Spaces:
Sleeping
Sleeping
File size: 51,469 Bytes
207e539 179134f 797ef46 135bef5 7e26368 797ef46 207e539 179134f f4239e7 295a917 f4239e7 207e539 295a917 207e539 11d1c92 3fbfba6 11d1c92 3fbfba6 797ef46 87c5acf 207e539 9075cf3 e9e3c76 207e539 e9e3c76 207e539 70ce2c6 207e539 7e2d7f4 70ce2c6 207e539 e9e3c76 207e539 e18847d e9e3c76 e18847d 207e539 bf1fd5b 207e539 bf1fd5b 207e539 bf1fd5b 207e539 bf1fd5b 207e539 bf1fd5b 207e539 bf1fd5b 207e539 bf1fd5b 207e539 bf1fd5b 207e539 bf1fd5b 207e539 bf1fd5b 207e539 bf1fd5b 207e539 bf1fd5b 207e539 bf1fd5b 207e539 bf1fd5b 207e539 bf1fd5b 207e539 bf1fd5b 207e539 bf1fd5b 207e539 bf1fd5b 207e539 bf1fd5b 9075cf3 0749314 a8f4e47 bf1fd5b a8f4e47 207e539 797ef46 207e539 797ef46 207e539 29cd28a 207e539 797ef46 |
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 99 100 101 102 103 104 105 106 107 108 109 110 111 112 113 114 115 116 117 118 119 120 121 122 123 124 125 126 127 128 129 130 131 132 133 134 135 136 137 138 139 140 141 142 143 144 145 146 147 148 149 150 151 152 153 154 155 156 157 158 159 160 161 162 163 164 165 166 167 168 169 170 171 172 173 174 175 176 177 178 179 180 181 182 183 184 185 186 187 188 189 190 191 192 193 194 195 196 197 198 199 200 201 202 203 204 205 206 207 208 209 210 211 212 213 214 215 216 217 218 219 220 221 222 223 224 225 226 227 228 229 230 231 232 233 234 235 236 237 238 239 240 241 242 243 244 245 246 247 248 249 250 251 252 253 254 255 256 257 258 259 260 261 262 263 264 265 266 267 268 269 270 271 272 273 274 275 276 277 278 279 280 281 282 283 284 285 286 287 288 289 290 291 292 293 294 295 296 297 298 299 300 301 302 303 304 305 306 307 308 309 310 311 312 313 314 315 316 317 318 319 320 321 322 323 324 325 326 327 328 329 330 331 332 333 334 335 336 337 338 339 340 341 342 343 344 345 346 347 348 349 350 351 352 353 354 355 356 357 358 359 360 361 362 363 364 365 366 367 368 369 370 371 372 373 374 375 376 377 378 379 380 381 382 383 384 385 386 387 388 389 390 391 392 393 394 395 396 397 398 399 400 401 402 403 404 405 406 407 408 409 410 411 412 413 414 415 416 417 418 419 420 421 422 423 424 425 426 427 428 429 430 431 432 433 434 435 436 437 438 439 440 441 442 443 444 445 446 447 448 449 450 451 452 453 454 455 456 457 458 459 460 461 462 463 464 465 466 467 468 469 470 471 472 473 474 475 476 477 478 479 480 481 482 483 484 485 486 487 488 489 490 491 492 493 494 495 496 497 498 499 500 501 502 503 504 505 506 507 508 509 510 511 512 513 514 515 516 517 518 519 520 521 522 523 524 525 526 527 528 529 530 531 532 533 534 535 536 537 538 539 540 541 542 543 544 545 546 547 548 549 550 551 552 553 554 555 556 557 558 559 560 561 562 563 564 565 566 567 568 569 570 571 572 573 574 575 576 577 578 579 580 581 582 583 584 585 586 587 588 589 590 591 592 593 594 595 596 597 598 599 600 601 602 603 604 605 606 607 608 609 610 611 612 613 614 615 616 617 618 619 620 621 622 623 624 625 626 627 628 629 630 631 632 633 634 635 636 637 638 639 640 641 642 643 644 645 646 647 648 649 650 651 652 653 654 655 656 657 658 659 660 661 662 663 664 665 666 667 668 669 670 671 672 673 674 675 676 677 678 679 680 681 682 683 684 685 686 687 688 689 690 691 692 693 694 695 696 697 698 699 700 701 702 703 704 705 706 707 708 709 710 711 712 713 714 715 716 717 718 719 720 721 722 723 724 725 726 727 728 729 730 731 732 733 734 735 736 737 738 739 740 741 742 743 744 745 746 747 748 749 750 751 752 753 754 755 756 757 758 759 760 761 762 763 764 765 766 767 768 769 770 771 772 773 774 775 776 777 778 779 780 781 782 783 784 785 786 787 788 789 790 791 792 793 794 795 796 797 798 799 800 801 802 803 804 805 806 807 808 809 810 811 812 813 814 815 816 817 818 819 820 821 822 823 824 825 826 827 828 829 830 831 832 833 834 835 836 837 838 839 840 841 842 843 844 845 846 847 848 849 850 851 852 853 854 855 856 857 858 859 860 861 862 863 864 865 866 867 868 869 870 871 872 873 874 875 876 877 878 879 880 881 882 883 884 885 886 887 888 889 890 891 892 893 894 895 896 897 898 899 900 901 902 903 904 905 906 907 908 909 910 911 912 913 914 915 916 917 918 919 920 921 922 923 924 925 926 927 928 929 930 931 932 933 934 935 936 937 938 939 940 941 942 943 944 945 946 947 948 949 950 951 952 953 954 955 956 957 958 959 960 961 962 963 964 965 966 967 968 969 970 971 972 973 974 975 976 977 978 979 980 981 982 983 984 985 986 987 988 989 990 991 992 993 994 995 996 997 998 999 1000 1001 1002 1003 1004 1005 1006 1007 1008 1009 1010 1011 1012 1013 1014 1015 1016 1017 1018 1019 1020 1021 1022 1023 1024 1025 1026 1027 1028 1029 1030 1031 1032 1033 1034 1035 1036 1037 1038 1039 1040 1041 1042 1043 1044 1045 1046 1047 1048 1049 1050 1051 1052 1053 1054 1055 1056 1057 1058 1059 1060 1061 1062 1063 1064 1065 1066 1067 1068 1069 1070 1071 1072 1073 1074 1075 1076 1077 1078 1079 1080 1081 1082 1083 1084 1085 1086 1087 1088 1089 1090 1091 1092 1093 1094 1095 1096 1097 1098 1099 1100 1101 1102 1103 1104 1105 1106 1107 1108 1109 1110 1111 1112 1113 1114 1115 1116 1117 1118 1119 1120 1121 1122 1123 1124 1125 1126 1127 1128 |
import gradio as gr
import os
import sys
import subprocess
import importlib
from pathlib import Path
import json
import pandas as pd
def install_private_package():
"""Install private package from GitHub using token"""
print("Installing private package...")
gh_token = os.environ.get("GH_TOKEN")
if not gh_token:
raise ValueError("GH_TOKEN not found in environment variables")
package_url = f"git+https://{gh_token}@github.com/tolulope/speech-model-analysis.git"
# Use subprocess for better error handling
result = subprocess.run(
[sys.executable, "-m", "pip", "install", "--no-cache-dir", package_url],
capture_output=True,
text=True
)
if result.returncode != 0:
print("STDOUT:", result.stdout)
print("STDERR:", result.stderr)
raise RuntimeError(f"Failed to install package: {result.stderr}")
print("✓ Package installed successfully!")
# Clear import caches so Python recognizes the new package
importlib.invalidate_caches()
# Install the package first
install_private_package()
# # Install private package at startup
# print("Installing private package...")
# gh_token = os.environ.get("GH_TOKEN")
# if not gh_token:
# raise ValueError("GH_TOKEN not found in environment variables")
# package_url = f"git+https://{gh_token}@github.com/tolulope/speech-model-analysis.git"
# os.system(f"{sys.executable} -m pip install {package_url}")
# Now import from your private package
from speech_model_analysis import (
VoxCommunisPreprocessor,
MultiModelAnalyzer,
create_hubert_configs,
)
from speech_model_analysis.phoneme_manager import PHONEMES, index_to_phoneme
from speech_model_analysis.voxcommunis_preprocessing import VoxCommunisPreprocessor, create_hubert_configs
from speech_model_analysis.gradio_viz import ClusterVisualizer
from speech_model_analysis.enhanced_analysis import calculate_all_metrics
from speech_model_analysis.audio_player import ClusterAudioExplorer, create_audio_grid
from speech_model_analysis.embedding_projector_viz import EmbeddingProjectorViz
from speech_model_analysis.context_pooling import ContextConfig, ContextAwarePooler, ContextAwareAnalyzer
print("Private package loaded successfully!")
from huggingface_hub import hf_hub_download, snapshot_download, login
login(os.environ["HF_TOKEN"])
# Download the full repo snapshot to a local dir
OUTPUT_DIR = snapshot_download("tolulope/speech-model-analysis", repo_type="dataset")
def get_top_level_dirs(root):
root = Path(root)
return [d for d in root.iterdir() if d.is_dir()]
def load_analyzer_for_subdir(subdir_path):
return MultiModelAnalyzer(str(subdir_path))
def toggle_tsne_params(method):
visible = method == "t-SNE"
return [
gr.update(visible=visible),
gr.update(visible=visible),
gr.update(visible=visible)
]
def create_integrated_gradio_interface(analyzer: MultiModelAnalyzer):
"""
Create comprehensive Gradio interface with model comparison.
Args:
analyzer: MultiModelAnalyzer instance
"""
# Extract feature options (same as before)
all_manners = sorted(set(p.manner.name for p in PHONEMES.values()
if p.manner))
all_places = sorted(set(p.place.name for p in PHONEMES.values()
if p.place))
all_voicings = ['voiced', 'voiceless']
all_heights = ['high', 'mid', 'low']
all_backness = ['front', 'central', 'back']
model_names = analyzer.get_model_names()
with gr.Blocks(title="Discrete Token Analysis") as demo:
gr.Markdown("# Discrete Token Phoneme Analysis")
# gr.Markdown("Compare HuBERT models and analyze discrete representations")
with gr.Tabs():
# Tab 1: Model Comparison
with gr.Tab("Model Comparison"):
gr.Markdown("### Compare Clustering Quality Across Models")
with gr.Row():
# comparison_plot = gr.Plot(label="Metrics Comparison")
metrics_table = gr.Dataframe(label="Detailed Metrics")
refresh_comparison_btn = gr.Button("Refresh Comparison", variant="primary")
def update_comparison():
# fig = analyzer.create_comparison_plot()
df = analyzer.compare_metrics()
df = df.round(2)
return df
# refresh_comparison_btn.click(
# fn=update_comparison,
# outputs=[comparison_plot, metrics_table]
# )
# Initialize
demo.load(
fn=update_comparison,
# outputs=[comparison_plot, metrics_table]
outputs=[metrics_table]
)
# Tab 2: Single Model Analysis
"""
with gr.Tab("Single Model Analysis"):
with gr.Row():
with gr.Column(scale=1):
gr.Markdown("### Model & Filters")
model_selector = gr.Dropdown(
model_names,
value=model_names[0] if model_names else None,
label="Select Model"
)
color_by = gr.Radio(
['cluster', 'phone'],
value='cluster',
label="Color by"
)
gr.Markdown("#### Articulatory Filters")
manner_filter = gr.Dropdown(
all_manners,
multiselect=True,
label="Manner"
)
place_filter = gr.Dropdown(
all_places,
multiselect=True,
label="Place"
)
voicing_filter = gr.Dropdown(
all_voicings,
multiselect=True,
label="Voicing"
)
vowel_height_filter = gr.Dropdown(
all_heights,
multiselect=True,
label="Vowel Height"
)
vowel_backness_filter = gr.Dropdown(
all_backness,
multiselect=True,
label="Vowel Backness"
)
update_btn = gr.Button("Update Visualization", variant="primary")
with gr.Column(scale=2):
plot_output = gr.Plot(label="Cluster Visualization")
gr.Markdown("💡 **Tip**: Click on points to hear audio in the Audio Explorer tab!")
with gr.Row():
with gr.Column():
metrics_output = gr.Markdown()
with gr.Column():
confusion_output = gr.Plot(label="Confusion Matrix")
def update_single_model(model_name, color, manner, place, voicing, height, backness):
if not model_name:
return None, "Select a model", None
visualizer = analyzer.visualizers[model_name]
# Create scatter plot
fig = visualizer.create_scatter_plot(
color_by=color,
filter_manner=manner if manner else None,
filter_place=place if place else None,
filter_voicing=voicing if voicing else None,
filter_vowel_height=height if height else None,
filter_vowel_backness=backness if backness else None
)
# Calculate metrics
metrics = visualizer.calculate_metrics(
filter_manner=manner if manner else None,
filter_place=place if place else None,
filter_voicing=voicing if voicing else None,
filter_vowel_height=height if height else None,
filter_vowel_backness=backness if backness else None
)
# Create confusion matrix
confusion_fig = analyzer.create_confusion_heatmap(model_name)
return fig, metrics, confusion_fig
update_btn.click(
fn=update_single_model,
inputs=[model_selector, color_by, manner_filter, place_filter,
voicing_filter, vowel_height_filter, vowel_backness_filter],
outputs=[plot_output, metrics_output, confusion_output]
)
"""
# Tab 3: Audio Explorer
"""
with gr.Tab("Audio Explorer"):
gr.Markdown("### Listen to Cluster Samples")
gr.Markdown("Explore audio segments from clusters and phonemes")
with gr.Row():
with gr.Column():
audio_model_selector = gr.Dropdown(
model_names,
value=model_names[0] if model_names else None,
label="Select Model"
)
exploration_mode = gr.Radio(
['By Cluster', 'By Phoneme', 'Compare Phoneme Across Clusters'],
value='By Cluster',
label="Exploration Mode"
)
# Cluster mode inputs
with gr.Group(visible=True) as cluster_inputs:
cluster_id_audio = gr.Number(
label="Cluster ID",
value=0,
precision=0
)
n_cluster_samples = gr.Slider(
1, 10, value=5,
step=1,
label="Number of samples"
)
# Phoneme mode inputs
with gr.Group(visible=False) as phoneme_inputs:
phoneme_select = gr.Dropdown(
sorted(list(PHONEMES.keys())),
label="Select Phoneme",
value="æ"
)
n_phoneme_samples = gr.Slider(
1, 10, value=5,
step=1,
label="Number of samples"
)
# Compare mode inputs
with gr.Group(visible=False) as compare_inputs:
phoneme_compare = gr.Dropdown(
sorted(list(PHONEMES.keys())),
label="Phoneme to Compare",
value="æ"
)
n_per_cluster = gr.Slider(
1, 5, value=3,
step=1,
label="Samples per cluster"
)
play_audio_btn = gr.Button("🎵 Load Audio Samples", variant="primary")
with gr.Column(scale=2):
audio_output = gr.HTML(label="Audio Player")
audio_info = gr.Markdown()
# Toggle visibility based on mode
def update_visibility(mode):
return (
gr.update(visible=(mode == 'By Cluster')),
gr.update(visible=(mode == 'By Phoneme')),
gr.update(visible=(mode == 'Compare Phoneme Across Clusters'))
)
exploration_mode.change(
fn=update_visibility,
inputs=[exploration_mode],
outputs=[cluster_inputs, phoneme_inputs, compare_inputs]
)
def load_audio_samples(model_name, mode, cluster_id, n_cluster,
phoneme, n_phoneme, phoneme_cmp, n_per_clust):
if not model_name or model_name not in analyzer.audio_explorers:
return "<p>Audio not available for this model</p>", "No audio data loaded"
explorer = analyzer.audio_explorers[model_name]
try:
if mode == 'By Cluster':
samples = explorer.get_cluster_samples(
cluster_id=int(cluster_id),
n_samples=int(n_cluster)
)
info = f"### Cluster {cluster_id}\n\nShowing {len(samples)} samples"
elif mode == 'By Phoneme':
samples = explorer.get_phoneme_samples(
phoneme=phoneme,
n_samples=int(n_phoneme)
)
info = f"### Phoneme: {phoneme}\n\nShowing {len(samples)} samples"
else: # Compare mode
cluster_samples = explorer.compare_phoneme_in_clusters(
phoneme=phoneme_cmp,
n_per_cluster=int(n_per_clust)
)
# Flatten samples and add cluster headers
html = ""
info_lines = [f"### Phoneme: {phoneme_cmp} across clusters\n"]
for cluster_id, samps in sorted(cluster_samples.items()):
html += f'<h4>Cluster {cluster_id}</h4>'
html += create_audio_grid(samps, columns=3)
info_lines.append(f"- Cluster {cluster_id}: {len(samps)} samples")
return html, "\n".join(info_lines)
if not samples:
return "<p>No samples found</p>", "No matching samples"
html = create_audio_grid(samples, columns=3)
return html, info
except Exception as e:
return f"<p>Error loading audio: {str(e)}</p>", f"Error: {str(e)}"
play_audio_btn.click(
fn=load_audio_samples,
inputs=[audio_model_selector, exploration_mode,
cluster_id_audio, n_cluster_samples,
phoneme_select, n_phoneme_samples,
phoneme_compare, n_per_cluster],
outputs=[audio_output, audio_info]
)
"""
# Tab 4: Export & Analysis
"""
with gr.Tab("Export & Analysis"):
gr.Markdown("### Export Results")
with gr.Row():
export_model = gr.Dropdown(
model_names,
label="Select Model to Export"
)
export_format = gr.Radio(
['CSV', 'JSON', 'NPZ'],
value='CSV',
label="Format"
)
export_btn = gr.Button("Export Data", variant="primary")
export_output = gr.File(label="Download")
def export_data(model_name, format_type):
if not model_name:
return None
data = analyzer.models[model_name]
output_path = f"{model_name}_export.{format_type.lower()}"
if format_type == 'CSV':
df = pd.DataFrame({
'cluster': data['cluster_labels'],
'phoneme': data['phoneme_strings'],
'phone_idx': data['phone_labels']
})
df.to_csv(output_path, index=False)
elif format_type == 'JSON':
export_dict = {
'clusters': data['cluster_labels'].tolist(),
'phonemes': data['phoneme_strings'].tolist(),
'phone_indices': data['phone_labels'].tolist()
}
with open(output_path, 'w') as f:
json.dump(export_dict, f, indent=2)
else: # NPZ
np.savez(
output_path,
features=data['features'],
clusters=data['cluster_labels'],
phones=data['phone_labels']
)
return output_path
export_btn.click(
fn=export_data,
inputs=[export_model, export_format],
outputs=[export_output]
)
"""
# Tab 6: Context Pooling Analysis
"""
with gr.Tab("Context Pooling"):
gr.Markdown("### Coarticulation Analysis")
gr.Markdown("Pool phoneme embeddings by context to account for coarticulation effects")
with gr.Row():
with gr.Column(scale=1):
gr.Markdown("#### Pooling Configuration")
context_model = gr.Dropdown(
model_names,
value=model_names[0] if model_names else None,
label="Select Model"
)
enable_pooling = gr.Checkbox(
label="Enable Context Pooling",
value=False
)
left_context = gr.Slider(
0, 3, value=1, step=1,
label="Left Context (# phones)",
info="How many phones before target"
)
right_context = gr.Slider(
0, 3, value=1, step=1,
label="Right Context (# phones)",
info="How many phones after target"
)
pooling_method = gr.Radio(
choices=['mean', 'median', 'max'],
value='mean',
label="Pooling Method"
)
min_samples = gr.Slider(
1, 10, value=2, step=1,
label="Min Samples per Context",
info="Minimum instances to pool"
)
compute_pooling_btn = gr.Button("Apply Pooling", variant="primary")
pooling_status = gr.Markdown("")
gr.Markdown("#### Analyze Specific Phone")
phone_to_analyze = gr.Textbox(
label="Phoneme",
placeholder="æ",
value="æ"
)
analyze_phone_btn = gr.Button("Analyze Contexts")
with gr.Column(scale=2):
pooling_comparison = gr.Markdown("*Apply pooling to see comparison*")
context_analysis = gr.Markdown("*Analyze a phone to see contexts*")
# with gr.Row():
# pooled_plot = gr.Plot(label="Pooled Embeddings (UMAP)")
# Context pooling callbacks
def apply_context_pooling(model_name, enable, left, right, method, min_samp):
if not model_name or model_name not in analyzer.models:
return "Model not available", ""
data = analyzer.models[model_name]
if not enable:
# No pooling
metrics = calculate_all_metrics(
data['cluster_labels'],
data['phone_labels']
)
comparison = "### No Pooling (Baseline)\n\n"
comparison += f"- **Points**: {len(data['features'])}\n"
comparison += f"- **Cluster Purity**: {metrics['cluster_purity']:.3f}\n"
comparison += f"- **Phone Purity**: {metrics['phone_purity']:.3f}\n"
comparison += f"- **V-Measure**: {metrics['v_measure']:.3f}\n"
comparison += f"- **NMI**: {metrics.get('nmi', 0):.3f}\n"
return "No pooling applied (baseline)", comparison
try:
# Create context config
config = ContextConfig(
enabled=True,
left_context=int(left),
right_context=int(right),
pooling_method=method,
min_samples=int(min_samp)
)
# Create pooler
pooler = ContextAwarePooler(config)
# Pool embeddings
# Note: This assumes sequential data. In practice, you'd need
# utterance boundaries from preprocessing
phone_sequence = data['phone_labels'] # Simplified
pooled_embeddings, context_info = pooler.create_context_clusters(
data['features'],
data['phone_labels'],
phone_sequence,
utterance_boundaries=None # Would come from data
)
# Calculate metrics on pooled space
# Need to re-cluster or map clusters
from sklearn.cluster import KMeans
n_clusters = len(np.unique(data['cluster_labels']))
kmeans = KMeans(n_clusters=n_clusters, random_state=42)
pooled_clusters = kmeans.fit_predict(pooled_embeddings)
metrics = calculate_all_metrics(
pooled_clusters,
context_info['labels']
)
# Create comparison
comparison = f"### Context Pooling Results\n\n"
comparison += f"**Configuration**: L{left}R{right} ({method})\n\n"
comparison += f"- **Original Points**: {context_info['n_original']}\n"
comparison += f"- **Pooled Points**: {context_info['n_pooled']}\n"
comparison += f"- **Reduction**: {(1 - context_info['reduction_ratio'])*100:.1f}%\n\n"
comparison += f"**Metrics**:\n"
comparison += f"- **Cluster Purity**: {metrics['cluster_purity']:.3f}\n"
comparison += f"- **Phone Purity**: {metrics['phone_purity']:.3f}\n"
comparison += f"- **V-Measure**: {metrics['v_measure']:.3f}\n"
comparison += f"- **NMI**: {metrics.get('nmi', 0):.3f}\n"
status = f"Pooled {context_info['n_original']} → {context_info['n_pooled']} points"
return status, comparison
except Exception as e:
return f"Error: {str(e)}", ""
def analyze_phone_contexts(model_name, phone, left, right):
if not model_name or not phone:
return "*Enter phone to analyze*"
if model_name not in analyzer.models:
return "Model not available"
try:
data = analyzer.models[model_name]
# Create analyzer
ctx_analyzer = ContextAwareAnalyzer(
embeddings=data['features'],
phone_labels=data['phone_labels'],
phone_sequence=data['phone_labels'],
cluster_labels=data['cluster_labels']
)
# Analyze phone
analysis = ctx_analyzer.analyze_context_effects(phone, PHONEMES)
if 'error' in analysis:
return f"{analysis['error']}"
# Format output
output = f"### Analysis of /{phone}/\n\n"
output += f"- **Total occurrences**: {analysis['total_occurrences']}\n"
output += f"- **Unique contexts**: {analysis['unique_contexts']}\n\n"
output += f"**Most Common Contexts**:\n\n"
# Sort by count
contexts_sorted = sorted(
analysis['contexts'].items(),
key=lambda x: x[1]['count'],
reverse=True
)
for ctx_str, info in contexts_sorted[:10]:
output += f"- **{ctx_str}**: {info['count']} times"
if info['cluster_distribution']:
clusters = ", ".join(f"C{c}({cnt})"
for c, cnt in info['cluster_distribution'].items())
output += f" → {clusters}"
output += "\n"
if len(contexts_sorted) > 10:
output += f"\n*... and {len(contexts_sorted) - 10} more contexts*"
return output
except Exception as e:
return f"Error: {str(e)}"
# Connect callbacks
compute_pooling_btn.click(
fn=apply_context_pooling,
inputs=[context_model, enable_pooling, left_context, right_context,
pooling_method, min_samples],
outputs=[pooling_status, pooling_comparison]
)
analyze_phone_btn.click(
fn=analyze_phone_contexts,
inputs=[context_model, phone_to_analyze, left_context, right_context],
outputs=[context_analysis]
)
"""
# def get_choices(model_name, label_type):
# viz = analyzer.projector_vizs[model_name]
# df = pd.DataFrame(viz.labels)
# choices = [str(x) for x in df[label_type].unique()]
# print(choices)
# value = choices[0] if choices else None
# return choices, value
def get_choices(model_name, label_type):
viz = analyzer.projector_vizs[model_name]
df = pd.DataFrame(viz.labels)
if label_type == "phone":
choices = df["phone"].unique()
elif label_type == "cluster":
choices = df["cluster"].unique()
else:
choices = df["language"].unique()
return gr.update(
choices=[str(x) for x in choices], # MUST be a Python list of strings
value=str(choices[0]) # MUST be one of the choices
)
with gr.Tab("Embedding Projector"):
gr.Markdown("### TensorFlow Projector-Style 3D Visualization")
gr.Markdown("Interactive exploration similar to TensorFlow's Embedding Projector")
with gr.Row():
# Left sidebar
with gr.Column(scale=1):
gr.Markdown("#### Model & Projection")
projector_model = gr.Dropdown(
model_names,
value=model_names[0] if model_names else None,
label="Select Model"
)
projection_method = gr.Radio(
choices=['PCA', 't-SNE', 'UMAP'],
# choices=['PCA', 'UMAP'],
value='UMAP',
label="Projection Method"
)
tsne_perplexity = gr.Slider(5, 50, value=30, step=1, label="t-SNE Perplexity", visible=False)
tsne_lr = gr.Slider(10, 1000, value=200, step=10, label="t-SNE Learning Rate", visible=False)
tsne_iters = gr.Slider(250, 5000, value=1000, step=250, label="t-SNE Iterations", visible=False)
projection_method.change(
fn=toggle_tsne_params,
inputs=[projection_method],
outputs=[tsne_perplexity, tsne_lr, tsne_iters]
)
dimension = gr.Radio(
choices=['3D', '2D'],
value='3D',
label="Dimensions"
)
projector_color_by = gr.Radio(
# choices=['cluster', 'phone', 'language'],
choices=['cluster', 'language'],
value='cluster',
label="Color by"
)
compute_btn = gr.Button("Compute Projections", variant="primary")
compute_status = gr.Markdown("*Click to compute projections*")
gr.Markdown("#### Search & Highlight")
search_mode = gr.Radio(
choices=['By Label', 'By Features'],
value='By Label',
label="Search Mode"
)
phones = ["æ", "ɑ", "ə", "i", "u"]
clusters = [0, 1, 2, 3]
languages = ["hi", "pa"]
# Label search (simple)
with gr.Group(visible=True) as label_search_group:
# search_label_type = gr.Radio(
# choices=['phone', 'cluster', 'language'],
# value='phone',
# label="Search in"
# )
# search_term = gr.Textbox(
# label="Search term",
# placeholder="e.g., 'æ' or '5'"
# )
# search_term = gr.Dropdown(
# choices=list(phones), # initial choices
# value=phones[0], # initial value
# label="Search term",
# allow_custom_value=True
# )
# # Update dropdown choices when the label type changes
# # Update search_term whenever the label type changes
# search_label_type.change(
# fn=get_choices,
# inputs=[projector_model, search_label_type],
# outputs=[search_term, search_term] # first = choices, second = value
# )
search_label_type = gr.Radio(
choices=["phone", "cluster", "language"],
value="phone",
label="Search in"
)
search_term = gr.Dropdown(
choices=[str(x) for x in phones],
value=str(phones[0]),
label="Search term"
)
search_label_type.change(
fn=get_choices,
inputs=[projector_model, search_label_type],
outputs=search_term
)
# Feature search (advanced)
with gr.Group(visible=False) as feature_search_group:
search_manner = gr.Dropdown(
choices=['stop', 'fricative', 'nasal', 'approximant',
'affricate', 'tap/flap'],
multiselect=True,
label="Manner"
)
search_place = gr.Dropdown(
choices=['bilabial', 'labiodental', 'dental', 'alveolar',
'postalveolar', 'palatal', 'velar', 'uvular',
'pharyngeal', 'glottal'],
multiselect=True,
label="Place"
)
search_voicing = gr.Dropdown(
choices=['voiced', 'voiceless'],
multiselect=True,
label="Voicing"
)
search_vowel_height = gr.Dropdown(
choices=['high', 'mid', 'low'],
multiselect=True,
label="Vowel Height"
)
search_vowel_backness = gr.Dropdown(
choices=['front', 'central', 'back'],
multiselect=True,
label="Vowel Backness"
)
search_btn = gr.Button("🔍 Search")
# gr.Markdown("#### Nearest Neighbors")
# point_idx = gr.Number(
# label="Point index",
# value=0,
# precision=0
# )
# n_neighbors = gr.Slider(
# 1, 50, value=10,
# step=1,
# label="Number of neighbors"
# )
# show_nn_btn = gr.Button("Show Neighbors")
info_display = gr.Markdown("*Select a point or search*")
# Main visualization area
with gr.Column(scale=3):
projector_plot = gr.Plot(label="Embedding Space")
# with gr.Row():
# comparison_btn = gr.Button("Show Comparison View (PCA | t-SNE | UMAP)")
# comparison_plot = gr.Plot(label="Comparison", visible=False)
# Projector callbacks
def compute_projections(model_name, method, tsne_perplexity, tsne_lr, tsne_iters):
if not model_name or model_name not in analyzer.projector_vizs:
return "Model not available", None
viz = analyzer.projector_vizs[model_name]
try:
method_lower = method.lower()
viz.compute_projections(method_lower, tsne_perplexity, tsne_lr, tsne_iters)
# Create initial plot
proj_key = f"{method_lower}_3d"
fig = viz.create_3d_scatter(
projection=proj_key,
color_by='cluster'
)
return f"{method} projections computed!", fig
except Exception as e:
return f"Error: {str(e)}", None
def toggle_search_mode(mode):
"""Toggle between label and feature search."""
if mode == 'By Label':
return gr.update(visible=True), gr.update(visible=False)
else:
return gr.update(visible=False), gr.update(visible=True)
def update_projector_plot(model_name, method, dim, color_by_val, highlight_indices=None):
if not model_name or model_name not in analyzer.projector_vizs:
return None
viz = analyzer.projector_vizs[model_name]
proj_key = f"{method.lower()}_{dim.lower()}"
# Check if projection exists
if proj_key not in viz.projections:
return None
try:
if dim == '3D':
fig = viz.create_3d_scatter(
projection=proj_key,
color_by=color_by_val.lower(),
highlight_indices=highlight_indices
)
else:
fig = viz.create_2d_scatter(
projection=proj_key,
color_by=color_by_val.lower(),
highlight_indices=highlight_indices
)
return fig
except Exception as e:
print(f"Error creating plot: {e}")
return None
def search_points(model_name, search_mode, search_type, term, method, dim,
color_by_val, manner, place, voicing, vheight, vbackness):
if not model_name or model_name not in analyzer.projector_vizs:
return None, "Model not available"
viz = analyzer.projector_vizs[model_name]
if search_mode == 'By Label':
if not term:
fig = update_projector_plot(model_name, method, dim, color_by_val)
return fig, "No search term provided"
matches = viz.search_by_label(term, search_type.lower())
info = f"Found {len(matches)} matches for '{term}' in {search_type}"
else: # By Features
matches = viz.search_by_articulatory_features(
PHONEMES,
manner=manner if manner else None,
place=place if place else None,
voicing=voicing if voicing else None,
vowel_height=vheight if vheight else None,
vowel_backness=vbackness if vbackness else None
)
# Get summary
summary = viz.get_articulatory_summary(matches, PHONEMES)
info = f"Found {len(matches)} points matching features:\n\n"
if manner:
info += f"**Manner**: {', '.join(manner)}\n"
if place:
info += f"**Place**: {', '.join(place)}\n"
if voicing:
info += f"**Voicing**: {', '.join(voicing)}\n"
if vheight:
info += f"**Vowel Height**: {', '.join(vheight)}\n"
if vbackness:
info += f"**Vowel Backness**: {', '.join(vbackness)}\n"
if summary and len(matches) > 0:
info += f"\n**Distribution**:\n"
if summary.get('manner'):
info += "- Manner: " + ", ".join(
f"{k}({v})" for k, v in sorted(summary['manner'].items())
) + "\n"
if summary.get('place'):
info += "- Place: " + ", ".join(
f"{k}({v})" for k, v in sorted(summary['place'].items())
) + "\n"
fig = update_projector_plot(model_name, method, dim, color_by_val,
highlight_indices=matches)
if matches:
if len(matches) <= 10:
info += f"\n\nIndices: {matches}"
else:
info += f"\n\nSample indices: {matches[:10]}... (+{len(matches)-10} more)"
return fig, info
def show_neighbors(model_name, idx, n, method, dim, color_by_val):
if not model_name or model_name not in analyzer.projector_vizs:
return None, "Model not available"
viz = analyzer.projector_vizs[model_name]
if viz.nn_model is None:
viz.build_nn_index()
neighbors, distances = viz.find_nearest_neighbors(int(idx), int(n))
# Show with lines to neighbors
line_pairs = [(int(idx), int(nn)) for nn in neighbors]
proj_key = f"{method.lower()}_{dim.lower()}"
if proj_key not in viz.projections:
return None, "Projections not computed"
if dim == '3D':
fig = viz.create_3d_scatter(
projection=proj_key,
color_by=color_by_val.lower(),
highlight_indices=[int(idx)] + list(neighbors),
show_lines=True,
line_pairs=line_pairs
)
else:
fig = viz.create_2d_scatter(
projection=proj_key,
color_by=color_by_val.lower(),
highlight_indices=[int(idx)] + list(neighbors)
)
info = f"Point {idx} - Nearest {n} neighbors:\n\n"
for i, (nn_idx, dist) in enumerate(zip(neighbors, distances), 1):
info += f"{i}. Index {nn_idx} (distance: {dist:.3f})\n"
return fig, info
def show_comparison_view(model_name, color_by_val):
if not model_name or model_name not in analyzer.projector_vizs:
return gr.update(visible=False), None
viz = analyzer.projector_vizs[model_name]
# Ensure all projections exist
for method in ['pca', 'tsne', 'umap']:
if f'{method}_3d' not in viz.projections:
return gr.update(visible=False), None
fig = viz.create_comparison_view(color_by=color_by_val.lower())
return gr.update(visible=True), fig
# Connect callbacks
# compute_btn.click(
# fn=compute_projections,
# inputs=[projector_model, projection_method],
# outputs=[compute_status, projector_plot]
# )
compute_btn.click(
fn=compute_projections,
inputs=[projector_model, projection_method,
tsne_perplexity, tsne_lr, tsne_iters],
outputs=[compute_status, projector_plot]
)
search_mode.change(
fn=toggle_search_mode,
inputs=[search_mode],
outputs=[label_search_group, feature_search_group]
)
for component in [projection_method, dimension, projector_color_by]:
component.change(
fn=lambda m, meth, d, c: update_projector_plot(m, meth, d, c),
inputs=[projector_model, projection_method, dimension, projector_color_by],
outputs=[projector_plot]
)
search_btn.click(
fn=search_points,
inputs=[projector_model, search_mode, search_label_type, search_term,
projection_method, dimension, projector_color_by,
search_manner, search_place, search_voicing,
search_vowel_height, search_vowel_backness],
outputs=[projector_plot, info_display]
)
# show_nn_btn.click(
# fn=show_neighbors,
# inputs=[projector_model, point_idx, n_neighbors,
# projection_method, dimension, projector_color_by],
# outputs=[projector_plot, info_display]
# )
# comparison_btn.click(
# fn=lambda m, c: show_comparison_view(m, c),
# inputs=[projector_model, projector_color_by],
# outputs=[comparison_plot, comparison_plot]
# )
return demo
def create_root_interface(output_dir):
subdirs = get_top_level_dirs(output_dir)
# Load config
try:
with open("config.json") as f:
config = json.load(f)
selected = config.get("selected_dirs", [])
if selected:
subdirs = [d for d in subdirs if d.name in selected]
except FileNotFoundError:
pass # Load all if no config
with gr.Blocks() as demo:
gr.Markdown("## Discrete Token Phoneme Analysis")
with gr.Tabs():
for subdir in subdirs:
with gr.Tab(subdir.name):
analyzer = load_analyzer_for_subdir(subdir)
create_integrated_gradio_interface(analyzer)
return demo
if __name__ == "__main__":
# # Create analyzer
# analyzer = MultiModelAnalyzer(OUTPUT_DIR)
# # Create and launch interface
# demo = create_integrated_gradio_interface(analyzer)
demo = create_root_interface(OUTPUT_DIR)
demo.launch(
theme=gr.themes.Soft()
# server_port=args.port,
# share=True # Creates public link
)
# # demo = create_interface()
# # demo.launch()
|