File size: 5,584 Bytes
eb3ba2e 6f1ebe2 5837809 6f1ebe2 eb3ba2e 6f1ebe2 eb3ba2e 6f1ebe2 5837809 6f1ebe2 eb3ba2e |
1 2 3 4 5 6 7 8 9 10 11 12 13 14 15 16 17 18 19 20 21 22 23 24 25 26 27 28 29 30 31 32 33 34 35 36 37 38 39 40 41 42 43 44 45 46 47 48 49 50 51 52 53 54 55 56 57 58 59 60 61 62 63 64 65 66 67 68 69 70 71 72 73 74 75 76 77 78 79 80 81 82 83 84 85 86 87 88 89 90 91 92 93 94 95 96 97 98 99 100 101 102 103 104 105 106 107 108 109 110 111 112 113 114 115 116 117 118 119 120 121 122 123 124 125 126 127 128 129 130 131 132 133 134 135 136 137 138 139 140 141 142 143 144 145 146 147 148 149 150 151 152 153 154 155 156 157 158 159 160 161 |
import os
import uuid
import numpy as np
import torch
import soundfile as sf
from gtts import gTTS
import edge_tts
from inference import Inference
import asyncio
from elevenlabs import voices, generate, save
from elevenlabs.api.error import UnauthenticatedRateLimitError
ELEVENLABS_VOICES_RAW = voices()
def get_elevenlabs_voice_names():
elevenlabs_voice_names = []
for voice in ELEVENLABS_VOICES_RAW:
elevenlabs_voice_names.append(voice.name)
return elevenlabs_voice_names
ELEVENLABS_VOICES_NAMES = get_elevenlabs_voice_names()
#git+https://github.com/suno-ai/bark.git
# from transformers import AutoProcessor, BarkModel
# import nltk
# from nltk.tokenize import sent_tokenize
# from bark import SAMPLE_RATE
# now_dir = os.getcwd()
def cast_to_device(tensor, device):
try:
return tensor.to(device)
except Exception as e:
print(e)
return tensor
# Buscar la forma de evitar descargar el archivo de 4gb cada vez que crea una instancia
# def _bark_conversion_(text, voice_preset):
# os.makedirs(os.path.join(now_dir, "tts"), exist_ok=True)
# device = "cuda:0" if torch.cuda.is_available() else "cpu"
# dtype = torch.float32 if "cpu" in device else torch.float16
# bark_processor = AutoProcessor.from_pretrained(
# "suno/bark",
# cache_dir=os.path.join(now_dir, "tts", "suno/bark"),
# torch_dtype=dtype,
# )
# bark_model = BarkModel.from_pretrained(
# "suno/bark",
# cache_dir=os.path.join(now_dir, "tts", "suno/bark"),
# torch_dtype=dtype,
# ).to(device)
# # bark_model.enable_cpu_offload()
# inputs = bark_processor(text=[text], return_tensors="pt", voice_preset=voice_preset)
# tensor_dict = {
# k: cast_to_device(v, device) if hasattr(v, "to") else v
# for k, v in inputs.items()
# }
# speech_values = bark_model.generate(**tensor_dict, do_sample=True)
# sampling_rate = bark_model.generation_config.sample_rate
# speech = speech_values.cpu().numpy().squeeze()
# return speech, sampling_rate
def tts_infer(tts_text, model_url, tts_method, tts_model, tts_api_key):
if not tts_text:
return 'Primero escribe el texto que quieres convertir.', None
if not tts_model:
return 'Selecciona un modelo TTS antes de convertir.', None
f0_method = "harvest"
output_folder = "audios"
os.makedirs(output_folder, exist_ok=True)
converted_tts_filename = os.path.join(output_folder, f"tts_out_{uuid.uuid4()}.wav")
success = False
if len(tts_text) > 60:
tts_text = tts_text[:60]
print("DEMO; limit to 60 characters")
language = tts_model[:2]
if tts_method == "Edge-tts":
try:
asyncio.run(
edge_tts.Communicate(
tts_text, "-".join(tts_model.split("-")[:-1])
).save(converted_tts_filename)
)
success = True
except Exception as e:
print("ERROR", e)
try:
tts = gTTS(tts_text, lang=language)
tts.save(converted_tts_filename)
print(
f"No audio was received. Please change the tts voice for {tts_model}. USING gTTS."
)
success = True
except:
tts = gTTS("a", lang=language)
tts.save(converted_tts_filename)
print("Error: Audio will be replaced.")
success = False
if tts_method == 'ElevenLabs':
try:
audio = generate(
text=tts_text,
voice=tts_model,
model="eleven_multilingual_v2",
api_key=tts_api_key
)
save(audio=audio, filename=converted_tts_filename)
success = True
except UnauthenticatedRateLimitError:
return "Necesitas configurar tu API Key para usar elevenlabs", None
if not model_url:
return 'Pon la url del modelo si quieres aplicarle otro tono.', converted_tts_filename
# elif tts_method == "Bark-tts":
# try:
# script = tts_text.replace("\n", " ").strip()
# sentences = sent_tokenize(script)
# silence = np.zeros(int(0.25 * SAMPLE_RATE))
# pieces = []
# for sentence in sentences:
# audio_array, _ = _bark_conversion_(sentence, tts_model.split("-")[0])
# pieces += [audio_array, silence.copy()]
# sf.write(
# file=converted_tts_filename, samplerate=SAMPLE_RATE, data=np.concatenate(pieces)
# )
# except Exception as e:
# print(f"{e}")
# return None, None
if success:
inference = Inference(
model_name=model_url,
f0_method=f0_method,
source_audio_path=converted_tts_filename,
output_file_name=os.path.join("./audio-outputs", os.path.basename(converted_tts_filename)),
)
output = inference.run()
if os.path.exists(converted_tts_filename):
os.remove(converted_tts_filename)
if os.path.exists(os.path.join("weights", inference.model_name)):
os.remove(os.path.join("weights", inference.model_name))
if 'success' in output and output['success']:
return output, output['file']
else:
return output, None
else:
return "Ocurrió un error durante la conversión", None
|