Spaces:
Running
on
Zero
Running
on
Zero
Upload app.py with huggingface_hub
Browse files
app.py
ADDED
|
@@ -0,0 +1,264 @@
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
| 1 |
+
from __future__ import annotations
|
| 2 |
+
|
| 3 |
+
import contextlib
|
| 4 |
+
import io
|
| 5 |
+
import os
|
| 6 |
+
from pathlib import Path
|
| 7 |
+
from typing import List, Tuple
|
| 8 |
+
|
| 9 |
+
import gradio as gr
|
| 10 |
+
import torch
|
| 11 |
+
import spaces
|
| 12 |
+
|
| 13 |
+
from dia2 import Dia2, GenerationConfig, SamplingConfig
|
| 14 |
+
|
| 15 |
+
DEFAULT_REPO = os.environ.get("DIA2_DEFAULT_REPO", "nari-labs/Dia2-2B")
|
| 16 |
+
MAX_TURNS = 10
|
| 17 |
+
INITIAL_TURNS = 2
|
| 18 |
+
|
| 19 |
+
_dia: Dia2 | None = None
|
| 20 |
+
|
| 21 |
+
|
| 22 |
+
def _get_dia() -> Dia2:
|
| 23 |
+
global _dia
|
| 24 |
+
if _dia is None:
|
| 25 |
+
_dia = Dia2.from_repo(DEFAULT_REPO, device="cuda", dtype="bfloat16")
|
| 26 |
+
return _dia
|
| 27 |
+
|
| 28 |
+
|
| 29 |
+
def _concat_script(turn_count: int, turn_values: List[str]) -> str:
|
| 30 |
+
lines: List[str] = []
|
| 31 |
+
for idx in range(min(turn_count, len(turn_values))):
|
| 32 |
+
text = (turn_values[idx] or "").strip()
|
| 33 |
+
if not text:
|
| 34 |
+
continue
|
| 35 |
+
speaker = "[S1]" if idx % 2 == 0 else "[S2]"
|
| 36 |
+
lines.append(f"{speaker} {text}")
|
| 37 |
+
return "\n".join(lines)
|
| 38 |
+
|
| 39 |
+
|
| 40 |
+
EXAMPLES: dict[str, dict[str, List[str] | str | None]] = {
|
| 41 |
+
"Intro": {
|
| 42 |
+
"turns": [
|
| 43 |
+
"Hello Dia2 fans! Today we're unveiling the new open TTS model.",
|
| 44 |
+
"Sounds exciting. Can you show a sample right now?",
|
| 45 |
+
"Absolutely. (laughs) Just press generate.",
|
| 46 |
+
],
|
| 47 |
+
"voice_s1": "example_prefix1.wav",
|
| 48 |
+
"voice_s2": "example_prefix2.wav",
|
| 49 |
+
},
|
| 50 |
+
"Customer Support": {
|
| 51 |
+
"turns": [
|
| 52 |
+
"Thanks for calling. How can I help you today?",
|
| 53 |
+
"My parcel never arrived and it's been two weeks.",
|
| 54 |
+
"I'm sorry about that. Let me check your tracking number.",
|
| 55 |
+
"Appreciate it. I really need that package soon.",
|
| 56 |
+
],
|
| 57 |
+
"voice_s1": "example_prefix1.wav",
|
| 58 |
+
"voice_s2": "example_prefix2.wav",
|
| 59 |
+
},
|
| 60 |
+
}
|
| 61 |
+
|
| 62 |
+
|
| 63 |
+
def _apply_turn_visibility(count: int) -> List[gr.Update]:
|
| 64 |
+
return [gr.update(visible=i < count) for i in range(MAX_TURNS)]
|
| 65 |
+
|
| 66 |
+
|
| 67 |
+
def _add_turn(count: int):
|
| 68 |
+
count = min(count + 1, MAX_TURNS)
|
| 69 |
+
return (count, *_apply_turn_visibility(count))
|
| 70 |
+
|
| 71 |
+
|
| 72 |
+
def _remove_turn(count: int):
|
| 73 |
+
count = max(1, count - 1)
|
| 74 |
+
return (count, *_apply_turn_visibility(count))
|
| 75 |
+
|
| 76 |
+
|
| 77 |
+
def _load_example(name: str, count: int):
|
| 78 |
+
data = EXAMPLES.get(name)
|
| 79 |
+
if not data:
|
| 80 |
+
return (count, *_apply_turn_visibility(count), None, None)
|
| 81 |
+
turns = data.get("turns", [])
|
| 82 |
+
voice_s1_path = data.get("voice_s1")
|
| 83 |
+
voice_s2_path = data.get("voice_s2")
|
| 84 |
+
new_count = min(len(turns), MAX_TURNS)
|
| 85 |
+
updates: List[gr.Update] = []
|
| 86 |
+
for idx in range(MAX_TURNS):
|
| 87 |
+
if idx < new_count:
|
| 88 |
+
updates.append(gr.update(value=turns[idx], visible=True))
|
| 89 |
+
else:
|
| 90 |
+
updates.append(gr.update(value="", visible=idx < INITIAL_TURNS))
|
| 91 |
+
return (new_count, *updates, voice_s1_path, voice_s2_path)
|
| 92 |
+
|
| 93 |
+
|
| 94 |
+
def _prepare_prefix(file_path: str | None) -> str | None:
|
| 95 |
+
if not file_path:
|
| 96 |
+
return None
|
| 97 |
+
path = Path(file_path)
|
| 98 |
+
if not path.exists():
|
| 99 |
+
return None
|
| 100 |
+
return str(path)
|
| 101 |
+
|
| 102 |
+
|
| 103 |
+
@spaces.GPU(duration=100)
|
| 104 |
+
def generate_audio(
|
| 105 |
+
turn_count: int,
|
| 106 |
+
*inputs,
|
| 107 |
+
):
|
| 108 |
+
turn_values = list(inputs[:MAX_TURNS])
|
| 109 |
+
voice_s1 = inputs[MAX_TURNS]
|
| 110 |
+
voice_s2 = inputs[MAX_TURNS + 1]
|
| 111 |
+
cfg_scale = float(inputs[MAX_TURNS + 2])
|
| 112 |
+
text_temperature = float(inputs[MAX_TURNS + 3])
|
| 113 |
+
audio_temperature = float(inputs[MAX_TURNS + 4])
|
| 114 |
+
text_top_k = int(inputs[MAX_TURNS + 5])
|
| 115 |
+
audio_top_k = int(inputs[MAX_TURNS + 6])
|
| 116 |
+
include_prefix = bool(inputs[MAX_TURNS + 7])
|
| 117 |
+
|
| 118 |
+
script = _concat_script(turn_count, turn_values)
|
| 119 |
+
if not script.strip():
|
| 120 |
+
raise gr.Error("Please enter at least one non-empty speaker turn.")
|
| 121 |
+
|
| 122 |
+
dia = _get_dia()
|
| 123 |
+
config = GenerationConfig(
|
| 124 |
+
cfg_scale=cfg_scale,
|
| 125 |
+
text=SamplingConfig(temperature=text_temperature, top_k=text_top_k),
|
| 126 |
+
audio=SamplingConfig(temperature=audio_temperature, top_k=audio_top_k),
|
| 127 |
+
use_cuda_graph=True,
|
| 128 |
+
)
|
| 129 |
+
kwargs = {
|
| 130 |
+
"prefix_speaker_1": _prepare_prefix(voice_s1),
|
| 131 |
+
"prefix_speaker_2": _prepare_prefix(voice_s2),
|
| 132 |
+
"include_prefix": include_prefix,
|
| 133 |
+
}
|
| 134 |
+
buffer = io.StringIO()
|
| 135 |
+
with contextlib.redirect_stdout(buffer):
|
| 136 |
+
result = dia.generate(
|
| 137 |
+
script,
|
| 138 |
+
config=config,
|
| 139 |
+
output_wav=None,
|
| 140 |
+
verbose=True,
|
| 141 |
+
**kwargs,
|
| 142 |
+
)
|
| 143 |
+
waveform = result.waveform.detach().cpu().numpy()
|
| 144 |
+
sample_rate = result.sample_rate
|
| 145 |
+
timestamps = result.timestamps
|
| 146 |
+
log_text = buffer.getvalue().strip()
|
| 147 |
+
table = [[w, round(t, 3)] for w, t in timestamps]
|
| 148 |
+
return (sample_rate, waveform), table, log_text or "Generation finished."
|
| 149 |
+
|
| 150 |
+
|
| 151 |
+
def build_interface() -> gr.Blocks:
|
| 152 |
+
with gr.Blocks(
|
| 153 |
+
title="Dia2 TTS", css=".compact-turn textarea {min-height: 60px}"
|
| 154 |
+
) as demo:
|
| 155 |
+
gr.Markdown(
|
| 156 |
+
"""## Dia2 — Open TTS Model
|
| 157 |
+
Compose dialogue, attach optional voice prompts, and generate audio (CUDA graphs enabled by default)."""
|
| 158 |
+
)
|
| 159 |
+
turn_state = gr.State(INITIAL_TURNS)
|
| 160 |
+
with gr.Row(equal_height=True):
|
| 161 |
+
example_dropdown = gr.Dropdown(
|
| 162 |
+
choices=["(select example)"] + list(EXAMPLES.keys()),
|
| 163 |
+
label="Examples",
|
| 164 |
+
value="(select example)",
|
| 165 |
+
)
|
| 166 |
+
with gr.Row(equal_height=True):
|
| 167 |
+
with gr.Column(scale=1):
|
| 168 |
+
with gr.Group():
|
| 169 |
+
gr.Markdown("### Script")
|
| 170 |
+
controls = []
|
| 171 |
+
for idx in range(MAX_TURNS):
|
| 172 |
+
speaker = "[S1]" if idx % 2 == 0 else "[S2]"
|
| 173 |
+
box = gr.Textbox(
|
| 174 |
+
label=f"{speaker} turn {idx + 1}",
|
| 175 |
+
lines=2,
|
| 176 |
+
elem_classes=["compact-turn"],
|
| 177 |
+
placeholder=f"Enter dialogue for {speaker}…",
|
| 178 |
+
visible=idx < INITIAL_TURNS,
|
| 179 |
+
)
|
| 180 |
+
controls.append(box)
|
| 181 |
+
with gr.Row():
|
| 182 |
+
add_btn = gr.Button("Add Turn")
|
| 183 |
+
remove_btn = gr.Button("Remove Turn")
|
| 184 |
+
with gr.Group():
|
| 185 |
+
gr.Markdown("### Voice Prompts")
|
| 186 |
+
with gr.Row():
|
| 187 |
+
voice_s1 = gr.File(
|
| 188 |
+
label="[S1] voice (wav/mp3)", type="filepath"
|
| 189 |
+
)
|
| 190 |
+
voice_s2 = gr.File(
|
| 191 |
+
label="[S2] voice (wav/mp3)", type="filepath"
|
| 192 |
+
)
|
| 193 |
+
with gr.Group():
|
| 194 |
+
gr.Markdown("### Sampling")
|
| 195 |
+
cfg_scale = gr.Slider(
|
| 196 |
+
1.0, 8.0, value=6.0, step=0.1, label="CFG Scale"
|
| 197 |
+
)
|
| 198 |
+
with gr.Group():
|
| 199 |
+
gr.Markdown("#### Text Sampling")
|
| 200 |
+
text_temperature = gr.Slider(
|
| 201 |
+
0.1, 1.5, value=0.6, step=0.05, label="Text Temperature"
|
| 202 |
+
)
|
| 203 |
+
text_top_k = gr.Slider(
|
| 204 |
+
1, 200, value=50, step=1, label="Text Top-K"
|
| 205 |
+
)
|
| 206 |
+
with gr.Group():
|
| 207 |
+
gr.Markdown("#### Audio Sampling")
|
| 208 |
+
audio_temperature = gr.Slider(
|
| 209 |
+
0.1, 1.5, value=0.8, step=0.05, label="Audio Temperature"
|
| 210 |
+
)
|
| 211 |
+
audio_top_k = gr.Slider(
|
| 212 |
+
1, 200, value=50, step=1, label="Audio Top-K"
|
| 213 |
+
)
|
| 214 |
+
include_prefix = gr.Checkbox(
|
| 215 |
+
label="Keep prefix audio in output", value=False
|
| 216 |
+
)
|
| 217 |
+
generate_btn = gr.Button("Generate", variant="primary")
|
| 218 |
+
with gr.Column(scale=1):
|
| 219 |
+
gr.Markdown("### Output")
|
| 220 |
+
audio_out = gr.Audio(label="Waveform", interactive=False)
|
| 221 |
+
timestamps = gr.Dataframe(
|
| 222 |
+
headers=["word", "seconds"], label="Timestamps"
|
| 223 |
+
)
|
| 224 |
+
log_box = gr.Textbox(label="Logs", lines=8)
|
| 225 |
+
|
| 226 |
+
add_btn.click(
|
| 227 |
+
lambda c: _add_turn(c),
|
| 228 |
+
inputs=turn_state,
|
| 229 |
+
outputs=[turn_state, *controls],
|
| 230 |
+
)
|
| 231 |
+
remove_btn.click(
|
| 232 |
+
lambda c: _remove_turn(c),
|
| 233 |
+
inputs=turn_state,
|
| 234 |
+
outputs=[turn_state, *controls],
|
| 235 |
+
)
|
| 236 |
+
example_dropdown.change(
|
| 237 |
+
lambda name, c: _load_example(name, c),
|
| 238 |
+
inputs=[example_dropdown, turn_state],
|
| 239 |
+
outputs=[turn_state, *controls, voice_s1, voice_s2],
|
| 240 |
+
)
|
| 241 |
+
|
| 242 |
+
generate_btn.click(
|
| 243 |
+
generate_audio,
|
| 244 |
+
inputs=[
|
| 245 |
+
turn_state,
|
| 246 |
+
*controls,
|
| 247 |
+
voice_s1,
|
| 248 |
+
voice_s2,
|
| 249 |
+
cfg_scale,
|
| 250 |
+
text_temperature,
|
| 251 |
+
audio_temperature,
|
| 252 |
+
text_top_k,
|
| 253 |
+
audio_top_k,
|
| 254 |
+
include_prefix,
|
| 255 |
+
],
|
| 256 |
+
outputs=[audio_out, timestamps, log_box],
|
| 257 |
+
)
|
| 258 |
+
return demo
|
| 259 |
+
|
| 260 |
+
|
| 261 |
+
if __name__ == "__main__":
|
| 262 |
+
app = build_interface()
|
| 263 |
+
app.queue(default_concurrency_limit=1)
|
| 264 |
+
app.launch(share=True)
|