File size: 23,386 Bytes
2941e2c e3aec0d 1264f26 2941e2c e3aec0d 2941e2c 20e96fd 2941e2c 0a59e8a 2941e2c 1264f26 2941e2c e3aec0d 2941e2c aceac12 2941e2c e3aec0d 2941e2c e3aec0d 2941e2c e3aec0d 2941e2c 20e96fd 2941e2c 20e96fd 2941e2c 20e96fd 2941e2c 20e96fd 2941e2c 20e96fd 1264f26 20e96fd 1264f26 20e96fd 1264f26 20e96fd 1264f26 20e96fd 2941e2c 0a59e8a 2941e2c 0a59e8a 2941e2c 20e96fd 2941e2c 20e96fd 2941e2c e3aec0d aceac12 2941e2c |
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 99 100 101 102 103 104 105 106 107 108 109 110 111 112 113 114 115 116 117 118 119 120 121 122 123 124 125 126 127 128 129 130 131 132 133 134 135 136 137 138 139 140 141 142 143 144 145 146 147 148 149 150 151 152 153 154 155 156 157 158 159 160 161 162 163 164 165 166 167 168 169 170 171 172 173 174 175 176 177 178 179 180 181 182 183 184 185 186 187 188 189 190 191 192 193 194 195 196 197 198 199 200 201 202 203 204 205 206 207 208 209 210 211 212 213 214 215 216 217 218 219 220 221 222 223 224 225 226 227 228 229 230 231 232 233 234 235 236 237 238 239 240 241 242 243 244 245 246 247 248 249 250 251 252 253 254 255 256 257 258 259 260 261 262 263 264 265 266 267 268 269 270 271 272 273 274 275 276 277 278 279 280 281 282 283 284 285 286 287 288 289 290 291 292 293 294 295 296 297 298 299 300 301 302 303 304 305 306 307 308 309 310 311 312 313 314 315 316 317 318 319 320 321 322 323 324 325 326 327 328 329 330 331 332 333 334 335 336 337 338 339 340 341 342 343 344 345 346 347 348 349 350 351 352 353 354 355 356 357 358 359 360 361 362 363 364 365 366 367 368 369 370 371 372 373 374 375 376 377 378 379 380 381 382 383 384 385 386 387 388 389 390 391 392 393 394 395 396 397 398 399 400 401 402 403 404 405 406 407 408 409 410 411 412 413 414 415 416 417 418 419 420 421 422 423 424 425 |
"""Gradio UI interface for Caribbean Voices OWSM platform."""
import gradio as gr
import time
import os
from pathlib import Path
from datetime import datetime
# Import modules
from utils.status import get_status_display, get_data_loading_status
from utils.entities import extract_entities_progress
from training.espnet_trainer import run_espnet_training_progress
from training.whisper_trainer import run_whisper_training_progress
from models.inference import transcribe_audio, run_inference_owsm
from models.loader import get_available_models, get_available_checkpoints
from data.loader import load_data_from_hf_dataset
from utils.logging import get_latest_log_file, get_all_log_files, get_log_directory
def create_interface():
"""Create and return the Gradio interface"""
interface_start = time.time()
with gr.Blocks(title="Caribbean Voices - OWSM Platform") as demo:
gr.Markdown("""
<div class="main-header">
<h1>🎤 Caribbean Voices Hackathon</h1>
<p>OWSM v3.1 Training & Inference Platform</p>
</div>
""")
with gr.Tabs() as tabs:
# Tab 1: Status & Setup (Homepage)
with gr.Tab("🏠 Home", id=0):
status_display = gr.HTML(value=get_status_display())
refresh_status_btn = gr.Button("🔄 Refresh Status", variant="secondary", size="lg")
# Navigation buttons
gr.Markdown("""
<div class="nav-buttons-grid">
""")
nav_buttons_row1 = gr.Row()
with nav_buttons_row1:
nav_load_data = gr.Button("📥 Load Data", variant="primary", size="lg", scale=1)
nav_entity_extraction = gr.Button("🔍 Entity Extraction", variant="primary", size="lg", scale=1)
nav_training = gr.Button("🏋️ Training", variant="primary", size="lg", scale=1)
nav_buttons_row2 = gr.Row()
with nav_buttons_row2:
nav_inference = gr.Button("🚀 Inference", variant="primary", size="lg", scale=1)
nav_single_file = gr.Button("🎯 Single File", variant="primary", size="lg", scale=1)
nav_about = gr.Button("📚 About", variant="secondary", size="lg", scale=1)
gr.Markdown("</div>")
# Add project info section
gr.Markdown("""
<div style="margin-top: 40px; padding: 20px; background: #fff; border-radius: 10px; border: 1px solid #e0e0e0;">
<h2 style="color: #667eea; margin-top: 0;">📊 About This Project</h2>
<p style="font-size: 1.05em; line-height: 1.6; color: #555;">
The <strong>Caribbean Voices Hackathon</strong> project focuses on building an advanced Automatic Speech Recognition (ASR)
system using OWSM v3.1 (Open Whisper-Style Model). This platform enables fine-tuning on Caribbean-accented speech
with specialized entity extraction and contextual biasing for improved recognition of Caribbean proper nouns,
locations, and organizations.
</p>
<h3 style="color: #667eea;">Key Features</h3>
<ul style="font-size: 1.05em; line-height: 1.8; color: #555;">
<li><strong>Entity Extraction:</strong> Automatically identifies Caribbean-specific entities from training transcripts</li>
<li><strong>OWSM Fine-tuning:</strong> Fine-tune the OWSM v3.1 model with entity-weighted loss</li>
<li><strong>Batch Inference:</strong> Process entire test sets efficiently</li>
<li><strong>Single File Testing:</strong> Quick transcription with multiple model options</li>
<li><strong>ESPnet Integration:</strong> Full support for ESPnet training recipes</li>
</ul>
</div>
""")
def refresh_status():
return get_status_display()
refresh_status_btn.click(
fn=refresh_status,
outputs=[status_display]
)
# Navigation button handlers - use JavaScript to switch tabs
nav_load_data.click(
None, None, None,
js="() => { setTimeout(() => { const tabs = document.querySelectorAll('button[role=\\'tab\\']'); if(tabs[1]) tabs[1].click(); }, 100); }"
)
nav_entity_extraction.click(
None, None, None,
js="() => { setTimeout(() => { const tabs = document.querySelectorAll('button[role=\\'tab\\']'); if(tabs[2]) tabs[2].click(); }, 100); }"
)
nav_training.click(
None, None, None,
js="() => { setTimeout(() => { const tabs = document.querySelectorAll('button[role=\\'tab\\']'); if(tabs[3]) tabs[3].click(); }, 100); }"
)
nav_inference.click(
None, None, None,
js="() => { setTimeout(() => { const tabs = document.querySelectorAll('button[role=\\'tab\\']'); if(tabs[4]) tabs[4].click(); }, 100); }"
)
nav_single_file.click(
None, None, None,
js="() => { setTimeout(() => { const tabs = document.querySelectorAll('button[role=\\'tab\\']'); if(tabs[5]) tabs[5].click(); }, 100); }"
)
nav_about.click(
None, None, None,
js="() => { setTimeout(() => { const tabs = document.querySelectorAll('button[role=\\'tab\\']'); if(tabs[6]) tabs[6].click(); }, 100); }"
)
# Tab 2: Data Loading
with gr.Tab("📥 Load Data"):
gr.Markdown("### Load Dataset into HF Space")
# Show current data status
data_status_display = gr.Markdown(value=get_data_loading_status())
refresh_data_status_btn = gr.Button("🔄 Refresh Status", variant="secondary", size="sm")
gr.Markdown("""
---
### Load Dataset
Data is automatically loaded from the Hugging Face dataset on startup.
You can manually load a different dataset below if needed.
""")
hf_dataset_name = gr.Textbox(
label="Hugging Face Dataset Name",
placeholder="username/dataset-name",
value=""
)
hf_load_btn = gr.Button("Load from HF Dataset", variant="primary")
hf_load_output = gr.Markdown()
# Refresh data status when buttons are clicked
def refresh_data_status():
return get_data_loading_status()
refresh_data_status_btn.click(
fn=refresh_data_status,
outputs=[data_status_display]
)
def load_hf_and_refresh(dataset_name, progress=gr.Progress()):
result = load_data_from_hf_dataset(dataset_name, progress)
return result, get_data_loading_status()
hf_load_btn.click(
fn=load_hf_and_refresh,
inputs=[hf_dataset_name],
outputs=[hf_load_output, data_status_display]
)
# Tab 3: Entity Extraction
with gr.Tab("🔍 Entity Extraction"):
gr.Markdown("### Extract Caribbean Entities from Training Data")
gr.Markdown("""
This extracts high-value Caribbean entities (proper nouns, locations, organizations)
from the training transcripts. These entities will be used for:
- Entity-weighted loss during training
- Contextual biasing during inference
""")
extract_btn = gr.Button("Extract Entities", variant="primary")
extract_output = gr.Markdown()
extract_json = gr.JSON(label="Entities JSON")
extract_btn.click(
fn=extract_entities_progress,
outputs=[extract_output, extract_json]
)
# Tab 4: Training (with sub-tabs for ESPnet and Whisper)
with gr.Tab("🏋️ Training"):
gr.Markdown("### Model Training")
gr.Markdown("""
Choose your training framework:
- **ESPnet Training**: For ESPnet OWSM models (requires ESPnet recipes)
- **Whisper Training**: For Whisper models (full HuggingFace integration)
""")
with gr.Tabs() as training_tabs:
# ESPnet Training Tab
with gr.Tab("🔧 ESPnet Training"):
gr.Markdown("### ESPnet OWSM Model Training")
gr.Markdown("""
**ESPnet Training** - Uses ESPnet's native framework.
This loads ESPnet models and prepares them for training with ESPnet recipes.
Full fine-tuning requires ESPnet training recipes.
""")
with gr.Row():
with gr.Column():
espnet_train_epochs = gr.Slider(1, 10, value=3, step=1, label="Epochs (for ESPnet recipes)")
espnet_train_batch_size = gr.Slider(1, 32, value=4, step=1, label="Batch Size (for ESPnet recipes)")
espnet_train_lr = gr.Slider(1e-6, 1e-3, value=3e-5, step=1e-6, label="Learning Rate (for ESPnet recipes)")
espnet_train_btn = gr.Button("Load ESPnet Model", variant="primary")
with gr.Column():
espnet_train_output = gr.Markdown()
espnet_train_metrics = gr.JSON(label="Model Info")
espnet_train_btn.click(
fn=run_espnet_training_progress,
inputs=[espnet_train_epochs, espnet_train_batch_size, espnet_train_lr],
outputs=[espnet_train_output, espnet_train_metrics]
)
# Whisper Training Tab
with gr.Tab("🎤 Whisper Training"):
gr.Markdown("### Whisper Model Training")
gr.Markdown("""
**Whisper Training** - Full HuggingFace transformers integration.
Fine-tune Whisper models with entity-weighted loss using HuggingFace's training framework.
Includes full support for HuggingFace features like early stopping, WER metrics, etc.
""")
with gr.Row():
with gr.Column():
gr.Markdown("#### Training Hyperparameters")
whisper_train_epochs = gr.Slider(1, 10, value=3, step=1, label="Epochs")
whisper_train_batch_size = gr.Slider(1, 32, value=4, step=1, label="Batch Size")
whisper_train_lr = gr.Slider(1e-6, 1e-3, value=3e-5, step=1e-6, label="Learning Rate")
gr.Markdown("#### Speed Augmentation")
gr.Markdown("Speed factors for dataset expansion (creates multiple versions of each sample)")
speed_aug_enabled = gr.Checkbox(value=True, label="Enable Speed Augmentation")
speed_factor_min = gr.Slider(0.8, 1.0, value=0.9, step=0.05, label="Min Speed Factor")
speed_factor_max = gr.Slider(1.0, 1.2, value=1.1, step=0.05, label="Max Speed Factor")
speed_factor_count = gr.Slider(2, 5, value=3, step=1, label="Number of Speed Variants")
gr.Markdown("#### SpecAugment Parameters")
gr.Markdown("Spectrogram augmentation settings (applied during training)")
specaug_enabled = gr.Checkbox(value=True, label="Enable SpecAugment")
specaug_time_mask = gr.Slider(0, 50, value=27, step=1, label="Time Mask Parameter")
specaug_freq_mask = gr.Slider(0, 20, value=10, step=1, label="Frequency Mask Parameter")
specaug_time_warp = gr.Checkbox(value=True, label="Enable Time Warping")
specaug_warp_param = gr.Slider(0, 80, value=40, step=5, label="Time Warp Parameter")
whisper_train_btn = gr.Button("Start Whisper Training", variant="primary", size="lg")
with gr.Column():
whisper_train_output = gr.Markdown()
whisper_train_metrics = gr.JSON(label="Training Metrics")
gr.Markdown("#### Training Logs")
log_info = gr.Markdown(f"Log directory: `{get_log_directory()}`")
latest_log_file = gr.File(
label="Download Latest Training Log",
visible=False
)
def update_log_download():
latest = get_latest_log_file("whisper_training")
if latest and os.path.exists(latest):
return gr.File(value=latest, visible=True)
return gr.File(visible=False)
refresh_log_btn = gr.Button("🔄 Refresh Logs", variant="secondary", size="sm")
refresh_log_btn.click(
fn=update_log_download,
outputs=[latest_log_file]
)
def run_training_with_log_refresh(
epochs, batch_size, lr,
speed_aug_enabled, speed_factor_min, speed_factor_max, speed_factor_count,
specaug_enabled, specaug_time_mask, specaug_freq_mask, specaug_time_warp, specaug_warp_param,
progress=gr.Progress()
):
"""Run training and refresh log download after completion."""
result = run_whisper_training_progress(
epochs, batch_size, lr,
speed_aug_enabled, speed_factor_min, speed_factor_max, speed_factor_count,
specaug_enabled, specaug_time_mask, specaug_freq_mask, specaug_time_warp, specaug_warp_param,
progress
)
latest_log = update_log_download()
return result[0], result[1], latest_log
whisper_train_btn.click(
fn=run_training_with_log_refresh,
inputs=[
whisper_train_epochs,
whisper_train_batch_size,
whisper_train_lr,
speed_aug_enabled,
speed_factor_min,
speed_factor_max,
speed_factor_count,
specaug_enabled,
specaug_time_mask,
specaug_freq_mask,
specaug_time_warp,
specaug_warp_param,
],
outputs=[whisper_train_output, whisper_train_metrics, latest_log_file]
)
# Tab 5: Inference
with gr.Tab("🚀 Inference"):
gr.Markdown("### Run Inference on Test Set")
gr.Markdown("Generate transcriptions for all test files using a trained checkpoint or base model")
# Checkpoint selection
checkpoint_choices = get_available_checkpoints()
if not checkpoint_choices:
checkpoint_choices = ["No checkpoints available - train a model first"]
checkpoint_default = checkpoint_choices[0]
else:
checkpoint_default = checkpoint_choices[0] if checkpoint_choices else None
checkpoint_dropdown = gr.Dropdown(
choices=checkpoint_choices,
value=checkpoint_default,
label="Select Checkpoint/Model",
info="Choose a trained checkpoint or base model for inference"
)
def refresh_checkpoints():
"""Refresh checkpoint list"""
checkpoints = get_available_checkpoints()
if not checkpoints:
return gr.Dropdown(choices=["No checkpoints available - train a model first"], value="No checkpoints available - train a model first")
return gr.Dropdown(choices=checkpoints, value=checkpoints[0])
refresh_checkpoints_btn = gr.Button("🔄 Refresh Checkpoint List", variant="secondary", size="sm")
refresh_checkpoints_btn.click(
fn=refresh_checkpoints,
outputs=[checkpoint_dropdown]
)
infer_btn = gr.Button("Run Inference", variant="primary")
infer_output = gr.Markdown()
infer_download = gr.File(label="Download Submission CSV")
infer_btn.click(
fn=run_inference_owsm,
inputs=[checkpoint_dropdown],
outputs=[infer_output, infer_download]
)
# Tab 6: Single File Transcription
with gr.Tab("🎯 Single File"):
gr.Markdown("### Transcribe a Single Audio File")
with gr.Row():
with gr.Column():
audio_input = gr.Audio(
label="Upload Audio File",
type="filepath",
sources=["upload", "microphone"]
)
model_choice = gr.Dropdown(
choices=get_available_models(),
value=get_available_models()[0],
label="Select Model"
)
max_seconds = gr.Slider(5, 60, value=30, step=5, label="Max Audio Length (seconds)")
transcribe_btn = gr.Button("Transcribe", variant="primary")
with gr.Column():
transcription_output = gr.Textbox(
label="Transcription",
lines=5,
placeholder="Transcription will appear here..."
)
info_output = gr.Markdown(label="Processing Info")
transcribe_btn.click(
fn=transcribe_audio,
inputs=[audio_input, model_choice, max_seconds],
outputs=[transcription_output, info_output]
)
# Tab 7: About
with gr.Tab("📚 About"):
gr.Markdown("""
## Caribbean Voices Hackathon - OWSM v3.1 Platform
### Features
- **Entity Extraction**: Extract Caribbean entities from training data
- **Model Training**: Fine-tune OWSM v3.1 with entity-weighted loss
- **Batch Inference**: Generate transcriptions for test set
- **Single File Transcription**: Quick transcription with multiple models
### OWSM v3.1 Features
- **Emergent Contextual Biasing**: Improves proper noun recognition
- **Entity-Weighted Loss**: Prioritizes Caribbean entities during training
- **Competition Compliant**: Single model, no external data
### Available Models
- **Wav2Vec2 Models**: Fast baseline models
- **OWSM v3.1 Small**: Open Whisper-style model with ESPnet
### Workflow
1. **Extract Entities**: Run entity extraction on training data
2. **Train Model**:
- **ESPnet Training**: Load ESPnet models (requires ESPnet recipes for fine-tuning)
- **Whisper Training**: Full HuggingFace fine-tuning with entity-weighted loss
3. **Run Inference**: Generate test set transcriptions
4. **Download Results**: Get submission CSV file
### Technical Details
- **ESPnet Framework**: ESPnet + PyTorch for ESPnet OWSM models
- **Whisper Framework**: HuggingFace transformers for Whisper models
- **Model**: OWSM v3.1 E-Branchformer (ESPnet) or Whisper (HuggingFace)
- **Entity Extraction**: Frequency + capitalization analysis
- **Training**: Entity-weighted cross-entropy loss
### Documentation
See `ESPNET_OWSM_SETUP.md` and `IMPLEMENTATION_SUMMARY.md` for details.
""")
interface_time = time.time() - interface_start
timestamp = datetime.now().strftime("%H:%M:%S.%f")[:-3]
print(f"[{timestamp}] ⏱️ Total interface creation: {interface_time:.3f}s")
# Return demo and CSS path for Gradio 6.x (CSS goes in launch())
css_path = Path(__file__).parent / "styles.css"
return demo, css_path
|