Spaces:
Running
on
Zero
Running
on
Zero
Update app.py
Browse files
app.py
CHANGED
@@ -4,7 +4,7 @@ import gradio as gr
|
|
4 |
import spaces
|
5 |
|
6 |
from zonos.model import Zonos
|
7 |
-
from zonos.conditioning import make_cond_dict
|
8 |
|
9 |
# We'll keep a global dictionary of loaded models to avoid reloading
|
10 |
MODELS_CACHE = {}
|
@@ -13,6 +13,15 @@ device = "cuda"
|
|
13 |
banner_url = "https://huggingface.co/datasets/Steveeeeeeen/random_images/resolve/main/ZonosHeader.png"
|
14 |
BANNER = f'<div style="display: flex; justify-content: space-around;"><img src="{banner_url}" alt="Banner" style="width: 40vw; min-width: 150px; max-width: 300px;"> </div>'
|
15 |
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
16 |
def load_model(model_name: str):
|
17 |
"""
|
18 |
Loads or retrieves a cached Zonos model, sets it to eval and bfloat16.
|
@@ -28,15 +37,20 @@ def load_model(model_name: str):
|
|
28 |
return MODELS_CACHE[model_name]
|
29 |
|
30 |
@spaces.GPU(duration=90)
|
31 |
-
def tts(text, speaker_audio,
|
32 |
"""
|
33 |
text: str (Text prompt to synthesize)
|
34 |
speaker_audio: (sample_rate, numpy_array) from Gradio if type="numpy"
|
35 |
-
|
36 |
model_choice: str (which Zonos model to use, e.g., "Zyphra/Zonos-v0.1-hybrid")
|
37 |
|
38 |
Returns (sr_out, wav_out_numpy).
|
39 |
"""
|
|
|
|
|
|
|
|
|
|
|
40 |
model = load_model(model_choice)
|
41 |
|
42 |
if not text:
|
@@ -52,12 +66,11 @@ def tts(text, speaker_audio, selected_language, model_choice):
|
|
52 |
# Convert to Torch tensor
|
53 |
wav_tensor = torch.from_numpy(wav_np).float()
|
54 |
|
55 |
-
# If stereo
|
56 |
-
# e.g. shape (2, samples) -> shape (samples,) by averaging
|
57 |
if wav_tensor.ndim == 2 and wav_tensor.shape[0] > 1:
|
58 |
-
wav_tensor = wav_tensor.mean(dim=0) #
|
59 |
|
60 |
-
#
|
61 |
wav_tensor = wav_tensor.unsqueeze(0)
|
62 |
|
63 |
# Get speaker embedding
|
@@ -66,12 +79,12 @@ def tts(text, speaker_audio, selected_language, model_choice):
|
|
66 |
spk_embedding = spk_embedding.to(device, dtype=torch.bfloat16)
|
67 |
|
68 |
# Prepare conditioning dictionary
|
69 |
-
cond_dict =
|
70 |
-
text
|
71 |
-
speaker
|
72 |
-
language
|
73 |
-
device
|
74 |
-
|
75 |
conditioning = model.prepare_conditioning(cond_dict)
|
76 |
|
77 |
# Generate codes
|
@@ -106,8 +119,6 @@ def build_demo():
|
|
106 |
ref_audio_input = gr.Audio(
|
107 |
label="Reference Audio (Speaker Cloning)",
|
108 |
type="numpy"
|
109 |
-
# Optionally add mono=True if you want Gradio to always downmix automatically:
|
110 |
-
# mono=True
|
111 |
)
|
112 |
|
113 |
model_dropdown = gr.Dropdown(
|
@@ -116,10 +127,12 @@ def build_demo():
|
|
116 |
value="Zyphra/Zonos-v0.1-hybrid",
|
117 |
interactive=True,
|
118 |
)
|
|
|
|
|
119 |
language_dropdown = gr.Dropdown(
|
120 |
-
label="Language
|
121 |
-
choices=
|
122 |
-
value="
|
123 |
interactive=True,
|
124 |
)
|
125 |
|
|
|
4 |
import spaces
|
5 |
|
6 |
from zonos.model import Zonos
|
7 |
+
from zonos.conditioning import make_cond_dict # Keep this; remove supported_language_codes
|
8 |
|
9 |
# We'll keep a global dictionary of loaded models to avoid reloading
|
10 |
MODELS_CACHE = {}
|
|
|
13 |
banner_url = "https://huggingface.co/datasets/Steveeeeeeen/random_images/resolve/main/ZonosHeader.png"
|
14 |
BANNER = f'<div style="display: flex; justify-content: space-around;"><img src="{banner_url}" alt="Banner" style="width: 40vw; min-width: 150px; max-width: 300px;"> </div>'
|
15 |
|
16 |
+
# Define a list of tuples: (Display Label, Language Code)
|
17 |
+
LANGUAGES = [
|
18 |
+
("English", "en-us"),
|
19 |
+
("Japanese", "ja"),
|
20 |
+
("Chinese", "cmn"),
|
21 |
+
("French", "fr-fr"),
|
22 |
+
("German", "de"),
|
23 |
+
]
|
24 |
+
|
25 |
def load_model(model_name: str):
|
26 |
"""
|
27 |
Loads or retrieves a cached Zonos model, sets it to eval and bfloat16.
|
|
|
37 |
return MODELS_CACHE[model_name]
|
38 |
|
39 |
@spaces.GPU(duration=90)
|
40 |
+
def tts(text, speaker_audio, selected_language_label, model_choice):
|
41 |
"""
|
42 |
text: str (Text prompt to synthesize)
|
43 |
speaker_audio: (sample_rate, numpy_array) from Gradio if type="numpy"
|
44 |
+
selected_language_label: str (the display name from the dropdown, e.g. "Chinese")
|
45 |
model_choice: str (which Zonos model to use, e.g., "Zyphra/Zonos-v0.1-hybrid")
|
46 |
|
47 |
Returns (sr_out, wav_out_numpy).
|
48 |
"""
|
49 |
+
# Map from label -> actual language code
|
50 |
+
label_to_code = dict(LANGUAGES)
|
51 |
+
# Convert the human-readable label back to the code
|
52 |
+
selected_language = label_to_code[selected_language_label]
|
53 |
+
|
54 |
model = load_model(model_choice)
|
55 |
|
56 |
if not text:
|
|
|
66 |
# Convert to Torch tensor
|
67 |
wav_tensor = torch.from_numpy(wav_np).float()
|
68 |
|
69 |
+
# If stereo or multi-channel, downmix to mono
|
|
|
70 |
if wav_tensor.ndim == 2 and wav_tensor.shape[0] > 1:
|
71 |
+
wav_tensor = wav_tensor.mean(dim=0) # => (samples,)
|
72 |
|
73 |
+
# Add batch dimension => (1, samples)
|
74 |
wav_tensor = wav_tensor.unsqueeze(0)
|
75 |
|
76 |
# Get speaker embedding
|
|
|
79 |
spk_embedding = spk_embedding.to(device, dtype=torch.bfloat16)
|
80 |
|
81 |
# Prepare conditioning dictionary
|
82 |
+
cond_dict = {
|
83 |
+
"text": text,
|
84 |
+
"speaker": spk_embedding,
|
85 |
+
"language": selected_language, # Use the code here
|
86 |
+
"device": device,
|
87 |
+
}
|
88 |
conditioning = model.prepare_conditioning(cond_dict)
|
89 |
|
90 |
# Generate codes
|
|
|
119 |
ref_audio_input = gr.Audio(
|
120 |
label="Reference Audio (Speaker Cloning)",
|
121 |
type="numpy"
|
|
|
|
|
122 |
)
|
123 |
|
124 |
model_dropdown = gr.Dropdown(
|
|
|
127 |
value="Zyphra/Zonos-v0.1-hybrid",
|
128 |
interactive=True,
|
129 |
)
|
130 |
+
|
131 |
+
# For the language dropdown, we display only the friendly label
|
132 |
language_dropdown = gr.Dropdown(
|
133 |
+
label="Language",
|
134 |
+
choices=[label for (label, code) in LANGUAGES],
|
135 |
+
value="English", # default display
|
136 |
interactive=True,
|
137 |
)
|
138 |
|