import os
import numpy as np
import soundfile
import msinference
foreign_accents_en = os.listdir('assets/wavs/mimic3_foreign/') # build by mimic3 USE assets/wavs/style_vector/ for native
foreign_accents_en = sorted(foreign_accents_en)
def tts_entry(text='A quick brown fox jumps over the lazy dog. Sweet dreams are made of this, I traveled the world and the seven seas.',
voice='af_ZA/google-nwu_1919', # 'serbian', # 'en_US/vctk_low#p276', 'isl', 'abi',
speed=1.4, # only for non-english
affect = True # False = high clarity for partially sight
):
'''returns 24kHZ np.array TTS
voice : 'en_US/vctk_low#p276' # from English voices -> https://audeering.github.io/shift/
or
voice : FOREIGN ACCENTS
or
voice : 'deu' # from LHS code -> https://huggingface.co/dkounadis/artificial-styletts2/blob/main/Utils/all_langs.csv
'''
# StyleTTS2 - English
# mimic-3 format of voice (English txt - English accent)
if ('en_US/' in voice) or ('en_UK/' in voice):
a = '' if affect else 'v2/'
style_vector = msinference.compute_style('assets/wavs/style_vector/' + a + voice.replace(
'/', '_').replace('#', '_').replace(
'cmu-arctic', 'cmu_arctic').replace(
'_low', '') + '.wav')
x = msinference.inference(text,
style_vector,
alpha=0.3,
beta=0.7,
diffusion_steps=7,
embedding_scale=1)
# mimic-3 format of voice (English text - Foreign accent)
elif '_' in voice: # MSS-TTS dont have '_' HOWEVER all assets/wavs/mimic3_foreign_4x have '_'
style_vector = msinference.compute_style('assets/wavs/mimic3_foreign_4x/' + voice.replace(
'/', '_').replace('#', '_').replace(
'cmu-arctic', 'cmu_arctic').replace(
'_low', '') + '.wav')
x = msinference.inference(text,
style_vector,
alpha=0.3,
beta=0.7,
diffusion_steps=7,
embedding_scale=1)
# Fallback - MMS TTS - Non-English
else:
# dont split foreign sentences: Avoids re-load of VITS & random speaker change issue
x = msinference.foreign(text=text,
lang=voice, # voice = 'romanian', 'serbian' 'hungarian'
speed=speed) # normalisation externally
# volume
x /= np.abs(x).max() + 1e-7 # amplify speech to full [-1,1]
print(x.shape, 'TTS OK')
return x
ONLY_PARSE_TABLE = False # also generate .wavs if not already uploaded to issue and links are taken
if not ONLY_PARSE_TABLE:
from pathlib import Path
Path('foreign_accents_en_dir/').mkdir(parents=True, exist_ok=True)
for v in foreign_accents_en:
_str = v[:-4] # strip .wav
print(_str)
try:
soundfile.write('foreign_accents_en_dir/' + v, tts_entry(voice=_str), 24000)
# checkt out folder to count all phonemized voices
except RuntimeError:
# return F.conv2d(input, weight, bias, self.stride, # ENGLISH letters not phonemisable for this ACCENT/LANG
# RuntimeError: Calculated padded input size per channel: (5 x 4). Kernel size: (5 x 5). Kernel size can't be greater than actual input size
print('SKIP', _str)
# --------------------------------------------------- Uploads in hf issue
upload_wav = ['',
'',
'',
'',
'',
'',
'',
'',
'',
'',
'',
'',
'',
'',
'',
'',
'',
'',
'',
'',
'',
'',
'',
'',
'',
'',
'',
'',
'',
'',
'',
'',
'',
'',
'',
'',
'',
'',
'',
'',
'',
'',
'',
'',
'',
'',
'',
'',
'',
'',
'',
'',
'',
'',
'',
'',
'',
'',
'',
'',
'',
'',
'',
'',
'',
'',
'',
'',
'',
'',
'',
'',
'',
'',
'',
'',
'',
'',
'',
'',
'',
'',
'',
'',
'',
'',
'',
'',
'',
'',
'',
'',
'',
'',
'',
'',
'',
'',
'',
'',
'',
'',
'',
'',
'',
'',
'',
'',
'',
'',
'',
'',
'',
'',
'',
'',
'',
'',
'',
'',
'',
'',
'',
'',
'',
'',
'',
'',
'',
'',
'',
'',
'',
'',
'',
'',
'',
'',
'',
'',
'',
'',
'',
'',
'',
'',
'',
'',
'',
'',
'',
'',
'',
'',
'',
'',
'',
'',
'',
'',
'',
'',
'',
'',
'',
'',
'',
'',
'',
'',
'',
'',
'',
'',
'',
'',
'',
'',
'',
'',
'',
'',
'',
'',
'',
'',
'',
'',
'',
'',
'',
'',
'',
'',
'',
'',
'',]
tmp = '| | voice | TTS |
|---|
'
for j, (voice, _f) in enumerate(zip(sorted(os.listdir('foreign_accents_en_dir/')), upload_wav)):
tmp += f'\n| \n\n{j}\n\n | \n\n\n {voice[:-4]} \n\n | \n\n\n {_f} \n\n | \n
\n'
tmp += '\n\n
'
with open('table_for_issue_accents_upload.txt', 'w') as f:
f.write(tmp)