upload_styles.py

# https://github.com/audeering/shift/tree/main  - MAKE Mimic-3 voice / harvard 1x 4x
import shutil
import csv
import io
import os
import typing
import wave
import sys
from mimic3_tts.__main__ import (CommandLineInterfaceState,
                                 get_args,
                                 initialize_args,
                                 initialize_tts,
                                 # print_voices,
                                 # process_lines,
                                 shutdown_tts,
                                 OutputNaming,
                                 process_line)
import time
import json
import os
import numpy as np

from pathlib import Path
import audiofile


# ================================================ LIST OF VOICES
ROOT_DIR = '/data/dkounadis/mimic3-voices/'
foreign_voices = []
english_voices = []
for lang in os.listdir(ROOT_DIR + 'voices'):
        
        for voice in os.listdir(ROOT_DIR + 'voices/' + lang):
            if 'en_' in lang:

                try:
                    with open(ROOT_DIR + 'voices/' + lang + '/' + voice + '/speakers.txt', 'r') as f:
                        for spk in f:
                            english_voices.append(lang + '/' + voice + '#' + spk.rstrip())
                        # voice_spk_string = lang + '/' + voice + '#' + spk.rstrip() for spk in f
                except FileNotFoundError:
                    english_voices.append(lang + '/' + voice)

            else:
                
                try:
                    with open(ROOT_DIR + 'voices/' + lang + '/' + voice + '/speakers.txt', 'r') as f:
                        for spk in f:
                            foreign_voices.append(lang + '/' + voice + '#' + spk.rstrip())
                        
                except FileNotFoundError:
                    foreign_voices.append(lang + '/' + voice)
# 
[print(i) for i in foreign_voices]
print('\n_______________________________\n')
[print(i) for i in english_voices]
# ====================================================== LIST Mimic-3 ALL VOICES
# list_voices = [
#     'en_US/m-ailabs_low#mary_ann',
#     'en_UK/apope_low',
#     'de_DE/thorsten-emotion_low#neutral',  # is the 4x really interesting we can just write it in Section
#     # 'ko_KO/kss_low',
#     'fr_FR/m-ailabs_low#gilles_g_le_blanc',

#     #'human',
#     ]  # special - for human we load specific style file - no Mimic3 is run

# ==================================    ====== END INTERFACE





def process_lines(state: CommandLineInterfaceState, wav_path=None):
    '''MIMIC3 INTERNAL CALL that yields the sigh sound'''

    args = state.args

    result_idx = 0
    print(f'why waitings in the for loop LIN {state.texts=}\n')
    for line in state.texts:
        # print(f'LIN {line=}\n')  # prints \n so is empty not getting the predifne text of state.texts
        line_voice: typing.Optional[str] = None
        line_id = ""
        line = line.strip()
        # if not line:
        #     continue

        if args.output_naming == OutputNaming.ID:
            # Line has the format id|text instead of just text
            with io.StringIO(line) as line_io:
                reader = csv.reader(line_io, delimiter=args.csv_delimiter)
                row = next(reader)
                line_id, line = row[0], row[-1]
                if args.csv_voice:
                    line_voice = row[1]

        process_line(line, state,
                     line_id=line_id,
                     line_voice=line_voice)
        result_idx += 1
    time.sleep(4)
    # Write combined audio to stdout
    if state.all_audio:
        # _LOGGER.debug("Writing WAV audio to stdout")

        if sys.stdout.isatty() and (not state.args.stdout):
            with io.BytesIO() as wav_io:
                wav_file_play: wave.Wave_write = wave.open(wav_io, "wb")
                with wav_file_play:
                    wav_file_play.setframerate(state.sample_rate_hz)
                    wav_file_play.setsampwidth(state.sample_width_bytes)
                    wav_file_play.setnchannels(state.num_channels)
                    wav_file_play.writeframes(state.all_audio)

                    # play_wav_bytes(state.args, wav_io.getvalue())
                # wav_path = '_direct_call_2.wav'
                with open(wav_path, 'wb') as wav_file:
                    wav_file.write(wav_io.getvalue())
                    wav_file.seek(0)
                    print('\n\n5T', wav_path)
    else:
        print('\n\nDOES NOT TTSING --> ADD SOME time.sleep(4)', wav_path)                    

# -----------------------------------------------------------------------------
# cat _tmp_ssml.txt | mimic3 --cuda --ssml --noise-w 0.90001 --length-scale 0.91 --noise-scale 0.04 > noise_w=0.90_en_happy_2.wav
# ======================================================================




for lang, list_voices in [
    ['english', english_voices],
    ['foreign', foreign_voices]
                            ]:
    for rate in [1, 4]:
        
                
        
        
        # # --
        # # assure mimic-3 generator .onnx exists
        # home_voice_dir = f'/home/audeering.local/dkounadis/.local/share/mycroft/mimic3/voices/{_voice.split("#")[0]}/'
        # Path(home_voice_dir).mkdir(parents=True, exist_ok=True)
        # speaker_free_voice_name = _voice.split("#")[0] if '#' in _voice else _voice


        # if (
        #     (not os.path.isfile(home_voice_dir + 'generator.onnx')) or 
        #     (os.path.getsize(home_voice_dir + 'generator.onnx') < 500)  # .onnx - is just LFS header
        #         ):

        #     # Copy

        #     shutil.copyfile(
        #         f'/data/dkounadis/mimic3-voices/voices/{speaker_free_voice_name}/generator.onnx',
        #         home_voice_dir + 'generator.onnx')
        # # --
        

        
        
        with open('harvard.json', 'r') as f:
            harvard_individual_sentences = json.load(f)['sentences']
        total_audio_mimic3 = []
        
        ix = 0
        for list_of_10 in harvard_individual_sentences[:4]:  # 77
            # text = ' '.join(list_of_10['sentences'])
            for text in list_of_10['sentences']:
            
                
                _voice = list_voices[ix % len(list_voices)]
                _str = _voice.replace('/', '_').replace('#', '_').replace('_low', '')

                if 'cmu-arctic' in _str:
                    _str = _str.replace('cmu-arctic', 'cmu_arctic') #+ '.wav'

                print(ix, lang, text)
                


                # Synthesis Mimic-3 then use it as prompt for StyleTTS2

                # MIMIC-3 if _voice is not HUMAN
                
                _ssml = (
                    '<speak>'
                    '<prosody volume=\'64\'>'
                    f'<prosody rate=\'{rate}\'>'
                    f'<voice name=\'{_voice}\'>'
                    '<s>'
                    f'{text[:-1] + ", .. !!!"}'
                    '</s>'
                    '</voice>'
                    '</prosody>'
                    '</prosody>'
                    '</speak>'
                )
                with open('_tmp_ssml.txt', 'w') as f:
                    f.write(_ssml)


                # ps = subprocess.Popen(f'cat _tmp_ssml.txt | mimic3 --ssml > {reference_wav}', shell=True)
                # ps.wait()  # using ps to call mimic3 because samples dont have time to be written in stdout buffer
                args = get_args()
                args.ssml = True
                args.text = [_ssml]  #['aa', 'bb'] #txt
                args.interactive = False
                # args.output_naming = OutputNaming.TIME

                state = CommandLineInterfaceState(args=args)
                initialize_args(state)
                initialize_tts(state)
                # args.texts = [txt] #['aa', 'bb'] #txt
                # state.stdout = '.' #None #'makeme.wav'
                # state.output_dir = '.noopy'
                # state.interactive = False
                # state.output_naming = OutputNaming.TIME
                # # state.ssml = 1234546575
                # state.stdout = True
                # state.tts = True
                style_path = 'tmp1.wav'
                process_lines(state, wav_path=style_path)
                shutdown_tts(state)
                x, fs = audiofile.read(style_path)
                ix += 1
                total_audio_mimic3.append(x)
            

        # save styletts2 .wav

        

        total_audio_mimic3 = np.concatenate(total_audio_mimic3) # -- concat 77x lists
        audiofile.write(f'harvards_upload_mimic3_{rate}_{lang}.wav', total_audio_mimic3, 22050)

        print(total_audio_mimic3.shape, 'LEN\n')