Spaces:
Running
on
Zero
Running
on
Zero
Refactor TTSModelV1 to load voice mappings from JSON and simplify voice selection
Browse files- tts_model_v1.py +10 -60
- voices/v1_voices.json +32 -0
tts_model_v1.py
CHANGED
@@ -1,4 +1,5 @@
|
|
1 |
import os
|
|
|
2 |
import torch
|
3 |
import numpy as np
|
4 |
import time
|
@@ -6,54 +7,32 @@ from typing import Tuple, List
|
|
6 |
import soundfile as sf
|
7 |
from kokoro import KPipeline
|
8 |
import spaces
|
9 |
-
from lib.file_utils import download_voice_files, ensure_dir
|
10 |
|
11 |
class TTSModelV1:
|
12 |
"""KPipeline-based TTS model for v1.0.0"""
|
13 |
|
14 |
def __init__(self):
|
15 |
self.pipeline = None
|
16 |
-
|
17 |
-
|
18 |
-
|
|
|
19 |
|
20 |
def initialize(self) -> bool:
|
21 |
-
"""Initialize KPipeline
|
22 |
try:
|
23 |
print("Initializing v1.0.0 model...")
|
24 |
-
|
25 |
self.pipeline = None # cannot be initialized outside of GPU decorator
|
26 |
-
|
27 |
-
# Download v1 voices if needed
|
28 |
-
ensure_dir(self.voices_dir)
|
29 |
-
if not os.path.exists(os.path.join(self.voices_dir, "voices")):
|
30 |
-
print("Downloading v1 voices...")
|
31 |
-
download_voice_files(self.model_repo, "voices", self.voices_dir)
|
32 |
-
|
33 |
-
# Verify voices were downloaded successfully
|
34 |
-
available_voices = self.list_voices()
|
35 |
-
if not available_voices:
|
36 |
-
print("Warning: No voices found after initialization")
|
37 |
-
else:
|
38 |
-
print(f"Found {len(available_voices)} voices")
|
39 |
-
|
40 |
print("Model initialization complete")
|
41 |
return True
|
42 |
-
|
43 |
except Exception as e:
|
44 |
print(f"Error initializing model: {str(e)}")
|
45 |
return False
|
46 |
|
47 |
def list_voices(self) -> List[str]:
|
48 |
"""List available voices"""
|
49 |
-
voices
|
50 |
-
|
51 |
-
if os.path.exists(voices_dir):
|
52 |
-
for file in os.listdir(voices_dir):
|
53 |
-
if file.endswith(".pt"):
|
54 |
-
voice_name = file[:-3]
|
55 |
-
voices.append(voice_name)
|
56 |
-
return voices
|
57 |
|
58 |
@spaces.GPU(duration=None) # Duration will be set by the UI
|
59 |
def generate_speech(self, text: str, voice_names: list[str], speed: float = 1.0, gpu_timeout: int = 60, progress_callback=None, progress_state=None, progress=None) -> Tuple[np.ndarray, float]:
|
@@ -76,35 +55,12 @@ class TTSModelV1:
|
|
76 |
if not text or not voice_names:
|
77 |
raise ValueError("Text and voice name are required")
|
78 |
|
79 |
-
# Handle voice
|
80 |
if isinstance(voice_names, list) and len(voice_names) > 1:
|
81 |
-
|
82 |
-
for voice in voice_names:
|
83 |
-
try:
|
84 |
-
voice_path = os.path.join(self.voices_dir, "voices", f"{voice}.pt")
|
85 |
-
try:
|
86 |
-
voicepack = torch.load(voice_path, weights_only=True)
|
87 |
-
except Exception as e:
|
88 |
-
print(f"Warning: weights_only load failed, attempting full load: {str(e)}")
|
89 |
-
voicepack = torch.load(voice_path, weights_only=False)
|
90 |
-
t_voices.append(voicepack)
|
91 |
-
except Exception as e:
|
92 |
-
print(f"Warning: Failed to load voice {voice}: {str(e)}")
|
93 |
-
|
94 |
-
# Combine voices by taking mean
|
95 |
-
voicepack = torch.mean(torch.stack(t_voices), dim=0)
|
96 |
voice_name = "_".join(voice_names)
|
97 |
-
# Save mixed voice temporarily
|
98 |
-
mixed_voice_path = os.path.join(self.voices_dir, "voices", f"{voice_name}.pt")
|
99 |
-
torch.save(voicepack, mixed_voice_path)
|
100 |
else:
|
101 |
voice_name = voice_names[0]
|
102 |
-
voice_path = os.path.join(self.voices_dir, "voices", f"{voice_name}.pt")
|
103 |
-
try:
|
104 |
-
voicepack = torch.load(voice_path, weights_only=True)
|
105 |
-
except Exception as e:
|
106 |
-
print(f"Warning: weights_only load failed, attempting full load: {str(e)}")
|
107 |
-
voicepack = torch.load(voice_path, weights_only=False)
|
108 |
|
109 |
# Initialize tracking
|
110 |
audio_chunks = []
|
@@ -172,12 +128,6 @@ class TTSModelV1:
|
|
172 |
# Concatenate audio chunks
|
173 |
audio = np.concatenate(audio_chunks)
|
174 |
|
175 |
-
# Cleanup temporary mixed voice if created
|
176 |
-
if len(voice_names) > 1:
|
177 |
-
try:
|
178 |
-
os.remove(mixed_voice_path)
|
179 |
-
except:
|
180 |
-
pass
|
181 |
|
182 |
# Return audio and metrics
|
183 |
return (
|
|
|
1 |
import os
|
2 |
+
import json
|
3 |
import torch
|
4 |
import numpy as np
|
5 |
import time
|
|
|
7 |
import soundfile as sf
|
8 |
from kokoro import KPipeline
|
9 |
import spaces
|
|
|
10 |
|
11 |
class TTSModelV1:
|
12 |
"""KPipeline-based TTS model for v1.0.0"""
|
13 |
|
14 |
def __init__(self):
|
15 |
self.pipeline = None
|
16 |
+
# Load v1 voice mappings
|
17 |
+
voice_map_path = os.path.join(os.path.dirname(__file__), "voices", "v1_voices.json")
|
18 |
+
with open(voice_map_path) as f:
|
19 |
+
self.voice_map = json.load(f)
|
20 |
|
21 |
def initialize(self) -> bool:
|
22 |
+
"""Initialize KPipeline"""
|
23 |
try:
|
24 |
print("Initializing v1.0.0 model...")
|
|
|
25 |
self.pipeline = None # cannot be initialized outside of GPU decorator
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
26 |
print("Model initialization complete")
|
27 |
return True
|
|
|
28 |
except Exception as e:
|
29 |
print(f"Error initializing model: {str(e)}")
|
30 |
return False
|
31 |
|
32 |
def list_voices(self) -> List[str]:
|
33 |
"""List available voices"""
|
34 |
+
# Return all voices from voice map
|
35 |
+
return self.voice_map["american"] + self.voice_map["british"]
|
|
|
|
|
|
|
|
|
|
|
|
|
36 |
|
37 |
@spaces.GPU(duration=None) # Duration will be set by the UI
|
38 |
def generate_speech(self, text: str, voice_names: list[str], speed: float = 1.0, gpu_timeout: int = 60, progress_callback=None, progress_state=None, progress=None) -> Tuple[np.ndarray, float]:
|
|
|
55 |
if not text or not voice_names:
|
56 |
raise ValueError("Text and voice name are required")
|
57 |
|
58 |
+
# Handle voice selection
|
59 |
if isinstance(voice_names, list) and len(voice_names) > 1:
|
60 |
+
# For multiple voices, join them with underscore
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
61 |
voice_name = "_".join(voice_names)
|
|
|
|
|
|
|
62 |
else:
|
63 |
voice_name = voice_names[0]
|
|
|
|
|
|
|
|
|
|
|
|
|
64 |
|
65 |
# Initialize tracking
|
66 |
audio_chunks = []
|
|
|
128 |
# Concatenate audio chunks
|
129 |
audio = np.concatenate(audio_chunks)
|
130 |
|
|
|
|
|
|
|
|
|
|
|
|
|
131 |
|
132 |
# Return audio and metrics
|
133 |
return (
|
voices/v1_voices.json
ADDED
@@ -0,0 +1,32 @@
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
1 |
+
{
|
2 |
+
"american": [
|
3 |
+
"af_alloy",
|
4 |
+
"af_aoede",
|
5 |
+
"af_bella",
|
6 |
+
"af_jessica",
|
7 |
+
"af_kore",
|
8 |
+
"af_nicole",
|
9 |
+
"af_nova",
|
10 |
+
"af_river",
|
11 |
+
"af_sarah",
|
12 |
+
"af_sky",
|
13 |
+
"am_adam",
|
14 |
+
"am_echo",
|
15 |
+
"am_eric",
|
16 |
+
"am_fenrir",
|
17 |
+
"am_liam",
|
18 |
+
"am_michael",
|
19 |
+
"am_onyx",
|
20 |
+
"am_puck"
|
21 |
+
],
|
22 |
+
"british": [
|
23 |
+
"bf_alice",
|
24 |
+
"bf_emma",
|
25 |
+
"bf_isabella",
|
26 |
+
"bf_lily",
|
27 |
+
"bm_daniel",
|
28 |
+
"bm_fable",
|
29 |
+
"bm_george",
|
30 |
+
"bm_lewis"
|
31 |
+
]
|
32 |
+
}
|