Spaces:
Sleeping
Sleeping
File size: 8,294 Bytes
3dba9d4 b4f1e5a 3dba9d4 d864fc1 d44c6d5 3dba9d4 b4f1e5a 3dba9d4 d864fc1 3dba9d4 b4f1e5a d44c6d5 b4f1e5a 3dba9d4 d864fc1 b4f1e5a 3dba9d4 b4f1e5a 27e6d88 909dbdf 27e6d88 b4f1e5a 27e6d88 b4f1e5a 27e6d88 3dba9d4 27e6d88 b4f1e5a 27e6d88 3dba9d4 27e6d88 d864fc1 27e6d88 b4f1e5a 27e6d88 b4f1e5a 27e6d88 3dba9d4 27e6d88 b4f1e5a 27e6d88 3dba9d4 27e6d88 |
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 |
import gradio as gr
import numpy as np
import os
import time
import torch
from scipy.io import wavfile
import soundfile as sf
import datasets
# Bark imports
from bark import generate_audio, SAMPLE_RATE
from bark.generation import preload_models, generate_text_semantic
# Hugging Face Transformers
from transformers import (
SpeechT5HifiGan,
SpeechT5ForTextToSpeech,
SpeechT5Processor
)
class VoiceSynthesizer:
def __init__(self):
# Create working directory
self.base_dir = os.path.dirname(os.path.abspath(__file__))
self.working_dir = os.path.join(self.base_dir, "working_files")
os.makedirs(self.working_dir, exist_ok=True)
# Store reference voice
self.reference_voice = None
# Initialize models dictionary
self.models = {
"bark": self._initialize_bark,
"speecht5": self._initialize_speecht5
}
# Default model
self.current_model = "bark"
# Initialize Bark models
try:
print("Attempting to load Bark models...")
preload_models()
print("Bark models loaded successfully.")
except Exception as e:
print(f"Bark model loading error: {e}")
def _initialize_bark(self):
"""Bark model initialization (already done in __init__)"""
return None
def _initialize_speecht5(self):
"""Initialize SpeechT5 model from Hugging Face"""
try:
# Load SpeechT5 model and processor
model = SpeechT5ForTextToSpeech.from_pretrained("microsoft/speecht5_tts")
processor = SpeechT5Processor.from_pretrained("microsoft/speecht5_tts")
vocoder = SpeechT5HifiGan.from_pretrained("microsoft/speecht5_hifigan")
# Load speaker embeddings
embeddings_dataset = datasets.load_dataset("Matthijs/cmu-arctic-xvectors", split="validation")
speaker_embeddings = torch.tensor(embeddings_dataset[0]["xvector"]).unsqueeze(0)
return {
"model": model,
"processor": processor,
"vocoder": vocoder,
"speaker_embeddings": speaker_embeddings
}
except Exception as e:
print(f"SpeechT5 model loading error: {e}")
return None
def process_reference_audio(self, reference_audio):
"""Process and store reference audio for voice cloning"""
try:
# Gradio can pass audio in different formats
if reference_audio is None:
return "No audio provided"
# Handle different input types
if isinstance(reference_audio, tuple):
# Gradio typically returns (sample_rate, audio_array)
if len(reference_audio) == 2:
sample_rate, audio_data = reference_audio
else:
audio_data = reference_audio[0]
sample_rate = SAMPLE_RATE # Default to Bark sample rate
elif isinstance(reference_audio, np.ndarray):
audio_data = reference_audio
sample_rate = SAMPLE_RATE
else:
return "Invalid audio format"
# Ensure audio is numpy array
audio_data = np.asarray(audio_data)
# Handle multi-channel audio
if audio_data.ndim > 1:
audio_data = audio_data.mean(axis=1)
# Trim or pad to standard length
max_duration = 10 # 10 seconds
max_samples = max_duration * sample_rate
if len(audio_data) > max_samples:
audio_data = audio_data[:max_samples]
# Resample if necessary
if sample_rate != SAMPLE_RATE:
from scipy.signal import resample
audio_data = resample(audio_data, int(len(audio_data) * SAMPLE_RATE / sample_rate))
# Save reference audio
ref_filename = os.path.join(self.working_dir, "reference_voice.wav")
sf.write(ref_filename, audio_data, SAMPLE_RATE)
# Store reference voice
self.reference_voice = ref_filename
return "Reference voice processed successfully"
except Exception as e:
print(f"Reference audio processing error: {e}")
import traceback
traceback.print_exc()
return f"Error processing reference audio: {str(e)}"
def _generate_bark_speech(self, text, voice_preset=None):
"""Generate speech using Bark"""
# Default Bark voice presets
voice_presets = [
"v2/en_speaker_6", # Female
"v2/en_speaker_3", # Male
"v2/en_speaker_9", # Neutral
]
# Prepare history prompt
history_prompt = None
# Check if a reference voice is available
if self.reference_voice is not None:
# Use saved reference voice file
history_prompt = self.reference_voice
elif voice_preset:
# Use predefined voice preset
history_prompt = voice_presets[0] if "v2/en_speaker" not in voice_preset else voice_preset
# Generate audio with or without history prompt
try:
if history_prompt:
audio_array = generate_audio(
text,
history_prompt=history_prompt
)
else:
# Fallback to default generation
audio_array = generate_audio(text)
# Save generated audio
filename = f"bark_speech_{int(time.time())}.wav"
filepath = os.path.join(self.working_dir, filename)
wavfile.write(filepath, SAMPLE_RATE, audio_array)
return filepath, None
except Exception as e:
print(f"Bark speech generation error: {e}")
import traceback
traceback.print_exc()
return None, f"Error in Bark speech generation: {str(e)}"
def generate_speech(self, text, model_name=None, voice_preset=None):
"""Generate speech using selected model"""
if not text or not text.strip():
return None, "Please enter some text to speak"
# Use specified model or current model
current_model = model_name or self.current_model
try:
if current_model == "bark":
return self._generate_bark_speech(text, voice_preset)
elif current_model == "speecht5":
return self._generate_speecht5_speech(text, voice_preset)
else:
raise ValueError(f"Unsupported model: {current_model}")
except Exception as e:
print(f"Speech generation error: {e}")
import traceback
traceback.print_exc()
return None, f"Error generating speech: {str(e)}"
def _generate_speecht5_speech(self, text, speaker_id=None):
"""Generate speech using SpeechT5"""
# Ensure model is initialized
speecht5_models = self.models["speecht5"]()
if not speecht5_models:
return None, "SpeechT5 model not loaded"
model = speecht5_models["model"]
processor = speecht5_models["processor"]
vocoder = speecht5_models["vocoder"]
speaker_embeddings = speecht5_models["speaker_embeddings"]
# Prepare inputs
inputs = processor(text=text, return_tensors="pt")
# Generate speech
speech = model.generate_speech(
inputs["input_ids"],
speaker_embeddings
)
# Convert to numpy array
audio_array = speech.numpy()
# Save generated audio
filename = f"speecht5_speech_{int(time.time())}.wav"
filepath = os.path.join(self.working_dir, filename)
wavfile.write(filepath, 16000, audio_array)
return filepath, None
# Rest of the code remains the same... |