Spaces:
Build error
Build error
File size: 9,388 Bytes
597e812 ce48e72 597e812 ce48e72 597e812 ce48e72 597e812 |
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 99 100 101 102 103 104 105 106 107 108 109 110 111 112 113 114 115 116 117 118 119 120 121 122 123 124 125 126 127 128 129 130 131 132 133 134 135 136 137 138 139 140 141 142 143 144 145 146 147 148 149 150 151 152 153 154 155 156 157 158 159 160 161 162 163 164 165 166 167 168 169 170 171 172 173 174 175 176 177 178 179 180 181 182 183 184 185 186 187 188 189 190 191 192 193 194 195 196 197 198 199 200 201 202 203 204 205 206 207 208 209 210 211 212 213 214 215 216 217 218 219 220 221 222 223 224 225 226 227 228 229 230 231 232 233 234 235 236 237 238 239 240 241 242 243 244 245 246 247 248 249 250 251 252 253 254 255 256 257 258 259 260 261 262 263 264 265 266 267 268 269 270 271 272 273 274 275 276 277 278 279 |
from auralis import TTS, TTSRequest, TTSOutput, setup_logger
from gradio import File, Files, Slider
import torch
from tts_ui.utils import (
calculate_byte_size,
split_text_into_chunks,
tmp_dir,
extract_text_from_epub,
text_from_file,
convert_audio,
)
from tts_ui.utils.doc_processor import DocumentProcessor
import hashlib
import torchaudio
import time
from pathlib import Path
import os
# Loading the TTS engine first and assign it to the class.
# This looks ugly, but it works
logger = setup_logger(__file__)
tts = TTS()
model_path = "AstraMindAI/xttsv2" # change this if you have a different model
gpt_model = "AstraMindAI/xtts2-gpt"
try:
cpu_only = torch.cuda.is_available()
if cpu_only:
os.environ["VLLM_NO_GPU"] = "1"
os.environ["TRITON_CPU_ONLY"] = "1"
tts: TTS = tts.from_pretrained(
model_name_or_path=model_path,
gpt_model=gpt_model,
enforce_eager=False,
max_seq_len_to_capture=4096, # Match WSL2 page size
scheduler_max_concurrency=4,
)
logger.info(f"Successfully loaded model {model_path}")
except Exception as e:
error_msg = f"Failed to load model: {e}."
logger.error(error_msg)
raise Exception(error_msg)
class AuralisTTSEngine:
def __init__(self):
self.logger = logger
self.tts: TTS = tts
self.model_path: str = model_path
self.gpt_model: str = gpt_model
self.tmp_dir: Path = tmp_dir
self.doc_processor = DocumentProcessor
def process_text_and_generate(
self,
input_text: str,
ref_audio_files: str | list[str] | bytes | list[bytes],
speed: float,
enhance_speech: bool,
temperature: float,
top_p: float,
top_k: float,
repetition_penalty: float,
language: str = "auto",
*args,
):
"""Process text and generate audio."""
log_messages: str = ""
if not ref_audio_files:
log_messages += "Please provide at least one reference audio!\n"
return None, log_messages
input_size = calculate_byte_size(input_text)
# use the chunking process if the text is too large
if input_size > 45000:
self.logger.info(
f"Found {input_size} bytes of text. Switching to chunk mode."
)
# todo: this function has a couple of overlapping functions as normal processing. I need to optimize the code
return self._process_large_text(
input_text,
ref_audio_files,
speed,
enhance_speech,
temperature,
top_p,
top_k,
repetition_penalty,
language,
)
else:
try:
with torch.no_grad():
# clone voices from all file paths (shorten them)
base64_voices: str | list[str] | bytes | list[bytes] = (
ref_audio_files[:5]
)
request = TTSRequest(
text=input_text,
speaker_files=base64_voices,
stream=False,
enhance_speech=enhance_speech,
temperature=temperature,
top_p=top_p,
top_k=top_k,
repetition_penalty=repetition_penalty,
language=language,
)
output: TTSOutput = self.tts.generate_speech(request)
if output:
if speed != 1:
output.change_speed(speed)
log_messages += f"✅ Successfully Generated audio\n"
self.logger.info(log_messages)
# return the sample rate and the audio file as a byte array
return (
output.sample_rate,
convert_audio(output.array),
), log_messages
else:
log_messages += "❌ No output was generated. Check that the model was correctly loaded\n"
return None, log_messages
except Exception as e:
self.logger.error(f"Error: {e}")
log_messages += f"❌ An Error occured: {e}\n"
return None, log_messages
def _process_large_text(
self,
input_full_text: str,
ref_audio_files: str | list[str] | bytes | list[bytes],
speed: float,
enhance_speech: bool,
temperature: float,
top_p: float,
top_k: float,
repetition_penalty: float,
language: str = "auto",
):
"""Process text in chunks and combine results"""
log_messages: str = ""
if not ref_audio_files:
log_messages += "Please provide at least one reference audio!\n"
return None, log_messages
base64_voices: str | list[str] | bytes | list[bytes] = ref_audio_files[:5]
chunks: list[str] = split_text_into_chunks(input_full_text)
print(f"Created {len(chunks)} chunks")
audio_segments: list[TTSOutput] = []
for idx, chunk in enumerate(chunks):
request = TTSRequest(
text=chunk,
speaker_files=base64_voices,
stream=False,
enhance_speech=enhance_speech,
temperature=temperature,
top_p=top_p,
top_k=top_k,
repetition_penalty=repetition_penalty,
language=language,
)
try:
with torch.no_grad():
audio = self.tts.generate_speech(request)
audio_segments.append(audio)
self.logger.info(f"Processed {idx + 1} chunks out of {len(chunks)}")
except Exception as e:
log_messages += f"❌ Chunk processing failed: {e}\n"
return None, log_messages
if len(audio_segments) <= 0:
log_messages += f"❌ Chunk processing failed. Chunk size: {len(chunks)}\n"
return None, log_messages
combined_output: TTSOutput = TTSOutput.combine_outputs(audio_segments)
if speed != 1:
combined_output.change_speed(speed)
log_messages += f"✅ Successfully Generated audio\n"
# return combined_output
return (
combined_output.sample_rate,
convert_audio(combined_output.array),
), log_messages
def process_file_and_generate(
self,
file_input: File,
ref_audio_files_file: Files,
speed_file: Slider,
enhance_speech_file,
temperature_file,
top_p_file,
top_k_file,
repetition_penalty_file,
language_file,
):
# todo: refactor this to use the document processor object
if file_input:
file_extension: str = Path(file_input.name).suffix
match file_extension:
case ".epub":
input_text: str = extract_text_from_epub(file_input.name)
case ".txt" | ".md":
input_text = text_from_file(file_input.name)
case _:
return (
None,
"Unsupported file format, it needs to be either .epub or .txt",
)
return self._process_large_text(
input_text,
ref_audio_files_file,
speed_file,
enhance_speech_file,
temperature_file,
top_p_file,
top_k_file,
repetition_penalty_file,
language_file,
)
else:
return None, "Please provide an .epub or .txt file!"
def process_mic_and_generate(
self,
input_text_mic,
mic_ref_audio,
speed_mic,
enhance_speech_mic,
temperature_mic,
top_p_mic,
top_k_mic,
repetition_penalty_mic,
language_mic,
):
if mic_ref_audio:
data: bytes = str(time.time()).encode("utf-8")
hash: str = hashlib.sha1(data).hexdigest()[:10]
output_path = self.tmp_dir / (f"mic_{hash}.wav")
torch_audio: torch.Tensor = torch.from_numpy(mic_ref_audio[1].astype(float))
try:
torchaudio.save(
str(output_path), torch_audio.unsqueeze(0), mic_ref_audio[0]
)
return self.process_text_and_generate(
input_text_mic,
[Path(output_path)],
speed_mic,
enhance_speech_mic,
temperature_mic,
top_p_mic,
top_k_mic,
repetition_penalty_mic,
language_mic,
)
except Exception as e:
self.logger.error(f"Error saving audio file: {e}")
return None, f"Error saving audio file: {e}"
else:
return None, "Please record an audio!"
|