputs = {key: inputs[key].to(device) for key in inputs}

    with torch.no_grad():
        logits = model.to(device)(**inputs).logits

    scores = F.softmax(logits, dim=1).detach().cpu().numpy()[0]
    outputs = {config.id2label[i]: f"{round(score * 100, 3):.1f}%" for i, score in enumerate(scores)}
    return outputs
 

TRUST = True
SR = 16000

config = AutoConfig.from_pretrained('Aniemore/wav2vec2-xlsr-53-russian-emotion-recognition', trust_remote_code=TRUST)
model = AutoModel.from_pretrained("Aniemore/wav2vec2-xlsr-53-russian-emotion-recognition", trust_remote_code=TRUST)
feature_extractor = Wav2Vec2FeatureExtractor.from_pretrained("Aniemore/wav2vec2-xlsr-53-russian-emotion-recognition")

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


def recognize(audio_path):
    return predict(audio_path)


with gr.Blocks() as blocks:
    audio = gr.Audio(source="microphone", type="filepath", label="Скажите что-нибудь...")
    success_button = gr.Button('Распознать эмоции')
    output = gr.JSON(label="Эмоции")
    
    success_button.click(fn=recognize, inputs=[audio], outputs=[output])
    
blocks.launch(enable_queue=True, debug=True)