Spaces:
Sleeping
Sleeping
Update app.py
Browse files
app.py
CHANGED
|
@@ -1,5 +1,5 @@
|
|
| 1 |
import os
|
| 2 |
-
# --- 1. FORCE CPU (
|
| 3 |
os.environ["CUDA_VISIBLE_DEVICES"] = "-1"
|
| 4 |
|
| 5 |
import sys
|
|
@@ -9,9 +9,9 @@ import edge_tts
|
|
| 9 |
import asyncio
|
| 10 |
import shutil
|
| 11 |
import zipfile
|
| 12 |
-
|
| 13 |
|
| 14 |
-
# PyTorch
|
| 15 |
torch.cuda.is_available = lambda : False
|
| 16 |
print(f"🚀 System Mode: CPU Only (Stable V2)")
|
| 17 |
|
|
@@ -22,19 +22,33 @@ if not os.path.exists("OpenVoice"):
|
|
| 22 |
|
| 23 |
sys.path.append(os.path.abspath("OpenVoice"))
|
| 24 |
|
| 25 |
-
#
|
| 26 |
os.makedirs("checkpoints_v2", exist_ok=True)
|
| 27 |
|
| 28 |
-
# V2 Model Download Function
|
| 29 |
def download_v2_models():
|
| 30 |
-
#
|
| 31 |
if not os.path.exists("checkpoints_v2/converter/checkpoint.pth"):
|
| 32 |
-
print("Downloading OpenVoice V2 Model...")
|
| 33 |
try:
|
| 34 |
-
#
|
| 35 |
-
|
| 36 |
-
|
| 37 |
-
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
| 38 |
except Exception as e:
|
| 39 |
print(f"Download Error: {e}")
|
| 40 |
|
|
@@ -51,18 +65,17 @@ except ImportError:
|
|
| 51 |
|
| 52 |
# --- 3. Load V2 Model ---
|
| 53 |
ckpt_converter = 'checkpoints_v2/converter'
|
| 54 |
-
if not os.path.exists(f"{ckpt_converter}/config.json"):
|
| 55 |
-
# Fallback path logic
|
| 56 |
-
ckpt_converter = 'OpenVoice/checkpoints_v2/converter'
|
| 57 |
|
| 58 |
-
print("
|
| 59 |
try:
|
| 60 |
-
# V2
|
| 61 |
tone_color_converter = ToneColorConverter(f'{ckpt_converter}/config.json', device='cpu')
|
| 62 |
tone_color_converter.load_ckpt(f'{ckpt_converter}/checkpoint.pth')
|
| 63 |
print("✅ V2 Model Loaded Successfully!")
|
| 64 |
except Exception as e:
|
| 65 |
print(f"Model Load Error: {e}")
|
|
|
|
|
|
|
| 66 |
|
| 67 |
# --- 4. Mastering Engine ---
|
| 68 |
def apply_mastering(input_wav, style="Radio"):
|
|
@@ -89,6 +102,8 @@ async def run_edge_tts(text, gender):
|
|
| 89 |
return output_file
|
| 90 |
|
| 91 |
def predict(text, ref_audio, gender, mastering_style):
|
|
|
|
|
|
|
| 92 |
if not text: return "စာရိုက်ထည့်ပါ", None
|
| 93 |
if not ref_audio: return "Reference Audio ထည့်ပါ", None
|
| 94 |
|
|
@@ -99,7 +114,7 @@ def predict(text, ref_audio, gender, mastering_style):
|
|
| 99 |
# Step B: OpenVoice V2 Conversion
|
| 100 |
os.makedirs("outputs", exist_ok=True)
|
| 101 |
|
| 102 |
-
# VAD Handling
|
| 103 |
try:
|
| 104 |
target_se, _ = se_extractor.get_se(ref_audio, tone_color_converter, target_dir='outputs', vad=True)
|
| 105 |
except:
|
|
@@ -109,7 +124,6 @@ def predict(text, ref_audio, gender, mastering_style):
|
|
| 109 |
|
| 110 |
raw_output = "outputs/raw_v2.wav"
|
| 111 |
|
| 112 |
-
# Actual Conversion
|
| 113 |
tone_color_converter.convert(
|
| 114 |
audio_src_path=base_audio,
|
| 115 |
src_se=source_se,
|
|
@@ -123,12 +137,14 @@ def predict(text, ref_audio, gender, mastering_style):
|
|
| 123 |
return "Success (V2 Ready)!", final_output
|
| 124 |
|
| 125 |
except Exception as e:
|
|
|
|
|
|
|
| 126 |
return f"Error: {str(e)}", None
|
| 127 |
|
| 128 |
# UI
|
| 129 |
with gr.Blocks(title="Myanmar OpenVoice V2") as demo:
|
| 130 |
gr.Markdown("# 🇲🇲 Myanmar Voice Cloning (OpenVoice V2)")
|
| 131 |
-
gr.Markdown("V2 Model + CPU Stable
|
| 132 |
|
| 133 |
with gr.Row():
|
| 134 |
with gr.Column():
|
|
|
|
| 1 |
import os
|
| 2 |
+
# --- 1. FORCE CPU (Driver Error ကာကွယ်ရန်) ---
|
| 3 |
os.environ["CUDA_VISIBLE_DEVICES"] = "-1"
|
| 4 |
|
| 5 |
import sys
|
|
|
|
| 9 |
import asyncio
|
| 10 |
import shutil
|
| 11 |
import zipfile
|
| 12 |
+
import urllib.request
|
| 13 |
|
| 14 |
+
# PyTorch CPU Mode
|
| 15 |
torch.cuda.is_available = lambda : False
|
| 16 |
print(f"🚀 System Mode: CPU Only (Stable V2)")
|
| 17 |
|
|
|
|
| 22 |
|
| 23 |
sys.path.append(os.path.abspath("OpenVoice"))
|
| 24 |
|
| 25 |
+
# V2 Checkpoints Folder ဖန်တီးခြင်း
|
| 26 |
os.makedirs("checkpoints_v2", exist_ok=True)
|
| 27 |
|
| 28 |
+
# V2 Model Download Function (Official S3 Link)
|
| 29 |
def download_v2_models():
|
| 30 |
+
# Model ရှိမရှိ စစ်ဆေးခြင်း
|
| 31 |
if not os.path.exists("checkpoints_v2/converter/checkpoint.pth"):
|
| 32 |
+
print("Downloading OpenVoice V2 Model from AWS...")
|
| 33 |
try:
|
| 34 |
+
# 1. Zip ဖိုင်ကို ဒေါင်းလုပ်ဆွဲခြင်း
|
| 35 |
+
url = "https://myshell-public-repo-hosting.s3.amazonaws.com/openvoice/checkpoints_v2_0417.zip"
|
| 36 |
+
zip_path = "checkpoints_v2.zip"
|
| 37 |
+
urllib.request.urlretrieve(url, zip_path)
|
| 38 |
+
|
| 39 |
+
# 2. ဖြည်ချခြင်း (Unzip)
|
| 40 |
+
print("Extracting Model...")
|
| 41 |
+
with zipfile.ZipFile(zip_path, 'r') as zip_ref:
|
| 42 |
+
zip_ref.extractall(".")
|
| 43 |
+
|
| 44 |
+
# 3. Folder နာမည်ပြန်ညှိခြင်း (လိုအပ်လျှင်)
|
| 45 |
+
if os.path.exists("checkpoints_v2_0417"):
|
| 46 |
+
# အဟောင်းဖျက်ပြီး အသစ်နေရာချ
|
| 47 |
+
if os.path.exists("checkpoints_v2"):
|
| 48 |
+
shutil.rmtree("checkpoints_v2")
|
| 49 |
+
os.rename("checkpoints_v2_0417", "checkpoints_v2")
|
| 50 |
+
|
| 51 |
+
print("✅ V2 Model Ready!")
|
| 52 |
except Exception as e:
|
| 53 |
print(f"Download Error: {e}")
|
| 54 |
|
|
|
|
| 65 |
|
| 66 |
# --- 3. Load V2 Model ---
|
| 67 |
ckpt_converter = 'checkpoints_v2/converter'
|
|
|
|
|
|
|
|
|
|
| 68 |
|
| 69 |
+
print(f"Loading V2 Model from {ckpt_converter}...")
|
| 70 |
try:
|
| 71 |
+
# V2 requires 'device' argument
|
| 72 |
tone_color_converter = ToneColorConverter(f'{ckpt_converter}/config.json', device='cpu')
|
| 73 |
tone_color_converter.load_ckpt(f'{ckpt_converter}/checkpoint.pth')
|
| 74 |
print("✅ V2 Model Loaded Successfully!")
|
| 75 |
except Exception as e:
|
| 76 |
print(f"Model Load Error: {e}")
|
| 77 |
+
# Error တက်ရင် Variable ကို None ပေးထားမည် (UI မှာ Crash မဖြစ်အောင်)
|
| 78 |
+
tone_color_converter = None
|
| 79 |
|
| 80 |
# --- 4. Mastering Engine ---
|
| 81 |
def apply_mastering(input_wav, style="Radio"):
|
|
|
|
| 102 |
return output_file
|
| 103 |
|
| 104 |
def predict(text, ref_audio, gender, mastering_style):
|
| 105 |
+
if tone_color_converter is None:
|
| 106 |
+
return "Critical Error: Model not loaded. Check logs.", None
|
| 107 |
if not text: return "စာရိုက်ထည့်ပါ", None
|
| 108 |
if not ref_audio: return "Reference Audio ထည့်ပါ", None
|
| 109 |
|
|
|
|
| 114 |
# Step B: OpenVoice V2 Conversion
|
| 115 |
os.makedirs("outputs", exist_ok=True)
|
| 116 |
|
| 117 |
+
# VAD Handling
|
| 118 |
try:
|
| 119 |
target_se, _ = se_extractor.get_se(ref_audio, tone_color_converter, target_dir='outputs', vad=True)
|
| 120 |
except:
|
|
|
|
| 124 |
|
| 125 |
raw_output = "outputs/raw_v2.wav"
|
| 126 |
|
|
|
|
| 127 |
tone_color_converter.convert(
|
| 128 |
audio_src_path=base_audio,
|
| 129 |
src_se=source_se,
|
|
|
|
| 137 |
return "Success (V2 Ready)!", final_output
|
| 138 |
|
| 139 |
except Exception as e:
|
| 140 |
+
import traceback
|
| 141 |
+
traceback.print_exc()
|
| 142 |
return f"Error: {str(e)}", None
|
| 143 |
|
| 144 |
# UI
|
| 145 |
with gr.Blocks(title="Myanmar OpenVoice V2") as demo:
|
| 146 |
gr.Markdown("# 🇲🇲 Myanmar Voice Cloning (OpenVoice V2)")
|
| 147 |
+
gr.Markdown("V2 Model (AWS Source) + CPU Stable")
|
| 148 |
|
| 149 |
with gr.Row():
|
| 150 |
with gr.Column():
|