Spaces:
Sleeping
Sleeping
File size: 33,777 Bytes
ac07929 ea401bc ac07929 | 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 | """
Sophisticated Voice Processor with ASR, TTS, and Audio Intelligence
Handles multi-language voice processing with emotion detection and adaptive responses
"""
import asyncio
import io
import numpy as np
import librosa
import soundfile as sf
import webrtcvad
from typing import Dict, List, Optional, Any, Union
from dataclasses import dataclass, field, asdict
import logging
from pathlib import Path
import tempfile
import hashlib
import torch
import torch.nn.functional as F
from transformers import AutoProcessor, AutoModelForSpeechSeq2Seq, VitsModel
from pydub import AudioSegment
from pydub.effects import normalize, compress_dynamic_range
import scipy.signal
import structlog
from caching_system import CacheManager
from production_components import (
ProductionModelOrchestrator,
ProductionMonitoringEngine,
)
# Initialize logging
logger = structlog.get_logger(__name__)
@dataclass
class AudioConfig:
"""Configuration for audio processing"""
sample_rate: int = 16000
channels: int = 1
bit_depth: int = 16
chunk_size: int = 1024
max_duration: int = 300 # 5 minutes
min_duration: float = 0.5 # 0.5 seconds
noise_reduction: bool = True
voice_activity_detection: bool = True
audio_enhancement: bool = True
emotion_detection: bool = True
speaker_identification: bool = True
@dataclass
class VoiceAnalysis:
"""Comprehensive voice analysis results"""
transcription: str
confidence_score: float
detected_language: str
language_confidence: float
emotion_analysis: Dict[str, float] = field(default_factory=dict)
speaker_characteristics: Dict[str, Any] = field(default_factory=dict)
audio_quality: Dict[str, float] = field(default_factory=dict)
speech_rate: float = 0.0
pause_analysis: Dict[str, Any] = field(default_factory=dict)
pronunciation_assessment: Dict[str, float] = field(default_factory=dict)
@dataclass
class TTSConfig:
"""Configuration for text-to-speech synthesis"""
speaking_rate: float = 1.0
pitch_shift: float = 0.0
emotion_style: str = "neutral"
language_specific_models: Dict[str, str] = field(default_factory=dict)
class ASREngine:
"""Automatic Speech Recognition with multi-model support"""
def __init__(self, config: AudioConfig):
self.config = config
self.asr_processors = {}
self.asr_models = {}
self.device = torch.device("cuda" if torch.cuda.is_available() else "cpu")
# Model configurations for different languages and use cases
self.model_configs = {
"general": "openai/whisper-small",
"rw": "mbazaNLP/Whisper-Small-Kinyarwanda",
}
async def initialize(self):
"""Initialize all ASR models"""
logger.info("Initializing ASR Engine...")
try:
for model_name, model_path in self.model_configs.items():
logger.info(
f"Loading Whisper model: {model_path} for {model_name} on {self.device}"
)
processor = AutoProcessor.from_pretrained(model_path)
model = AutoModelForSpeechSeq2Seq.from_pretrained(model_path).to(
self.device
)
self.asr_processors[model_name] = processor
self.asr_models[model_name] = model
logger.info(f"{model_name} model loaded.")
logger.info("ASR Engine initialized successfully")
except Exception as e:
logger.error(f"Failed to initialize ASR Engine: {e}", exc_info=True)
raise
async def transcribe_audio(
self, audio_data: bytes, expected_language: str = "auto"
) -> VoiceAnalysis:
"""Comprehensive audio transcription with analysis"""
try:
# Preprocess audio
processed_audio = await self._preprocess_audio(audio_data)
# Detect language if not specified or fallback
detected_language = expected_language
if expected_language == "auto":
detected_language = await self._detect_language(processed_audio)
language_confidence = 0.8 # Default
if expected_language == "auto":
# Language confidence from detection
language_confidence = await self._get_language_confidence(
processed_audio, detected_language
)
# Select best model for language
best_model_name = "rw" if detected_language == "rw" else "general"
# Perform transcription
transcription_result = await self._transcribe_with_model(
processed_audio, best_model_name, detected_language
)
# Analyze audio characteristics
audio_analysis = await self._analyze_audio_characteristics(processed_audio)
# Combine results
voice_analysis = VoiceAnalysis(
transcription=transcription_result["text"],
confidence_score=transcription_result["confidence"],
detected_language=detected_language,
language_confidence=language_confidence,
emotion_analysis=audio_analysis.get("emotions", {}),
speaker_characteristics=audio_analysis.get("speaker", {}),
audio_quality=audio_analysis.get("quality", {}),
speech_rate=audio_analysis.get("speech_rate", 0.0),
pause_analysis=audio_analysis.get("pauses", {}),
pronunciation_assessment=audio_analysis.get("pronunciation", {}),
)
return voice_analysis
except Exception as e:
logger.error(f"Error in audio transcription: {e}", exc_info=True)
raise
async def _preprocess_audio(self, audio_data: bytes) -> np.ndarray:
"""audio preprocessing pipeline"""
# Convert bytes to audio array
import noisereduce as nr
audio_segment = AudioSegment.from_file(io.BytesIO(audio_data))
# Convert to mono and resample
audio_segment = audio_segment.set_channels(1).set_frame_rate(
self.config.sample_rate
)
# Convert to numpy array
audio_array = np.array(audio_segment.get_array_of_samples(), dtype=np.float32)
audio_array = audio_array / np.max(np.abs(audio_array)) # Normalize to -1 to 1
# Apply noise reduction
if self.config.noise_reduction:
audio_array = nr.reduce_noise(y=audio_array, sr=self.config.sample_rate)
# Apply audio enhancement
if self.config.audio_enhancement:
audio_array = self._enhance_audio(audio_array)
# Voice activity detection
if self.config.voice_activity_detection:
audio_array = self._apply_vad(audio_array)
return audio_array
def _enhance_audio(self, audio: np.ndarray) -> np.ndarray:
"""Apply audio enhancement techniques."""
processed_audio = np.asarray(audio, dtype=np.float32).flatten()
sos = scipy.signal.butter(
5, 80, btype="high", fs=self.config.sample_rate, output="sos"
)
filtered_audio_output = scipy.signal.sosfilt(sos, processed_audio)
filtered_audio: np.ndarray
if isinstance(filtered_audio_output, tuple):
filtered_audio = filtered_audio_output[0]
else:
filtered_audio = filtered_audio_output
if filtered_audio.size == 0 or np.all(filtered_audio == 0):
logger.warning(
"Audio array became silent or empty after filtering. Skipping further enhancement."
)
return filtered_audio
audio_segment = AudioSegment(
data=filtered_audio.tobytes(),
frame_rate=self.config.sample_rate,
sample_width=filtered_audio.dtype.itemsize,
channels=1,
)
compressed_segment = compress_dynamic_range(
audio_segment, threshold=-20.0, ratio=4.0
)
enhanced_audio = np.array(
compressed_segment.get_array_of_samples(), dtype=np.float32
)
max_val = np.max(np.abs(enhanced_audio))
if max_val > 0:
enhanced_audio = enhanced_audio / max_val
return enhanced_audio
def _apply_vad(self, audio: np.ndarray) -> np.ndarray:
"""Apply Voice Activity Detection to remove silence"""
audio_16bit = (audio * 32767).astype(np.int16)
vad = webrtcvad.Vad(2)
frame_duration_ms = 30
frame_size = int(self.config.sample_rate * frame_duration_ms / 1000)
if self.config.sample_rate == 16000:
frame_size = 480
voiced_frames_list = []
for i in range(0, len(audio_16bit) - frame_size + 1, frame_size):
frame = audio_16bit[i : i + frame_size]
if len(frame) == frame_size:
is_speech = vad.is_speech(frame.tobytes(), self.config.sample_rate)
if is_speech:
voiced_frames_list.extend(frame)
if voiced_frames_list:
return np.array(voiced_frames_list, dtype=np.float32) / 32767.0
else:
return audio
async def _detect_language(self, audio: np.ndarray) -> str:
"""Detect language from audio using general model"""
try:
processor = self.asr_processors["general"]
model = self.asr_models["general"]
input_features = processor(
audio, sampling_rate=self.config.sample_rate, return_tensors="pt"
).input_features.to(self.device)
predicted = model.generate(
input_features,
max_new_tokens=1,
output_scores=True,
return_dict_in_generate=True,
)
lang_token = predicted.sequences[0, 1]
lang = processor.decode(lang_token)
# Map to our supported languages
language_mapping = {
"en": "en",
"fr": "fr",
"rw": "rw",
"sw": "rw", # Fallback Swahili to Kinyarwanda
}
return language_mapping.get(lang, "en")
except Exception as e:
logger.error(f"Language detection failed: {e}", exc_info=True)
return "en"
async def _get_language_confidence(
self, audio: np.ndarray, detected_language: str
) -> float:
"""Get confidence for detected language"""
try:
processor = self.asr_processors["general"]
model = self.asr_models["general"]
input_features = processor(
audio, sampling_rate=self.config.sample_rate, return_tensors="pt"
).input_features.to(self.device)
predicted = model.generate(
input_features,
max_new_tokens=1,
output_scores=True,
return_dict_in_generate=True,
)
prob = F.softmax(predicted.scores[0], dim=-1)[
0, predicted.sequences[0, 1]
].item()
return float(prob)
except Exception as e:
logger.error(f"Language confidence calculation failed: {e}", exc_info=True)
return 0.8
def _select_best_model(self, language: str) -> str:
"""Select the best ASR model for the detected language"""
if language == "rw":
return "rw"
return "general"
async def _transcribe_with_model(
self, audio: np.ndarray, model_name: str, language: str
) -> Dict[str, Any]:
"""Transcribe audio using specified model"""
try:
processor = self.asr_processors[model_name]
model = self.asr_models[model_name]
lang_code = "rw" if model_name == "rw" else language
forced_decoder_ids = processor.get_decoder_prompt_ids(
language=lang_code, task="transcribe"
)
input_features = processor(
audio, sampling_rate=self.config.sample_rate, return_tensors="pt"
).input_features.to(self.device)
predicted = model.generate(
input_features,
forced_decoder_ids=forced_decoder_ids,
output_scores=True,
return_dict_in_generate=True,
)
transcription = processor.batch_decode(
predicted.sequences, skip_special_tokens=True
)[0]
# Calculate confidence
scores = predicted.scores
log_probs = [torch.log_softmax(score, dim=-1) for score in scores]
seq_len = predicted.sequences.shape[1]
forced_len = len(forced_decoder_ids) if forced_decoder_ids else 0
token_log_probs = [
log_prob[0, predicted.sequences[0, i].item()]
for i, log_prob in enumerate(log_probs, start=forced_len)
if i >= forced_len
]
avg_logprob = (
sum(token_log_probs) / len(token_log_probs) if token_log_probs else 0
)
confidence = float(np.exp(avg_logprob))
return {
"text": transcription.strip(),
"confidence": confidence,
"language_confidence": 0.8,
}
except Exception as e:
logger.error(f"Transcription failed for {model_name}: {e}", exc_info=True)
raise
async def _analyze_audio_characteristics(self, audio: np.ndarray) -> Dict[str, Any]:
"""Analyze various audio characteristics"""
analysis = {}
try:
if self.config.emotion_detection:
analysis["emotions"] = await self._analyze_emotions(audio)
analysis["speech_rate"] = self._calculate_speech_rate(audio)
analysis["quality"] = self._assess_audio_quality(audio)
analysis["pauses"] = self._analyze_pauses(audio)
if self.config.speaker_identification:
analysis["speaker"] = await self._analyze_speaker_characteristics(audio)
except Exception as e:
logger.error(f"Error in audio characteristics analysis: {e}", exc_info=True)
return analysis
async def _analyze_emotions(self, audio: np.ndarray) -> Dict[str, float]:
"""Analyze emotional content of speech"""
try:
emotions = {
"neutral": 0.4,
"happy": 0.2,
"sad": 0.1,
"angry": 0.1,
"fear": 0.1,
"surprise": 0.1,
}
mfccs = librosa.feature.mfcc(y=audio, sr=self.config.sample_rate, n_mfcc=13)
energy = float(np.mean(librosa.feature.rms(y=audio)))
spectral_centroid = float(
np.mean(
librosa.feature.spectral_centroid(
y=audio, sr=self.config.sample_rate
)
)
)
if energy > 0.1 and spectral_centroid > 2000:
emotions["happy"] += 0.3
emotions["neutral"] -= 0.2
elif energy < 0.05:
emotions["sad"] += 0.2
emotions["neutral"] -= 0.1
total = sum(emotions.values())
if total > 0:
emotions = {k: v / total for k, v in emotions.items()}
return emotions
except Exception as e:
logger.error(f"Emotion analysis failed: {e}", exc_info=True)
return {"neutral": 1.0}
def _calculate_speech_rate(self, audio: np.ndarray) -> float:
"""Calculate speech rate (words per minute)"""
try:
duration = len(audio) / self.config.sample_rate
intervals = librosa.effects.split(audio, top_db=20)
speech_duration = sum(
(end - start) / self.config.sample_rate for start, end in intervals
)
if speech_duration > 0:
estimated_words = speech_duration * 2.5 # Rough syllable rate
speech_rate = (
estimated_words / speech_duration
) * 60 # Words per minute
return float(min(speech_rate, 300))
return 0.0
except Exception as e:
logger.error(f"Speech rate calculation failed: {e}", exc_info=True)
return 0.0
def _assess_audio_quality(self, audio: np.ndarray) -> Dict[str, float]:
"""Assess audio quality metrics"""
try:
quality = {}
signal_power = float(np.mean(audio**2))
noise_power = float(np.mean((audio - np.mean(audio)) ** 2) * 0.1)
snr = 10 * np.log10(signal_power / max(noise_power, 1e-10))
quality["snr"] = float(max(0.0, min(1.0, snr / 30)))
dynamic_range = float(np.max(audio) - np.min(audio))
quality["dynamic_range"] = float(min(1.0, dynamic_range))
fft = np.fft.fft(audio)
spectral_energy = float(np.mean(np.abs(fft)))
quality["spectral_quality"] = float(min(1.0, spectral_energy * 10))
quality["overall"] = float(np.mean(list(quality.values())))
return quality
except Exception as e:
logger.error(f"Audio quality assessment failed: {e}", exc_info=True)
return {"overall": 0.5}
def _analyze_pauses(self, audio: np.ndarray) -> Dict[str, Any]:
"""Analyze pause patterns in speech"""
try:
intervals = librosa.effects.split(audio, top_db=20)
if len(intervals) < 2:
return {
"pause_count": 0,
"average_pause_duration": 0.0,
"total_pause_time": 0.0,
"pause_distribution": {},
}
pause_durations = []
for i in range(len(intervals) - 1):
pause_start = intervals[i][1]
pause_end = intervals[i + 1][0]
pause_duration = (pause_end - pause_start) / self.config.sample_rate
if pause_duration > 0.1:
pause_durations.append(pause_duration)
return {
"pause_count": len(pause_durations),
"average_pause_duration": (
float(np.mean(pause_durations)) if pause_durations else 0.0
),
"total_pause_time": float(sum(pause_durations)),
"pause_distribution": {
"short": sum(1 for p in pause_durations if p < 0.5),
"medium": sum(1 for p in pause_durations if 0.5 <= p < 1.0),
"long": sum(1 for p in pause_durations if p >= 1.0),
},
}
except Exception as e:
logger.error(f"Pause analysis failed: {e}", exc_info=True)
return {
"pause_count": 0,
"average_pause_duration": 0.0,
"total_pause_time": 0.0,
"pause_distribution": {},
}
async def _analyze_speaker_characteristics(
self, audio: np.ndarray
) -> Dict[str, Any]:
"""Analyze speaker characteristics"""
try:
characteristics = {}
f0 = librosa.yin(audio, fmin=50, fmax=400, sr=self.config.sample_rate)
f0_clean = f0[f0 > 0]
if len(f0_clean) > 0:
characteristics["average_pitch"] = float(np.mean(f0_clean))
characteristics["pitch_range"] = float(
np.max(f0_clean) - np.min(f0_clean)
)
characteristics["pitch_variance"] = float(np.var(f0_clean))
mfccs = librosa.feature.mfcc(y=audio, sr=self.config.sample_rate, n_mfcc=13)
characteristics["vocal_tract_length"] = (
float(np.mean(mfccs[1:4])) if mfccs.size > 0 else 0.0
)
spectral_centroid = librosa.feature.spectral_centroid(
y=audio, sr=self.config.sample_rate
)
characteristics["brightness"] = (
float(np.mean(spectral_centroid)) if spectral_centroid.size > 0 else 0.0
)
zero_crossing_rate = librosa.feature.zero_crossing_rate(audio)
characteristics["roughness"] = (
float(np.mean(zero_crossing_rate))
if zero_crossing_rate.size > 0
else 0.0
)
return characteristics
except Exception as e:
logger.error(f"Speaker characteristics analysis failed: {e}", exc_info=True)
return {}
class TTSEngine:
"""Text-to-Speech with emotion and voice cloning"""
def __init__(self, config: TTSConfig):
self.config = config
self.tts_processors = {}
self.tts_models = {}
self.device = torch.device("cuda" if torch.cuda.is_available() else "cpu")
# TTS model configurations
self.model_configs = {
"en": "facebook/mms-tts-eng",
"fr": "facebook/mms-tts-fra",
"rw": "facebook/mms-tts-kin",
}
async def initialize(self):
"""Initialize TTS models"""
logger.info("Initializing TTS Engine...")
try:
for lang, model_path in self.model_configs.items():
logger.info(
f"Loading MMS-TTS model: {model_path} for {lang} on {self.device}"
)
processor = AutoProcessor.from_pretrained(model_path)
model = VitsModel.from_pretrained(model_path)
if hasattr(model, "to"):
model = model.to(self.device) # type: ignore
self.tts_processors[lang] = processor
self.tts_models[lang] = model
logger.info(f"{lang} TTS model loaded.")
logger.info("TTS Engine initialized successfully")
except Exception as e:
logger.error(f"Failed to initialize TTS Engine: {e}", exc_info=True)
raise
async def synthesize_speech(
self,
text: str,
language: str = "en",
emotion_context: Optional[Dict[str, float]] = None,
) -> bytes:
"""Synthesize speech with advanced features"""
try:
# Select appropriate model
model_key = language if language in self.tts_models else "en"
# Preprocess text
processed_text = await self._preprocess_text(text, language)
# Generate speech
audio = await self._mms_tts_synthesize(processed_text, model_key)
# Post-process audio
enhanced_audio = await self._postprocess_audio(audio, emotion_context)
# Convert to bytes
audio_bytes = self._audio_to_bytes(enhanced_audio)
return audio_bytes
except Exception as e:
logger.error(f"Speech synthesis failed: {e}", exc_info=True)
raise
async def _preprocess_text(self, text: str, language: str) -> str:
"""Preprocess text for better TTS output"""
processed_text = text.strip()
if language == "rw":
processed_text = self._normalize_kinyarwanda_text(processed_text)
elif language == "en":
processed_text = self._normalize_english_text(processed_text)
elif language == "fr":
processed_text = self._normalize_french_text(processed_text)
processed_text = self._add_prosodic_markers(processed_text)
return processed_text
def _normalize_kinyarwanda_text(self, text: str) -> str:
"""Normalize Kinyarwanda text for TTS (Placeholder)"""
return text
def _normalize_english_text(self, text: str) -> str:
"""Normalize English text for TTS"""
import re
abbreviations = {
"Dr.": "Doctor",
"Mr.": "Mister",
"Mrs.": "Missus",
"Ms.": "Miss",
"Prof.": "Professor",
"etc.": "etcetera",
"vs.": "versus",
}
for abbr, expansion in abbreviations.items():
text = text.replace(abbr, expansion)
text = re.sub(
r"\b(\d+)\b", lambda m: self._number_to_words(int(m.group(1))), text
)
return text
def _normalize_french_text(self, text: str) -> str:
"""Normalize French text for TTS (Placeholder)"""
return text
def _number_to_words(self, number: int) -> str:
"""Convert numbers to words (simplified placeholder)"""
if number == 0:
return "zero"
if number == 1:
return "one"
if number == 2:
return "two"
return str(number)
def _add_prosodic_markers(self, text: str) -> str:
"""Add prosodic markers for better speech synthesis"""
text = text.replace(".", ". <break time='0.5s'/>")
text = text.replace(",", ", <break time='0.2s'/>")
text = text.replace("?", "? <break time='0.7s'/>")
text = text.replace("!", "! <break time='0.7s'/>")
return text
async def _mms_tts_synthesize(self, text: str, model_key: str) -> np.ndarray:
"""Synthesize using MMS-TTS model"""
try:
processor = self.tts_processors[model_key]
model = self.tts_models[model_key]
inputs = processor(text=text, return_tensors="pt").to(self.device)
with torch.no_grad():
outputs = model(**inputs)
audio = outputs.waveform[0].cpu().numpy()
return audio
except Exception as e:
logger.error(f"MMS-TTS synthesis failed: {e}", exc_info=True)
raise
async def _postprocess_audio(
self, audio: np.ndarray, emotion_context: Optional[Dict[str, float]] = None
) -> np.ndarray:
"""Post-process synthesized audio"""
if emotion_context:
audio = self._apply_emotional_processing(audio, emotion_context)
audio = self._enhance_synthesized_audio(audio)
audio = audio / np.max(np.abs(audio)) # Final normalization
return audio
def _apply_emotional_processing(
self, audio: np.ndarray, emotion_context: Dict[str, float]
) -> np.ndarray:
"""Apply emotion-based audio processing"""
dominant_emotion_item = max(emotion_context.items(), key=lambda x: x[1])
emotion, intensity = dominant_emotion_item
sr_for_pitch_shift = 16000
if emotion == "happy" and intensity > 0.5:
audio = librosa.effects.pitch_shift(audio, sr=sr_for_pitch_shift, n_steps=1)
elif emotion == "sad" and intensity > 0.5:
audio = librosa.effects.pitch_shift(
audio, sr=sr_for_pitch_shift, n_steps=-1
)
elif emotion == "angry" and intensity > 0.5:
audio = np.tanh(audio * 1.2) # Simple distortion
return audio
def _enhance_synthesized_audio(self, audio: np.ndarray) -> np.ndarray:
"""Enhance synthesized audio quality"""
audio_segment = AudioSegment(
audio.tobytes(),
frame_rate=16000,
sample_width=audio.dtype.itemsize,
channels=1,
)
normalized = normalize(audio_segment)
compressed = compress_dynamic_range(normalized, threshold=-15.0, ratio=2.0)
enhanced_audio = np.array(compressed.get_array_of_samples(), dtype=np.float32)
enhanced_audio = enhanced_audio / np.max(np.abs(enhanced_audio))
return enhanced_audio
def _audio_to_bytes(self, audio: np.ndarray, format: str = "mp3") -> bytes:
"""Convert audio array to bytes"""
audio_16bit = (audio * 32767).astype(np.int16)
audio_segment = AudioSegment(
audio_16bit.tobytes(), frame_rate=16000, sample_width=2, channels=1
)
buffer = io.BytesIO()
audio_segment.export(buffer, format=format)
return buffer.getvalue()
class VoiceProcessor:
"""Main voice processor orchestrating ASR and TTS"""
def __init__(
self,
cache_manager: CacheManager,
model_orchestrator: ProductionModelOrchestrator,
monitoring_engine: ProductionMonitoringEngine,
):
self.cache_manager = cache_manager
self.model_orchestrator = model_orchestrator
self.monitoring_engine = monitoring_engine
self.audio_config = AudioConfig()
self.tts_config = TTSConfig()
self.asr_engine = ASREngine(self.audio_config)
self.tts_engine = TTSEngine(self.tts_config)
self.is_initialized = False
async def initialize(self):
"""Initialize the sophisticated voice processor"""
logger.info("Initializing Sophisticated Voice Processor...")
try:
# Initialize ASR engine
await self.asr_engine.initialize()
# Initialize TTS engine
await self.tts_engine.initialize()
self.is_initialized = True
logger.info("Sophisticated Voice Processor initialized successfully")
except Exception as e:
logger.error(
f"Failed to initialize Sophisticated Voice Processor: {e}",
exc_info=True,
)
raise
async def enhanced_transcribe_audio(
self,
audio_data: bytes,
expected_language: str = "auto",
session_id: Optional[str] = None,
) -> VoiceAnalysis:
"""Enhanced audio transcription with comprehensive analysis"""
if not self.is_initialized:
raise RuntimeError("Voice Processor not initialized")
try:
# Check cache first
audio_hash = hashlib.md5(audio_data).hexdigest()
cached_result = await self.cache_manager.get_transcription_cache(audio_hash)
if cached_result:
logger.info("Returning cached transcription.", audio_hash=audio_hash, session_id=session_id)
return VoiceAnalysis(**cached_result)
# Perform transcription
voice_analysis = await self.asr_engine.transcribe_audio(audio_data, expected_language)
# Prepare result
result_dict = asdict(voice_analysis)
# Cache result
await self.cache_manager.cache_transcription(audio_hash, result_dict)
# Log metrics
await self.monitoring_engine.log_query_metrics(
{
"session_id": session_id,
"transcription_confidence": voice_analysis.confidence_score,
"detected_language": voice_analysis.detected_language,
"audio_duration": len(audio_data)
/ (
self.audio_config.sample_rate
* (self.audio_config.bit_depth / 8)
),
"audio_quality": voice_analysis.audio_quality.get("overall", 0.5),
}
)
return voice_analysis
except Exception as e:
logger.error(f"Enhanced transcription failed: {e}", exc_info=True)
raise
async def enhanced_text_to_speech(
self,
text: str,
language: str = "en",
session_context: Optional[Dict[str, Any]] = None,
emotion_context: Optional[Dict[str, float]] = None,
) -> bytes:
"""Enhanced text-to-speech with contextual adaptation"""
if not self.is_initialized:
raise RuntimeError("Voice Processor not initialized")
try:
# Check cache first
emotion_key = (
"_".join(f"{k}{v:.2f}" for k, v in sorted(emotion_context.items()))
if emotion_context
else "neutral"
)
tts_key = f"{text}_{language}_{emotion_key}"
tts_hash = hashlib.md5(tts_key.encode()).hexdigest()
cached_audio = await self.cache_manager.get_tts_cache(tts_hash)
if cached_audio:
return cached_audio
# Synthesize speech
audio_bytes = await self.tts_engine.synthesize_speech(
text=text, language=language, emotion_context=emotion_context
)
# Cache result
await self.cache_manager.cache_tts(tts_hash, audio_bytes)
return audio_bytes
except Exception as e:
logger.error(f"Enhanced TTS failed: {e}", exc_info=True)
raise
async def health_check(self) -> bool:
"""Check health of voice processor components"""
try:
if not self.is_initialized:
logger.warning("Voice Processor is not initialized.")
return False
# Check ASR models
if not self.asr_engine.asr_models:
logger.error("ASR models not loaded.")
return False
# Check TTS models
if not self.tts_engine.tts_models:
logger.error("TTS models not loaded.")
return False
# Check dependencies health
if not await self.cache_manager.health_check():
return False
if not await self.model_orchestrator.health_check():
return False
if not await self.monitoring_engine.health_check():
return False
logger.info("Voice Processor health check passed.")
return True
except Exception as e:
logger.error(f"Voice processor health check failed: {e}", exc_info=True)
return False
|