import gdown 
import random

import numpy as np

from audio import read_mfcc
from batcher import sample_from_mfcc
from constants import SAMPLE_RATE, NUM_FRAMES
from conv_models import DeepSpeakerModel
from test import batch_cosine_similarity

class speaker_recognition:
    def __init__(self):

      np.random.seed(123)
      random.seed(123)

      self.speakers = {}
      self.weights = ""
      self.by_name = True

      self.SAMPLE_RATE = SAMPLE_RATE
      self.NUM_FRAMES = NUM_FRAMES

      self.spin_up()
    
    def spin_up(self):
      if self.weights == "":
        output = "weights.h5"
        gdown.download("https://drive.google.com/uc?id=1F9NvdrarWZNktdX9KlRYWWHDwRkip_aP", output, quiet=False)
        self.weights = "weights.h5"

      self.model = DeepSpeakerModel()
      self.model.m.load_weights(self.weights, by_name=True)

    def create_speaker(self, data, id=""):
      id = id if id != "" else f"{len(self.speakers)}"
      self.speakers[id] = data
      return id

    def check_speakers(self, data, id="", threshold = 0.5):
      us = ""
      n = 0
      for speaker in self.speakers:
        k = batch_cosine_similarity(self.speakers[speaker], data)
        if k > threshold:
          if k > n:
            n = k
            us = speaker
          else:pass
      if n == 0:
        id = self.create_speaker(data, id)
        return f"created new speaker : {id}"

      return (us, k[0])

    def run_transform(self, audio, pcm = False):
      data = sample_from_mfcc(read_mfcc(audio, self.SAMPLE_RATE), self.NUM_FRAMES)
      data = self.model.m.predict(np.expand_dims(data, axis=0))
      return data