Spaces:
Build error
Build error
File size: 48,918 Bytes
a775cf4 | 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 99 100 101 102 103 104 105 106 107 108 109 110 111 112 113 114 115 116 117 118 119 120 121 122 123 124 125 126 127 128 129 130 131 132 133 134 135 136 137 138 139 140 141 142 143 144 145 146 147 148 149 150 151 152 153 154 155 156 157 158 159 160 161 162 163 164 165 166 167 168 169 170 171 172 173 174 175 176 177 178 179 180 181 182 183 184 185 186 187 188 189 190 191 192 193 194 195 196 197 198 199 200 201 202 203 204 205 206 207 208 209 210 211 212 213 214 215 216 217 218 219 220 221 222 223 224 225 226 227 228 229 230 231 232 233 234 235 236 237 238 239 240 241 242 243 244 245 246 247 248 249 250 251 252 253 254 255 256 257 258 259 260 261 262 263 264 265 266 267 268 269 270 271 272 273 274 275 276 277 278 279 280 281 282 283 284 285 286 287 288 289 290 291 292 293 294 295 296 297 298 299 300 301 302 303 304 305 306 307 308 309 310 311 312 313 314 315 316 317 318 319 320 321 322 323 324 325 326 327 328 329 330 331 332 333 334 335 336 337 338 339 340 341 342 343 344 345 346 347 348 349 350 351 352 353 354 355 356 357 358 359 360 361 362 363 364 365 366 367 368 369 370 371 372 373 374 375 376 377 378 379 380 381 382 383 384 385 386 387 388 389 390 391 392 393 394 395 396 397 398 399 400 401 402 403 404 405 406 407 408 409 410 411 412 413 414 415 416 417 418 419 420 421 422 423 424 425 426 427 428 429 430 431 432 433 434 435 436 437 438 439 440 441 442 443 444 445 446 447 448 449 450 451 452 453 454 455 456 457 458 459 460 461 462 463 464 465 466 467 468 469 470 471 472 473 474 475 476 477 478 479 480 481 482 483 484 485 486 487 488 489 490 491 492 493 494 495 496 497 498 499 500 501 502 503 504 505 506 507 508 509 510 511 512 513 514 515 516 517 518 519 520 521 522 523 524 525 526 527 528 529 530 531 532 533 534 535 536 537 538 539 540 541 542 543 544 545 546 547 548 549 550 551 552 553 554 555 556 557 558 559 560 561 562 563 564 565 566 567 568 569 570 571 572 573 574 575 576 577 578 579 580 581 582 583 584 585 586 587 588 589 590 591 592 593 594 595 596 597 598 599 600 601 602 603 604 605 606 607 608 609 610 611 612 613 614 615 616 617 618 619 620 621 622 623 624 625 626 627 628 629 630 631 632 633 634 635 636 637 638 639 640 641 642 643 644 645 646 647 648 649 650 651 652 653 654 655 656 657 658 659 660 661 662 663 664 665 666 667 668 669 670 671 672 673 674 675 676 677 678 679 680 681 682 683 684 685 686 687 688 689 690 691 692 693 694 695 696 697 698 699 700 701 702 703 704 705 706 707 708 709 710 711 712 713 714 715 716 717 718 719 720 721 722 723 724 725 726 727 728 729 730 731 732 733 734 735 736 737 738 739 740 741 742 743 744 745 746 747 748 749 750 751 752 753 754 755 756 757 758 759 760 761 762 763 764 765 766 767 768 769 770 771 772 773 774 775 776 777 778 779 780 781 782 783 784 785 786 787 788 789 790 791 792 793 794 795 796 797 798 799 800 801 802 803 804 805 806 807 808 809 810 811 812 813 814 815 816 817 818 819 820 821 822 823 824 825 826 827 828 829 830 831 832 833 834 835 836 837 838 839 840 841 842 843 844 845 846 847 848 849 850 851 852 853 854 855 856 857 858 859 860 861 862 863 864 865 866 867 868 869 870 871 872 873 874 875 876 877 878 879 880 881 882 883 884 885 886 887 888 889 890 891 892 893 894 895 896 897 898 899 900 901 902 903 904 905 906 907 908 909 910 911 912 913 914 915 916 917 918 919 920 921 922 923 924 925 926 927 928 929 930 931 932 933 934 935 936 937 938 939 940 941 942 943 944 945 946 947 948 949 950 951 952 953 954 955 956 957 958 959 960 961 962 963 964 965 966 967 968 969 970 971 972 973 974 975 976 977 978 979 980 981 982 983 984 985 986 987 988 989 990 991 992 993 994 995 996 997 998 999 1000 1001 1002 1003 1004 1005 1006 1007 1008 1009 1010 1011 1012 1013 1014 1015 1016 1017 1018 1019 1020 1021 1022 1023 1024 1025 1026 1027 1028 1029 1030 1031 1032 1033 1034 1035 1036 1037 1038 1039 1040 1041 1042 1043 1044 1045 1046 1047 1048 1049 1050 1051 1052 1053 1054 1055 1056 1057 1058 1059 1060 1061 1062 1063 1064 1065 1066 1067 1068 1069 1070 1071 1072 1073 1074 1075 1076 1077 1078 1079 1080 1081 1082 1083 1084 1085 1086 1087 1088 1089 1090 1091 | """
IndexTTS2 Handler for Ultimate TTS Studio
Provides integration with IndexTTS-2 Text-to-Speech system with advanced emotion control
"""
import os
import sys
import warnings
import numpy as np
import torch
import tempfile
import json
from pathlib import Path
from typing import Optional, Union, Tuple, Dict, Any, List
from datetime import datetime
import librosa
import soundfile as sf
# Suppress warnings
warnings.filterwarnings('ignore')
# Setup IndexTTS2 module path and cache directories
current_dir = os.path.dirname(os.path.abspath(__file__))
indextts2_base_path = os.path.join(current_dir, 'indextts2')
indextts2_module_path = os.path.join(indextts2_base_path, 'indextts')
# Set up cache directory early to avoid ModelScope conflicts
cache_dir = os.path.join(current_dir, 'checkpoints', 'indextts2', 'cache')
os.makedirs(cache_dir, exist_ok=True)
# Set environment variables before any imports
os.environ['HF_HOME'] = cache_dir
os.environ['TRANSFORMERS_CACHE'] = cache_dir
os.environ['HF_HUB_CACHE'] = cache_dir
os.environ['MODELSCOPE_CACHE'] = cache_dir
os.environ['HUGGINGFACE_HUB_CACHE'] = cache_dir
# Add paths to sys.path for proper module resolution
# Ensure the app root is included so implicit namespace package 'indextts2' works
paths_to_add = [current_dir, indextts2_base_path, indextts2_module_path]
for path in paths_to_add:
if path not in sys.path:
sys.path.insert(0, path)
# Global handler instance
_indextts2_handler = None
INDEXTTS2_AVAILABLE = False
# Try multiple import strategies
def try_import_indextts2():
"""Try different import strategies for IndexTTS2"""
global INDEXTTS2_AVAILABLE
# Strategy 1: Set up proper module structure and import
try:
import importlib
# Import the bundled package and alias as top-level 'indextts' for internal absolute imports
indextts_pkg = importlib.import_module('indextts2.indextts')
sys.modules['indextts'] = indextts_pkg
# Import IndexTTS2
from indextts2.indextts.infer_v2 import IndexTTS2
INDEXTTS2_AVAILABLE = True
print("β
IndexTTS2 loaded successfully (Strategy 1)")
return IndexTTS2
except Exception as e1:
print(f"β οΈ Strategy 1 failed: {e1}")
# Strategy 2: Direct file loading with module setup
try:
import importlib
import importlib.util
# Ensure 'indextts' alias points to bundled package for internal imports
try:
indextts_pkg = importlib.import_module('indextts2.indextts')
sys.modules['indextts'] = indextts_pkg
except Exception:
pass
# Load infer_v2.py directly but set up the environment first
infer_v2_path = os.path.join(indextts2_module_path, "infer_v2.py")
spec = importlib.util.spec_from_file_location("indextts2.indextts.infer_v2", infer_v2_path)
infer_v2_module = importlib.util.module_from_spec(spec)
# Register both names for compatibility
sys.modules['indextts2.indextts.infer_v2'] = infer_v2_module
sys.modules['indextts.infer_v2'] = infer_v2_module
# Execute the module
spec.loader.exec_module(infer_v2_module)
IndexTTS2 = infer_v2_module.IndexTTS2
INDEXTTS2_AVAILABLE = True
print("β
IndexTTS2 loaded successfully (Strategy 2)")
return IndexTTS2
except Exception as e2:
print(f"β οΈ Strategy 2 failed: {e2}")
# All strategies failed
INDEXTTS2_AVAILABLE = False
print("β All import strategies failed")
print(" IndexTTS2 requires complex module dependencies that couldn't be resolved")
print(" This is a known limitation with the current IndexTTS2 package structure")
return None
# Try to import IndexTTS2
IndexTTS2 = try_import_indextts2()
def check_indextts2_models():
"""Check if IndexTTS2 models are available"""
model_dir = Path("checkpoints/indextts2")
config_path = model_dir / "config.yaml"
if not config_path.exists():
return False
# Check for essential IndexTTS-2 model files
essential_files = ["gpt.pth", "s2mel.pth", "bpe.model"]
for filename in essential_files:
if not (model_dir / filename).exists():
return False
return True
def ensure_indextts2_dependencies():
"""Ensure all IndexTTS2 dependencies are available"""
try:
from transformers import SeamlessM4TFeatureExtractor, Wav2Vec2BertModel
from huggingface_hub import hf_hub_download, snapshot_download
model_dir = Path("checkpoints/indextts2")
cache_dir = model_dir / "cache"
cache_dir.mkdir(parents=True, exist_ok=True)
# Set environment variables to use local cache
os.environ['HF_HOME'] = str(cache_dir)
os.environ['TRANSFORMERS_CACHE'] = str(cache_dir)
os.environ['HF_HUB_CACHE'] = str(cache_dir)
os.environ['MODELSCOPE_CACHE'] = str(cache_dir)
print("π§ Ensuring IndexTTS2 dependencies are available...")
# Check and download facebook/w2v-bert-2.0 (feature extractor and model)
try:
print(" π Checking facebook/w2v-bert-2.0...")
repo_id = "facebook/w2v-bert-2.0"
# Prefer a concrete snapshot path inside our cache if it already exists
snapshots_dir = cache_dir / f"models--{repo_id.replace('/', '--')}" / "snapshots"
local_repo_path = None
if snapshots_dir.exists() and any(snapshots_dir.iterdir()):
# pick most recent snapshot
local_repo_path = str(max(snapshots_dir.iterdir(), key=lambda p: p.stat().st_mtime))
if local_repo_path is None:
local_repo_path = snapshot_download(repo_id=repo_id, cache_dir=str(cache_dir))
# Validate both processor and model load locally (no remote access)
SeamlessM4TFeatureExtractor.from_pretrained(local_repo_path, local_files_only=True)
Wav2Vec2BertModel.from_pretrained(local_repo_path, local_files_only=True)
print(" β
facebook/w2v-bert-2.0 ready")
except Exception:
# Non-fatal: real loading will happen later with fallback logic
print(" β οΈ facebook/w2v-bert-2.0 not fully validated; will download on demand")
# Check and download MaskGCT semantic codec
try:
print(" π Checking amphion/MaskGCT...")
hf_hub_download(
"amphion/MaskGCT",
filename="semantic_codec/model.safetensors",
cache_dir=str(cache_dir)
)
print(" β
MaskGCT semantic codec ready")
except Exception as e:
print(f" β οΈ MaskGCT issue: {e}")
# Check and download campplus
try:
print(" π Checking funasr/campplus...")
hf_hub_download(
"funasr/campplus",
filename="campplus_cn_common.bin",
cache_dir=str(cache_dir)
)
print(" β
campplus ready")
except Exception as e:
print(f" β οΈ campplus issue: {e}")
return True
except Exception as e:
print(f"β Error ensuring dependencies: {e}")
return False
def download_indextts2_models():
"""Download IndexTTS2 models and dependencies from HuggingFace"""
try:
from huggingface_hub import hf_hub_download
from transformers import SeamlessM4TFeatureExtractor
import requests
except ImportError:
print("β οΈ Cannot auto-download IndexTTS2 models - missing huggingface_hub or transformers")
print(" Install with: pip install huggingface_hub transformers requests")
return False
repo_id = "IndexTeam/IndexTTS-2"
model_dir = Path("checkpoints/indextts2")
# Create directory if it doesn't exist
model_dir.mkdir(parents=True, exist_ok=True)
print("π― Auto-downloading IndexTTS-2 models and dependencies...")
print(" This may take several minutes on first run...")
# Step 1: Download main IndexTTS-2 model files
required_files = [
"config.yaml",
"bpe.model",
"gpt.pth",
"s2mel.pth",
"feat1.pt",
"feat2.pt",
"wav2vec2bert_stats.pt"
]
# Qwen emotion model files (in subfolder)
qwen_files = [
"qwen0.6bemo4-merge/config.json",
"qwen0.6bemo4-merge/generation_config.json",
"qwen0.6bemo4-merge/model.safetensors",
"qwen0.6bemo4-merge/tokenizer.json",
"qwen0.6bemo4-merge/tokenizer_config.json",
"qwen0.6bemo4-merge/vocab.json"
]
all_files = required_files + qwen_files
print("π₯ Downloading IndexTTS-2 main models...")
for filename in all_files:
file_path = model_dir / filename
if file_path.exists():
print(f" β
{filename} already exists")
continue
try:
print(f" β¬οΈ Downloading {filename}...")
# Create subdirectories if needed
file_path.parent.mkdir(parents=True, exist_ok=True)
hf_hub_download(
repo_id=repo_id,
filename=filename,
local_dir=str(model_dir),
local_dir_use_symlinks=False
)
print(f" β
{filename} downloaded")
except Exception as e:
print(f" β Failed to download {filename}: {e}")
continue
# Step 2: Download dependency models
print("π₯ Downloading IndexTTS-2 dependencies...")
try:
print(" β¬οΈ Downloading facebook/w2v-bert-2.0 feature extractor...")
# Set cache directory before downloading
cache_dir = str(model_dir / "cache")
os.makedirs(cache_dir, exist_ok=True)
# Set environment variables to use local cache
os.environ['HF_HOME'] = cache_dir
os.environ['TRANSFORMERS_CACHE'] = cache_dir
os.environ['HF_HUB_CACHE'] = cache_dir
# Download to local cache
SeamlessM4TFeatureExtractor.from_pretrained(
"facebook/w2v-bert-2.0",
cache_dir=cache_dir
)
print(" β
facebook/w2v-bert-2.0 downloaded to local cache")
except Exception as e:
print(f" β Failed to download facebook/w2v-bert-2.0: {e}")
print(" π‘ This model will be downloaded automatically when needed")
try:
print(" β¬οΈ Downloading amphion/MaskGCT semantic codec...")
# Download the semantic codec model
semantic_codec_path = hf_hub_download(
"amphion/MaskGCT",
filename="semantic_codec/model.safetensors",
cache_dir=str(model_dir / "cache")
)
print(f" β
MaskGCT semantic codec downloaded to {semantic_codec_path}")
except Exception as e:
print(f" β Failed to download MaskGCT semantic codec: {e}")
try:
print(" β¬οΈ Downloading funasr/campplus model...")
# Download campplus model
campplus_path = hf_hub_download(
"funasr/campplus",
filename="campplus_cn_common.bin",
cache_dir=str(model_dir / "cache")
)
print(f" β
campplus model downloaded to {campplus_path}")
except Exception as e:
print(f" β Failed to download campplus model: {e}")
# Step 3: Check if essential files are present
essential_files = ["config.yaml", "gpt.pth", "s2mel.pth", "bpe.model"]
missing_essential = []
for filename in essential_files:
if not (model_dir / filename).exists():
missing_essential.append(filename)
if missing_essential:
print(f"β Essential files missing: {missing_essential}")
return False
print("π IndexTTS-2 models and dependencies ready!")
print("π‘ Note: Some models are cached by transformers/huggingface_hub")
return True
def get_indextts2_handler():
"""Get the global IndexTTS2 handler instance (singleton)"""
global _indextts2_handler
if _indextts2_handler is None:
_indextts2_handler = IndexTTS2Handler()
return _indextts2_handler
class IndexTTS2Handler:
"""Handler for IndexTTS2 TTS system with advanced emotion control"""
def __init__(self):
self.model = None
self.device = self._get_device()
self.sample_rate = 22050
self.model_path = "IndexTeam/IndexTTS-2"
self.checkpoints_dir = Path("checkpoints/indextts2")
self.checkpoints_dir.mkdir(parents=True, exist_ok=True)
# Emotion control modes
self.emotion_modes = {
'audio_reference': 'Use audio file for emotion reference',
'vector_control': 'Manual emotion vector adjustment',
'text_description': 'Natural language emotion description'
}
# Emotion vectors for manual control
self.emotion_vectors = {
'happy': 0.0,
'angry': 0.0,
'sad': 0.0,
'afraid': 0.0,
'disgusted': 0.0,
'melancholic': 0.0,
'surprised': 0.0,
'calm': 0.0
}
def _get_device(self):
"""Get the appropriate device for inference"""
if torch.cuda.is_available():
return "cuda"
elif hasattr(torch.backends, 'mps') and torch.backends.mps.is_available():
return "mps"
else:
return "cpu"
def initialize_model(self):
"""Initialize the IndexTTS2 model"""
if not INDEXTTS2_AVAILABLE or IndexTTS2 is None:
return False, "β IndexTTS2 not available"
try:
print("π― Initializing IndexTTS2 model...")
# Set up cache directories to avoid ModelScope conflicts
cache_dir = str(self.checkpoints_dir / "cache")
os.makedirs(cache_dir, exist_ok=True)
# Set environment variables to use local cache (multiple cache systems)
os.environ['HF_HOME'] = cache_dir
os.environ['TRANSFORMERS_CACHE'] = cache_dir
os.environ['HF_HUB_CACHE'] = cache_dir
os.environ['MODELSCOPE_CACHE'] = cache_dir
os.environ['HUGGINGFACE_HUB_CACHE'] = cache_dir
# Also set the cache directory in the IndexTTS2 infer_v2.py file
# This overrides the hardcoded cache path
original_hf_cache = os.environ.get('HF_HUB_CACHE', './checkpoints/hf_cache')
os.environ['HF_HUB_CACHE'] = cache_dir
# Check if models are available, download if needed
if not check_indextts2_models():
print("π₯ IndexTTS2 models not found, downloading...")
if not download_indextts2_models():
return False, "β Failed to download IndexTTS2 models"
# Ensure all dependencies are available
print("π§ Ensuring dependencies are ready...")
try:
ensure_indextts2_dependencies()
except Exception as dep_error:
print(f"β οΈ Dependency check failed: {dep_error}")
print("π‘ Continuing with initialization - dependencies will be downloaded as needed")
# Initialize IndexTTS2 with config path and model directory
config_path = str(self.checkpoints_dir / "config.yaml")
model_dir = str(self.checkpoints_dir)
print(f"π§ Using config: {config_path}")
print(f"π§ Using model dir: {model_dir}")
print(f"π§ Using cache dir: {cache_dir}")
self.model = IndexTTS2(
cfg_path=config_path,
model_dir=model_dir,
device=self.device,
use_fp16=self.device != "cpu"
)
print(f"β
IndexTTS2 model loaded on {self.device}")
return True, "β
IndexTTS2 model loaded successfully"
except Exception as e:
import traceback
traceback.print_exc()
print(f"β Error initializing IndexTTS2: {e}")
return False, f"β Error initializing IndexTTS2: {str(e)}"
def unload_model(self):
"""Unload the IndexTTS2 model to free memory"""
try:
if self.model is not None:
del self.model
self.model = None
# Force garbage collection
import gc
gc.collect()
if torch.cuda.is_available():
torch.cuda.empty_cache()
return "β
IndexTTS2 model unloaded successfully"
except Exception as e:
return f"β οΈ Error unloading IndexTTS2: {str(e)}"
def is_model_loaded(self):
"""Check if the model is loaded"""
return self.model is not None
def preprocess_audio(self, audio_path: str, max_duration: float = 15.0):
"""Preprocess reference audio for optimal performance"""
try:
# Load audio
audio, sr = librosa.load(audio_path, sr=self.sample_rate)
# Trim silence
audio, _ = librosa.effects.trim(audio, top_db=20)
# Limit duration for optimal performance
max_samples = int(max_duration * self.sample_rate)
if len(audio) > max_samples:
audio = audio[:max_samples]
return audio, sr
except Exception as e:
print(f"β Error preprocessing audio: {e}")
return None, None
def _preprocess_text_for_tensor_safety(self, text: str) -> str:
"""Preprocess text to avoid tensor dimension mismatches in IndexTTS2"""
import re
# Remove excessive punctuation that might cause issues
text = re.sub(r'[.]{3,}', '...', text) # Limit ellipsis
text = re.sub(r'[!]{2,}', '!', text) # Limit exclamation marks
text = re.sub(r'[?]{2,}', '?', text) # Limit question marks
# Clean up excessive whitespace
text = re.sub(r'\s+', ' ', text)
text = text.strip()
# Remove or replace problematic character sequences
text = re.sub(r'[^\w\s.,!?;:\'"()-]', '', text) # Keep only safe characters
# Ensure text doesn't end abruptly without punctuation
if text and text[-1] not in '.!?':
text += '.'
# Limit very long sentences that might cause tensor issues
sentences = re.split(r'([.!?]+)', text)
processed_sentences = []
for i in range(0, len(sentences), 2):
if i < len(sentences):
sentence = sentences[i].strip()
punctuation = sentences[i + 1] if i + 1 < len(sentences) else '.'
# If sentence is too long, split it at commas or conjunctions
if len(sentence) > 150:
# Split at commas, semicolons, or conjunctions
parts = re.split(r'(,|;|\s+and\s+|\s+but\s+|\s+or\s+)', sentence)
current_part = ""
for j, part in enumerate(parts):
if part.strip() in [',', ';', 'and', 'but', 'or']:
current_part += part
if len(current_part) > 80: # Split here
processed_sentences.append(current_part.strip() + '.')
current_part = ""
else:
if len(current_part + part) > 120 and current_part:
processed_sentences.append(current_part.strip() + '.')
current_part = part
else:
current_part += part
if current_part.strip():
processed_sentences.append(current_part.strip() + punctuation)
else:
processed_sentences.append(sentence + punctuation)
return ' '.join(processed_sentences)
def generate_speech(
self,
text: str,
reference_audio: Optional[str] = None,
emotion_mode: str = "audio_reference",
emotion_audio: Optional[str] = None,
emotion_vectors: Optional[Dict[str, float]] = None,
emotion_description: str = "",
temperature: float = 0.8,
top_p: float = 0.9,
top_k: int = 50,
repetition_penalty: float = 1.1,
max_mel_tokens: int = 1500,
seed: Optional[int] = None,
use_random: bool = True,
emo_alpha: float = 1.0
) -> Tuple[Optional[np.ndarray], str]:
"""
Generate speech using IndexTTS2 with emotion control
Args:
text: Text to synthesize
reference_audio: Path to reference audio for voice cloning
emotion_mode: Emotion control mode ('audio_reference', 'vector_control', 'text_description')
emotion_audio: Path to emotion reference audio
emotion_vectors: Dictionary of emotion intensities
emotion_description: Natural language emotion description
temperature: Sampling temperature
top_p: Top-p sampling parameter
top_k: Top-k sampling parameter
repetition_penalty: Repetition penalty
max_mel_tokens: Maximum mel tokens to generate
seed: Random seed for reproducibility
use_random: Enable random sampling
emo_alpha: Emotion blending alpha
"""
if not self.is_model_loaded():
return None, "β IndexTTS2 model not loaded. Please initialize first."
if not reference_audio or not os.path.exists(reference_audio):
return None, "β Reference audio is required for IndexTTS2"
try:
print(f"π― Generating speech with IndexTTS2...")
print(f" Text: {text[:50]}...")
print(f" Emotion mode: {emotion_mode}")
# Preprocess long text to avoid tensor dimension mismatch
# Split very long text into smaller chunks to prevent issues
if len(text) > 500: # If text is longer than 500 characters
print(f" β οΈ Long text detected ({len(text)} chars), splitting into chunks...")
# Split by sentences first, then by length if needed
import re
sentences = re.split(r'[.!?]+', text)
processed_chunks = []
current_chunk = ""
for sentence in sentences:
sentence = sentence.strip()
if not sentence:
continue
# If adding this sentence would make chunk too long, save current chunk
if len(current_chunk) + len(sentence) > 300 and current_chunk:
processed_chunks.append(current_chunk.strip())
current_chunk = sentence
else:
current_chunk += (" " + sentence if current_chunk else sentence)
# Add the last chunk
if current_chunk.strip():
processed_chunks.append(current_chunk.strip())
# If we have multiple chunks, process them separately and concatenate
if len(processed_chunks) > 1:
print(f" π Processing {len(processed_chunks)} text chunks...")
chunk_audios = []
for i, chunk in enumerate(processed_chunks):
print(f" π Processing chunk {i+1}/{len(processed_chunks)}: {chunk[:30]}...")
# Recursive call with shorter text
chunk_audio, chunk_message = self.generate_speech(
chunk, reference_audio, emotion_mode, emotion_audio,
emotion_vectors, emotion_description, temperature, top_p, top_k,
repetition_penalty, min(max_mel_tokens, 800), seed, use_random, emo_alpha
)
if chunk_audio is None:
return None, f"β Error processing chunk {i+1}: {chunk_message}"
chunk_audios.append(chunk_audio)
# Concatenate all chunk audios with small pauses
print(f" π΅ Combining {len(chunk_audios)} audio chunks...")
pause_samples = int(0.3 * self.sample_rate) # 0.3 second pause
pause_audio = np.zeros(pause_samples)
combined_audio = chunk_audios[0]
for chunk_audio in chunk_audios[1:]:
combined_audio = np.concatenate([combined_audio, pause_audio, chunk_audio])
return combined_audio, "β
Long text processed successfully in chunks"
else:
# Single chunk, continue with normal processing
text = processed_chunks[0]
# Set random seed if provided
if seed is not None:
torch.manual_seed(seed)
np.random.seed(seed)
# Prepare generation parameters
generation_kwargs = {
'do_sample': True,
'temperature': temperature,
'top_p': top_p,
'top_k': top_k,
'repetition_penalty': repetition_penalty,
'max_mel_tokens': max_mel_tokens
}
# Handle emotion control based on mode
emo_audio_prompt = None
emo_vector = None
use_emo_text = False
emo_text = None
if emotion_mode == "audio_reference" and emotion_audio and os.path.exists(emotion_audio):
emo_audio_prompt = emotion_audio
elif emotion_mode == "vector_control" and emotion_vectors:
# Convert emotion vectors to the format expected by IndexTTS2
emo_vector = []
for emotion in ['happy', 'angry', 'sad', 'afraid', 'disgusted', 'melancholic', 'surprised', 'calm']:
emo_vector.append(emotion_vectors.get(emotion, 0.0))
elif emotion_mode == "text_description" and emotion_description:
use_emo_text = True
emo_text = emotion_description
# Preprocess text to avoid tensor dimension issues
# Clean up text that might cause tensor mismatches
original_text = text
text = self._preprocess_text_for_tensor_safety(text)
if text != original_text:
print(f"π§ Text preprocessed to avoid tensor issues")
# Generate speech using the actual IndexTTS2 API with retry logic for tensor dimension issues
# Use smaller max_text_tokens_per_segment to prevent tensor dimension issues
max_text_tokens_per_segment = min(80, max_mel_tokens // 20) if len(text) > 200 else 120
# Implement retry logic with progressively smaller parameters
max_retries = 3
retry_count = 0
result = None
while retry_count < max_retries and result is None:
try:
# Adjust parameters based on retry count
if retry_count > 0:
print(f" π Tensor dimension mismatch (attempt {retry_count}/{max_retries})")
# Progressively reduce parameters to avoid tensor issues
max_text_tokens_per_segment = max(20, max_text_tokens_per_segment // 2)
generation_kwargs['max_mel_tokens'] = max(300, generation_kwargs['max_mel_tokens'] // 2)
print(f" π§ Retrying with max_text_tokens_per_segment={max_text_tokens_per_segment}, max_mel_tokens={generation_kwargs['max_mel_tokens']}")
result = self.model.infer(
spk_audio_prompt=reference_audio,
text=text,
output_path=None, # Return audio data instead of saving
emo_audio_prompt=emo_audio_prompt,
emo_alpha=emo_alpha,
emo_vector=emo_vector,
use_emo_text=use_emo_text,
emo_text=emo_text,
use_random=use_random,
max_text_tokens_per_segment=max_text_tokens_per_segment,
**generation_kwargs
)
except Exception as retry_error:
retry_error_msg = str(retry_error)
# Check if this is a tensor dimension mismatch error
if ("Sizes of tensors must match" in retry_error_msg or
"Expected size" in retry_error_msg or
"dimension" in retry_error_msg.lower()):
retry_count += 1
if retry_count >= max_retries:
# Final attempt: force text chunking with very small segments
print(f" π Final attempt: forcing text chunking...")
try:
# Split text into very small chunks and process separately
words = text.split()
chunk_size = max(5, len(words) // 4) # Very small chunks
text_chunks = []
for i in range(0, len(words), chunk_size):
chunk = " ".join(words[i:i + chunk_size])
text_chunks.append(chunk)
if len(text_chunks) > 1:
print(f" π Processing {len(text_chunks)} micro-chunks...")
chunk_audios = []
for j, chunk in enumerate(text_chunks):
print(f" π Micro-chunk {j+1}/{len(text_chunks)}: {chunk[:20]}...")
chunk_result = self.model.infer(
spk_audio_prompt=reference_audio,
text=chunk,
output_path=None,
emo_audio_prompt=emo_audio_prompt,
emo_alpha=emo_alpha,
emo_vector=emo_vector,
use_emo_text=use_emo_text,
emo_text=emo_text,
use_random=use_random,
max_text_tokens_per_segment=20,
max_mel_tokens=300
)
if chunk_result is not None:
if isinstance(chunk_result, tuple) and len(chunk_result) == 2:
_, chunk_audio = chunk_result
if isinstance(chunk_audio, torch.Tensor):
chunk_audio = chunk_audio.cpu().numpy()
# Ensure chunk_audio is 1D
if chunk_audio.ndim == 2:
if chunk_audio.shape[0] == 1:
chunk_audio = chunk_audio.flatten()
elif chunk_audio.shape[1] == 1:
chunk_audio = chunk_audio.flatten()
else:
# Take first channel if stereo
chunk_audio = chunk_audio[0] if chunk_audio.shape[0] < chunk_audio.shape[1] else chunk_audio[:, 0]
chunk_audios.append(chunk_audio)
if chunk_audios:
try:
# Ensure all chunks are 1D arrays before combining
normalized_chunks = []
for i, chunk_audio in enumerate(chunk_audios):
if chunk_audio.ndim > 1:
chunk_audio = chunk_audio.flatten()
# Ensure it's a valid audio array
if len(chunk_audio) == 0:
print(f" β οΈ Skipping empty chunk {i+1}")
continue
normalized_chunks.append(chunk_audio)
if normalized_chunks:
# Combine chunks with small pauses
pause_samples = int(0.1 * self.sample_rate)
pause_audio = np.zeros(pause_samples)
combined_audio = normalized_chunks[0]
for chunk_audio in normalized_chunks[1:]:
combined_audio = np.concatenate([combined_audio, pause_audio, chunk_audio])
print(f" β
Successfully combined {len(normalized_chunks)} micro-chunks")
result = (self.sample_rate, combined_audio)
break
else:
print(f" β No valid chunks to combine")
except Exception as combine_error:
print(f" β Error combining chunks: {combine_error}")
# Continue to raise the original error
except Exception as chunk_error:
print(f" β Chunking also failed: {chunk_error}")
raise retry_error # Re-raise original error
else:
continue # Try again with smaller parameters
else:
# Not a tensor dimension error, re-raise immediately
raise retry_error
if result is None:
return None, "β Failed to generate audio"
# IndexTTS2 returns (sample_rate, audio_data) tuple
if isinstance(result, tuple) and len(result) == 2:
sample_rate, audio_data = result
# Convert to numpy array if needed
if isinstance(audio_data, torch.Tensor):
audio_data = audio_data.cpu().numpy()
# Handle different audio data formats more robustly
if audio_data.ndim == 2:
# If stereo or transposed, take first channel or transpose
if audio_data.shape[0] == 2:
audio_data = audio_data[0] # Take first channel
elif audio_data.shape[1] == 1:
audio_data = audio_data.flatten() # Flatten single channel
elif audio_data.shape[0] == 1:
audio_data = audio_data.flatten() # Flatten single channel
else:
# Choose the dimension that makes more sense for audio
if audio_data.shape[0] < audio_data.shape[1]:
audio_data = audio_data[0] # Take first row
else:
audio_data = audio_data[:, 0] # Take first column
elif audio_data.ndim > 2:
# Handle higher dimensional arrays by flattening
audio_data = audio_data.flatten()
# Normalize audio to prevent clipping
if len(audio_data) > 0:
max_val = np.max(np.abs(audio_data))
if max_val > 0:
audio_data = audio_data / max_val * 0.95
print(f"β
Generated {len(audio_data)} samples at {sample_rate}Hz")
return audio_data, "β
Speech generated successfully"
else:
return None, "β Unexpected audio format returned"
except Exception as e:
import traceback
error_msg = str(e)
# Provide specific guidance for common tensor dimension errors
if "Sizes of tensors must match" in error_msg:
print("π Tensor dimension mismatch detected - this is likely due to long text processing")
print("π‘ Try reducing max_mel_tokens or splitting your text into shorter segments")
error_msg = f"Tensor dimension mismatch (likely due to long text): {error_msg}"
elif "Expected size" in error_msg and "but got size" in error_msg:
print("π Tensor size mismatch detected")
print("π‘ This may be resolved by using shorter text segments")
error_msg = f"Tensor size mismatch: {error_msg}"
traceback.print_exc()
return None, f"β Error generating speech: {error_msg}"
def get_model_info(self):
"""Get information about the loaded model"""
if not self.is_model_loaded():
return "β No model loaded"
info = {
'model_name': 'IndexTTS-2',
'device': self.device,
'sample_rate': self.sample_rate,
'emotion_modes': list(self.emotion_modes.keys()),
'supported_languages': ['English', 'Chinese'],
'features': [
'Zero-shot voice cloning',
'Advanced emotion control',
'Emotion-speaker disentanglement',
'Duration control',
'Multi-modal emotion input'
]
}
return info
# Model management functions
def init_indextts2():
"""Initialize IndexTTS2 model"""
if not INDEXTTS2_AVAILABLE:
return False, "β IndexTTS2 not available"
try:
handler = get_indextts2_handler()
success, message = handler.initialize_model()
return success, message
except Exception as e:
return False, f"β Error initializing IndexTTS2: {str(e)}"
def unload_indextts2():
"""Unload IndexTTS2 model"""
try:
handler = get_indextts2_handler()
return handler.unload_model()
except Exception as e:
return f"β οΈ Error unloading IndexTTS2: {str(e)}"
def generate_indextts2_tts(
text: str,
reference_audio: Optional[str] = None,
emotion_mode: str = "audio_reference",
emotion_audio: Optional[str] = None,
emotion_vectors: Optional[Dict[str, float]] = None,
emotion_description: str = "",
temperature: float = 0.8,
top_p: float = 0.9,
top_k: int = 50,
repetition_penalty: float = 1.1,
max_mel_tokens: int = 1500,
seed: Optional[int] = None,
use_random: bool = True,
emo_alpha: float = 1.0,
effects_settings: Optional[Dict] = None,
audio_format: str = "wav",
skip_file_saving: bool = False
) -> Tuple[Optional[Union[str, Tuple]], str]:
"""
Generate TTS using IndexTTS2 with comprehensive emotion control
Returns:
Tuple of (audio_data, info_message)
audio_data can be either file path (str) or (sample_rate, audio_array) tuple
"""
try:
handler = get_indextts2_handler()
if not handler.is_model_loaded():
return None, "β IndexTTS2 model not loaded. Please load the model first."
# Generate speech
audio_array, message = handler.generate_speech(
text=text,
reference_audio=reference_audio,
emotion_mode=emotion_mode,
emotion_audio=emotion_audio,
emotion_vectors=emotion_vectors,
emotion_description=emotion_description,
temperature=temperature,
top_p=top_p,
top_k=top_k,
repetition_penalty=repetition_penalty,
max_mel_tokens=max_mel_tokens,
seed=seed,
use_random=use_random,
emo_alpha=emo_alpha
)
if audio_array is None:
return None, message
# Apply audio effects if specified
if effects_settings:
try:
# Import audio effects processing (assuming it exists in the main app)
from launch import apply_audio_effects
audio_array = apply_audio_effects(audio_array, handler.sample_rate, effects_settings)
except ImportError:
print("β οΈ Audio effects not available")
if skip_file_saving:
# Return audio data directly as tuple
return (handler.sample_rate, audio_array), message
# Save to file
timestamp = datetime.now().strftime("%Y%m%d_%H%M%S")
filename = f"indextts2_output_{timestamp}.{audio_format}"
output_path = os.path.join("outputs", filename)
# Ensure outputs directory exists
os.makedirs("outputs", exist_ok=True)
# Save audio file
if audio_format.lower() == "wav":
sf.write(output_path, audio_array, handler.sample_rate)
elif audio_format.lower() == "mp3":
# Convert to MP3 using pydub
try:
from pydub import AudioSegment
import tempfile
# Create temporary WAV file
temp_wav = tempfile.NamedTemporaryFile(suffix=".wav", delete=False)
temp_wav.close()
try:
# Save as high-quality WAV first
sf.write(temp_wav.name, audio_array, handler.sample_rate)
# Convert WAV to MP3 with high quality settings
audio_segment = AudioSegment.from_wav(temp_wav.name)
# Export with high quality settings
audio_segment.export(
output_path,
format="mp3",
bitrate="320k", # High quality
parameters=["-q:a", "0"] # Highest quality
)
finally:
# Clean up temporary file
try:
os.unlink(temp_wav.name)
except:
pass
except ImportError:
# Fallback to WAV if pydub not available
print("β οΈ Warning: pydub not available for MP3 conversion, saving as WAV instead")
wav_filename = f"indextts2_output_{timestamp}.wav"
wav_output_path = os.path.join("outputs", wav_filename)
sf.write(wav_output_path, audio_array, handler.sample_rate)
output_path = wav_output_path
except Exception as e:
print(f"β οΈ MP3 conversion failed: {e}, saving as WAV instead")
wav_filename = f"indextts2_output_{timestamp}.wav"
wav_output_path = os.path.join("outputs", wav_filename)
sf.write(wav_output_path, audio_array, handler.sample_rate)
output_path = wav_output_path
else:
# For other formats, try soundfile (may not work for all formats)
try:
sf.write(output_path, audio_array, handler.sample_rate, format=audio_format)
except Exception as e:
print(f"β οΈ Format {audio_format} not supported, saving as WAV instead")
wav_filename = f"indextts2_output_{timestamp}.wav"
wav_output_path = os.path.join("outputs", wav_filename)
sf.write(wav_output_path, audio_array, handler.sample_rate)
output_path = wav_output_path
# Calculate duration and create enhanced status message
duration = len(audio_array) / handler.sample_rate
filename = os.path.basename(output_path)
status_message = f"β
IndexTTS2 synthesis completed\n"
status_message += f"π Saved as: {filename}\n"
status_message += f"β±οΈ Duration: {duration:.2f}s\n"
status_message += f"π Sample Rate: {handler.sample_rate}Hz"
return output_path, status_message
except Exception as e:
import traceback
traceback.print_exc()
return None, f"β Error in IndexTTS2 generation: {str(e)}"
def get_indextts2_status():
"""Get IndexTTS2 model status"""
if not INDEXTTS2_AVAILABLE:
return "β IndexTTS2 not available"
handler = get_indextts2_handler()
if handler.is_model_loaded():
return "β
IndexTTS2 model loaded and ready"
else:
return "β οΈ IndexTTS2 model not loaded"
# Emotion presets for easy use
EMOTION_PRESETS = {
'neutral': {'happy': 0.0, 'angry': 0.0, 'sad': 0.0, 'afraid': 0.0, 'disgusted': 0.0, 'melancholic': 0.0, 'surprised': 0.0, 'calm': 1.0},
'happy': {'happy': 1.0, 'angry': 0.0, 'sad': 0.0, 'afraid': 0.0, 'disgusted': 0.0, 'melancholic': 0.0, 'surprised': 0.2, 'calm': 0.0},
'sad': {'happy': 0.0, 'angry': 0.0, 'sad': 1.0, 'afraid': 0.0, 'disgusted': 0.0, 'melancholic': 0.8, 'surprised': 0.0, 'calm': 0.0},
'angry': {'happy': 0.0, 'angry': 1.0, 'sad': 0.0, 'afraid': 0.0, 'disgusted': 0.3, 'melancholic': 0.0, 'surprised': 0.0, 'calm': 0.0},
'excited': {'happy': 0.8, 'angry': 0.0, 'sad': 0.0, 'afraid': 0.0, 'disgusted': 0.0, 'melancholic': 0.0, 'surprised': 0.6, 'calm': 0.0},
'melancholic': {'happy': 0.0, 'angry': 0.0, 'sad': 0.6, 'afraid': 0.0, 'disgusted': 0.0, 'melancholic': 1.0, 'surprised': 0.0, 'calm': 0.2},
'surprised': {'happy': 0.2, 'angry': 0.0, 'sad': 0.0, 'afraid': 0.3, 'disgusted': 0.0, 'melancholic': 0.0, 'surprised': 1.0, 'calm': 0.0},
'afraid': {'happy': 0.0, 'angry': 0.0, 'sad': 0.3, 'afraid': 1.0, 'disgusted': 0.0, 'melancholic': 0.0, 'surprised': 0.4, 'calm': 0.0}
} |