#imports
import os 
import pandas as pd

from asrassessment.utils.timit_load import TIMIT_file 

#Load TIMIT Files
timit_dir = f"{os.getcwd()}/{TIMIT_PACKAGE_NAME}" #note this is the folder containing the 'test' & 'train' folders. Usually they are named 'TIMIT'/'timit'
TIMIT_dict = TIMIT_file(timit_dir)

#Take sample phoneme string from TIMIT file 
phn_file_dir = TIMIT_dict['train']['dr1']['fecd0']['phn'][0]

#Load ASR Model
...

#Calculate Phoneneme Error Rate btw. 2 strings

from asrassessment.utils.data_input import convert_wav
from asrassessment.utils.generalfunc import *
from asrassessment.utils.standardizer import *

#file directory
wav_file_dir = TIMIT_dict['train']['dr1']['fecd0']['wav'][0]

#convert and overwrite wav file so it is useable 
convert_wav(file_dir,overwrite=True)

#test model 
asr_phn = allosaurus_model(file_dir)

#standardize phoneme string
asr_phn_conv = IPA_to_TIMIT(asr_phn)

#load TIMIT phn
timit_phn = read_phn(phn_file_dir,string=True)

#standardize phoneme string 
timit_phn_conv = TIMIT_to_IPA(timit_phn)

from asrassessment.utils.phone_error_rate import error_rate

output, error_df = error_rate(timit_phn_conv,asr_phn_conv)

#Final dataframe showing phoneme comparison & type of error
print(output)
print(error_df)

from asrassessment import main as asrtest

#plot 
asrtest.full_phn_boxplot(asr_model=allosaurus_model,file_set="TRAIN", DR=[0,1])

#Load ASR model
...

#Load Speech to Text Model 
...

#import 
from asrassessment import main as asrtest

#plot
asrtest.full_noise_stackedplot(audio_dict=TIMIT_dict['train'],
                               noise_wav="audiocheck.net_whitenoisegaussian.wav",
                               cfg_filedir= 'noisyspeech.cfg',
                               asr_phn_model=allosaurus_model,
                               asr_txt_model=speech_recog,
                               DR = [0,1],
                               SPK = [0,1],
                               louder_volumes=[],
                               softer_volumes=[0,5,10,15,20,25,30])

#import 
from asrassessment import main as asrtest

#file directory
phn_file_dir = TIMIT_dict['train']['dr1']['fecd0']['phn'][0]
wav_file_dir = TIMIT_dict['train']['dr1']['fecd0']['wav'][0]


asrtest.phoneme_wavchart(timit_phndir = phn_file_dir, 
                         timit_wavdir = wav_file_dir,
                         asr_model=allosaurus_model,
                         vlinecolor='grey',
                         print_df=False)

%pip install allosaurus
from allosaurus.app import read_recognizer

def allosaurus_model(file_directory,fr=16000,dataframe=False):
  
    model = read_recognizer()
    str_output = model.recognize(file_directory,lang_id='eng',timestamp=True)
    lst_output = str_output.split("\n")
    df = pd.DataFrame(lst_output, columns=['header'])
    df = df.header.str.split(pat=' ',expand=True)
    df.columns = ['start','timing','phoneme']
    
    #edit dataframe (add 'timing' to 'start' to get 'end' time/change start & end to milliseconds)
    df['start'] = df['start'].astype(float)
    df['start'] = df['start'].values*fr
    
    df['timing'] = df['timing'].astype(float)
    df['timing'] = df['timing'].values*fr
    
    df['end'] = df.apply(lambda row: row.start + row.timing, axis = 1)
    finaldf = df[['start','end', 'phoneme']]
    
    if dataframe == True:
        return finaldf
    else:
        allosaurus_phn = col_to_string(finaldf,colname='phoneme')
    return allosaurus_phn

import speech_recognition as sr

def speech_recog(timit_wav):
  r = sr.Recognizer()
  with sr.AudioFile(timit_wav) as source:
    audio = r.record(source) 
  
  return r.recognize_google(audio)