Upload folder using huggingface_hub

.gitattributes

@@ -33,3 +33,167 @@ saved_model/**/* filter=lfs diff=lfs merge=lfs -text
 *.zip filter=lfs diff=lfs merge=lfs -text
 *.zst filter=lfs diff=lfs merge=lfs -text
 *tfevents* filter=lfs diff=lfs merge=lfs -text
+data/freesound/audio/fs_001_nature_Autumn_leaves_falling_on_forest_floor_(close,_loopable).wav filter=lfs diff=lfs merge=lfs -text
+data/freesound/audio/fs_002_nature_woodpeckers_Haanwijk_1206_PM_250306_1056.wav filter=lfs diff=lfs merge=lfs -text
+data/freesound/audio/fs_003_nature_Bird_-_Tawny_owl,_female.wav filter=lfs diff=lfs merge=lfs -text
+data/freesound/audio/fs_004_nature_Bird_Chirps.wav filter=lfs diff=lfs merge=lfs -text
+data/freesound/audio/fs_005_nature_Birds_Chirping_Busy.wav.wav filter=lfs diff=lfs merge=lfs -text
+data/freesound/audio/fs_006_nature_bird_in_oostduinen_17.wav.wav filter=lfs diff=lfs merge=lfs -text
+data/freesound/audio/fs_007_nature_wind_bushes_coast_903_PM_240311_0680.wav filter=lfs diff=lfs merge=lfs -text
+data/freesound/audio/fs_008_nature_estate_farm_noise_1153AM3_220509_0343.wav.wav filter=lfs diff=lfs merge=lfs -text
+data/freesound/audio/fs_009_nature_Howler_monkey_howling_in_the_evening_jungle.wav filter=lfs diff=lfs merge=lfs -text
+data/freesound/audio/fs_010_nature_forest_heavy_rain_loop.wav.wav filter=lfs diff=lfs merge=lfs -text
+data/freesound/audio/fs_011_nature_CardinalChirpingParabolicMP3VersionMay62013.mp3.wav filter=lfs diff=lfs merge=lfs -text
+data/freesound/audio/fs_012_nature_Superb_Lyrebird_II.wav filter=lfs diff=lfs merge=lfs -text
+data/freesound/audio/fs_013_nature_Night_Wind_Chimes.wav filter=lfs diff=lfs merge=lfs -text
+data/freesound/audio/fs_014_nature_Southern_Summer_Evening_Ambience_with_Crickets_4.wav filter=lfs diff=lfs merge=lfs -text
+data/freesound/audio/fs_015_nature_Li_River._Crickets_&_Frogs.wav filter=lfs diff=lfs merge=lfs -text
+data/freesound/audio/fs_016_nature_Thunder_sound_effect_2.wav filter=lfs diff=lfs merge=lfs -text
+data/freesound/audio/fs_017_nature_Rain_medium_-_falling_on_soft_surface.wav filter=lfs diff=lfs merge=lfs -text
+data/freesound/audio/fs_018_nature_Thunder_6.wav filter=lfs diff=lfs merge=lfs -text
+data/freesound/audio/fs_019_nature_Water_Flowing_into_Underground_Sinkhole.wav filter=lfs diff=lfs merge=lfs -text
+data/freesound/audio/fs_020_nature_Water_Flowing_Through_Close_Rapids_6.wav filter=lfs diff=lfs merge=lfs -text
+data/freesound/audio/fs_021_nature_Water_Flowing_Through_Close_Rapids_5.wav filter=lfs diff=lfs merge=lfs -text
+data/freesound/audio/fs_022_urban_CITY_AMBIENCE_for_CLEAN_LOOP.wav.wav filter=lfs diff=lfs merge=lfs -text
+data/freesound/audio/fs_023_urban_Kitchen_Ambience_During_Daytime_.wav.wav filter=lfs diff=lfs merge=lfs -text
+data/freesound/audio/fs_024_urban_police_in_Paris.wav.wav filter=lfs diff=lfs merge=lfs -text
+data/freesound/audio/fs_025_urban_BerlinGameScene.com_Crowd_Cheer_4.wav filter=lfs diff=lfs merge=lfs -text
+data/freesound/audio/fs_026_urban_Да_ладно.wav filter=lfs diff=lfs merge=lfs -text
+data/freesound/audio/fs_027_urban_Restaurant_5.mp3.wav filter=lfs diff=lfs merge=lfs -text
+data/freesound/audio/fs_028_urban_Industrial_grinder.wav filter=lfs diff=lfs merge=lfs -text
+data/freesound/audio/fs_029_urban_Continuous_mechanical_whir_.wav filter=lfs diff=lfs merge=lfs -text
+data/freesound/audio/fs_030_urban_Electric_saw.wav filter=lfs diff=lfs merge=lfs -text
+data/freesound/audio/fs_031_urban_Renault_Master_F3500_dCi135_Foley_Horn_Outside_Mono.wav.wav filter=lfs diff=lfs merge=lfs -text
+data/freesound/audio/fs_032_urban_Street_atmosphere_and_sounds_in_Rome.wav filter=lfs diff=lfs merge=lfs -text
+data/freesound/audio/fs_033_urban_Distant_horn_beeps_09.flac.wav filter=lfs diff=lfs merge=lfs -text
+data/freesound/audio/fs_034_urban_Underground_Train_Station_Ambience.wav filter=lfs diff=lfs merge=lfs -text
+data/freesound/audio/fs_035_urban_subwaystartsequence.wav.wav filter=lfs diff=lfs merge=lfs -text
+data/freesound/audio/fs_036_urban_Train,_Subway,_Pass_By,_Under_Bridge_Perspective.wav filter=lfs diff=lfs merge=lfs -text
+data/freesound/audio/fs_037_urban_SFX_Donkey_Cart.MP3.wav filter=lfs diff=lfs merge=lfs -text
+data/freesound/audio/fs_038_urban_people_talking_at_street_market.wav filter=lfs diff=lfs merge=lfs -text
+data/freesound/audio/fs_039_urban_Japan_Tokyo_Shinjuku_Street_Promoter_Yelling_City.wav.wav filter=lfs diff=lfs merge=lfs -text
+data/freesound/audio/fs_040_water_waves_close_coast_small_004.wav.wav filter=lfs diff=lfs merge=lfs -text
+data/freesound/audio/fs_041_water_wave_sand_beach_012.wav.wav filter=lfs diff=lfs merge=lfs -text
+data/freesound/audio/fs_042_water_waves_over_mall_shells_01_150515_00.wav.wav filter=lfs diff=lfs merge=lfs -text
+data/freesound/audio/fs_043_water_Rain_on_a_Plastic_Roof.wav filter=lfs diff=lfs merge=lfs -text
+data/freesound/audio/fs_044_water_Ambiance_Waterfall_Big_Skogafoss_Far_Loop_Stereo_02.wav.wav filter=lfs diff=lfs merge=lfs -text
+data/freesound/audio/fs_045_water_Water_Stream_1.wav filter=lfs diff=lfs merge=lfs -text
+data/freesound/audio/fs_046_water_Rushing_River_Loop.wav filter=lfs diff=lfs merge=lfs -text
+data/freesound/audio/fs_047_water_Water_Game_Theme_Loop_2.wav.wav filter=lfs diff=lfs merge=lfs -text
+data/freesound/audio/fs_048_water_waves_sand_beach_048.wav.wav filter=lfs diff=lfs merge=lfs -text
+data/freesound/audio/fs_049_water_waves_sand_beach_013.wav.wav filter=lfs diff=lfs merge=lfs -text
+data/freesound/audio/fs_050_water_Rain_Hitting_Puddle.wav filter=lfs diff=lfs merge=lfs -text
+data/freesound/audio/fs_051_water_EXT_RainInPuddle.wav.wav filter=lfs diff=lfs merge=lfs -text
+data/freesound/audio/fs_052_water_Heavy_Rain,_Outside.wav filter=lfs diff=lfs merge=lfs -text
+data/freesound/audio/fs_053_water_09_Stream,_Dirty_Sewer,_Dark,_Trickling,_Burbling,_Flowing,_Underwa.wav filter=lfs diff=lfs merge=lfs -text
+data/freesound/audio/fs_054_water_15_ca_pads.wav.wav filter=lfs diff=lfs merge=lfs -text
+data/freesound/audio/fs_055_water_12_Stream,_Muddy_Current,_Dark,_Trickling,_Drips,_Flowing,_Underwat.wav filter=lfs diff=lfs merge=lfs -text
+data/freesound/audio/fs_056_water_r18_babbling_brook.wav filter=lfs diff=lfs merge=lfs -text
+data/freesound/audio/fs_057_water_Running_water_flowing.wav filter=lfs diff=lfs merge=lfs -text
+data/freesound/audio/fs_058_water_Creek_02.wav filter=lfs diff=lfs merge=lfs -text
+data/freesound/audio/fs_059_water_2019-06-14_Nyord_Hafen_2_mono_niere.wav.wav filter=lfs diff=lfs merge=lfs -text
+data/freesound/audio/fs_060_water_Ambiance_marina_1154_AM_220801_0468.wav filter=lfs diff=lfs merge=lfs -text
+data/freesound/audio/fs_061_nature_Thunder_08.wav filter=lfs diff=lfs merge=lfs -text
+data/freesound/audio/fs_062_nature_Close_lightning_strike.wav filter=lfs diff=lfs merge=lfs -text
+data/freesound/audio/fs_063_nature_Rain_and_Thunder_3.wav filter=lfs diff=lfs merge=lfs -text
+data/freesound/audio/fs_064_nature_heavy_rain_outside.wav filter=lfs diff=lfs merge=lfs -text
+data/freesound/audio/fs_065_nature_Heavy_Rain_Thunder_UK_Cambridge_Short.wav.wav filter=lfs diff=lfs merge=lfs -text
+data/freesound/audio/fs_066_nature_Heavy_Rain.wav filter=lfs diff=lfs merge=lfs -text
+data/freesound/audio/fs_067_nature_room-tone_wind_rain_11_200216_0114.wav.wav filter=lfs diff=lfs merge=lfs -text
+data/freesound/audio/fs_068_nature_wind_gap_indoors_013_170305_1100.wav.wav filter=lfs diff=lfs merge=lfs -text
+data/freesound/audio/fs_069_nature_20190809.howling131.wav.wav filter=lfs diff=lfs merge=lfs -text
+data/freesound/audio/fs_070_nature_Tires_on_Gravel_Road_1.wav filter=lfs diff=lfs merge=lfs -text
+data/freesound/audio/fs_071_nature_novasoundhail260.wav.wav filter=lfs diff=lfs merge=lfs -text
+data/freesound/audio/fs_072_nature_Raekuuro_ikkunaa_vasten___Hail_hitting_against_a_window.wav filter=lfs diff=lfs merge=lfs -text
+data/freesound/audio/fs_073_nature_dogs_barking_mono_4824.wav.wav filter=lfs diff=lfs merge=lfs -text
+data/freesound/audio/fs_074_nature_dog_barking.wav.wav filter=lfs diff=lfs merge=lfs -text
+data/freesound/audio/fs_075_nature_Barking_Dog.wav filter=lfs diff=lfs merge=lfs -text
+data/freesound/audio/fs_076_nature_Cat_purring.wav filter=lfs diff=lfs merge=lfs -text
+data/freesound/audio/fs_077_nature_cat_eating.wav.wav filter=lfs diff=lfs merge=lfs -text
+data/freesound/audio/fs_078_nature_Cat_Purring.wav filter=lfs diff=lfs merge=lfs -text
+data/freesound/audio/fs_079_nature_estate_spring_NL_EU_1235PM_220509_0434.wav filter=lfs diff=lfs merge=lfs -text
+data/freesound/audio/fs_080_nature_farmfield_817_AM_NL_EU_220515_0345.wav filter=lfs diff=lfs merge=lfs -text
+data/freesound/audio/fs_081_nature_edge_forest_735AM_210221_0257.wav.wav filter=lfs diff=lfs merge=lfs -text
+data/freesound/audio/fs_082_nature_frogs_lake_night_spot_3.wav.wav filter=lfs diff=lfs merge=lfs -text
+data/freesound/audio/fs_083_nature_night-frogs2.wav.wav filter=lfs diff=lfs merge=lfs -text
+data/freesound/audio/fs_084_nature_200703101950PacificChorusFrogs1.wav.wav filter=lfs diff=lfs merge=lfs -text
+data/freesound/audio/fs_085_nature_bees_and_birds_170601_1190.wav.wav filter=lfs diff=lfs merge=lfs -text
+data/freesound/audio/fs_086_nature_Cicada_Insects_8_25_13_10_06_PM_1.wav.wav filter=lfs diff=lfs merge=lfs -text
+data/freesound/audio/fs_087_nature_Cicadas.wav filter=lfs diff=lfs merge=lfs -text
+data/freesound/audio/fs_088_urban_Typing_2.wav.wav filter=lfs diff=lfs merge=lfs -text
+data/freesound/audio/fs_089_urban_Busy_Office_No_People_Loop.wav filter=lfs diff=lfs merge=lfs -text
+data/freesound/audio/fs_090_urban_Keyboard_typing.WAV.wav filter=lfs diff=lfs merge=lfs -text
+data/freesound/audio/fs_091_urban_pan_fry2.wav.wav filter=lfs diff=lfs merge=lfs -text
+data/freesound/audio/fs_092_urban_meat_grill.wav.wav filter=lfs diff=lfs merge=lfs -text
+data/freesound/audio/fs_093_urban_boilingtomatosauce.wav.wav filter=lfs diff=lfs merge=lfs -text
+data/freesound/audio/fs_094_urban_Fire_-_Swooshes_-_Burst_Evolving_32.wav.wav filter=lfs diff=lfs merge=lfs -text
+data/freesound/audio/fs_095_urban_Fireplace.wav.wav filter=lfs diff=lfs merge=lfs -text
+data/freesound/audio/fs_096_urban_Fire_-_Chimney___Stove.wav filter=lfs diff=lfs merge=lfs -text
+data/freesound/audio/fs_097_urban_Clock_sound.mp3.wav filter=lfs diff=lfs merge=lfs -text
+data/freesound/audio/fs_098_urban_Clock_ticking.wav.wav filter=lfs diff=lfs merge=lfs -text
+data/freesound/audio/fs_099_urban_Clock_Ticking.wav filter=lfs diff=lfs merge=lfs -text
+data/processed/audio/city_people_traffic.wav filter=lfs diff=lfs merge=lfs -text
+data/processed/audio/forest_birds_wind_01.wav filter=lfs diff=lfs merge=lfs -text
+data/processed/audio/ocean_waves_wind_01.wav filter=lfs diff=lfs merge=lfs -text
+data/processed/audio/ocean_waves_wind_02.wav filter=lfs diff=lfs merge=lfs -text
+data/processed/audio/rain_city_traffic_01.wav filter=lfs diff=lfs merge=lfs -text
+data/processed/images/beach_waves_01.png filter=lfs diff=lfs merge=lfs -text
+data/processed/images/beach_waves_02.png filter=lfs diff=lfs merge=lfs -text
+data/processed/images/beach_waves_03.png filter=lfs diff=lfs merge=lfs -text
+data/processed/images/city01.png filter=lfs diff=lfs merge=lfs -text
+data/processed/images/city_02.png filter=lfs diff=lfs merge=lfs -text
+data/processed/images/city_neo_02.png filter=lfs diff=lfs merge=lfs -text
+data/processed/images/city_neon_01.png filter=lfs diff=lfs merge=lfs -text
+data/processed/images/forest_morning_fog_02.png filter=lfs diff=lfs merge=lfs -text
+data/processed/images/forest_mountain01.png filter=lfs diff=lfs merge=lfs -text
+data/wikimedia/images/beach_waves_01.png filter=lfs diff=lfs merge=lfs -text
+data/wikimedia/images/city01.png filter=lfs diff=lfs merge=lfs -text
+data/wikimedia/images/city_02.png filter=lfs diff=lfs merge=lfs -text
+data/wikimedia/images/city_neon_01.png filter=lfs diff=lfs merge=lfs -text
+data/wikimedia/images/forest_morning_fog_02.png filter=lfs diff=lfs merge=lfs -text
+data/wikimedia/images/wm_001_nature_01-พระที่นั่งคูหาคฤหาสน์.jpg filter=lfs diff=lfs merge=lfs -text
+data/wikimedia/images/wm_002_nature_1_pano_cuiping_yangshuo_2016.jpg filter=lfs diff=lfs merge=lfs -text
+data/wikimedia/images/wm_004_nature_2013_Cogden_Bridge.jpg filter=lfs diff=lfs merge=lfs -text
+data/wikimedia/images/wm_005_nature_2013_Rainbow_over_Washfold.jpg filter=lfs diff=lfs merge=lfs -text
+data/wikimedia/images/wm_006_nature_2014_Track_on_Fremington_Edge.jpg filter=lfs diff=lfs merge=lfs -text
+data/wikimedia/images/wm_007_nature_2014_Yorkshire_Dales_country_road_Swaledale_Askrigg.jpg filter=lfs diff=lfs merge=lfs -text
+data/wikimedia/images/wm_008_nature_2015_Ribblehead_Viaduct_1.jpg filter=lfs diff=lfs merge=lfs -text
+data/wikimedia/images/wm_009_nature_2015_Swaledale_from_Kisdon_Hill.jpg filter=lfs diff=lfs merge=lfs -text
+data/wikimedia/images/wm_010_nature_2018_-_Château_fort_de_Lourdes.jpg filter=lfs diff=lfs merge=lfs -text
+data/wikimedia/images/wm_011_nature_01_Calanche_Piana.jpg filter=lfs diff=lfs merge=lfs -text
+data/wikimedia/images/wm_012_nature_01_Gorges_du_Tarn_Roc_des_Hourtous.jpg filter=lfs diff=lfs merge=lfs -text
+data/wikimedia/images/wm_013_nature_1_lake_louise_pano_2019.jpg filter=lfs diff=lfs merge=lfs -text
+data/wikimedia/images/wm_014_nature_1_tianzishan_wulingyuan_zhangjiajie_2012.jpg filter=lfs diff=lfs merge=lfs -text
+data/wikimedia/images/wm_015_nature_1_zhangjiajie_huangshizhai_wulingyuan_panorama_2012.jpg filter=lfs diff=lfs merge=lfs -text
+data/wikimedia/images/wm_016_nature_150906-001_Wilpena_Pound_from_Moralana_Scenic_Drive_Pano_50pc.jpg filter=lfs diff=lfs merge=lfs -text
+data/wikimedia/images/wm_017_nature_20090719_Crkva_Gospa_od_Zdravlja_Kotor_Bay_Montenegro.jpg filter=lfs diff=lfs merge=lfs -text
+data/wikimedia/images/wm_019_nature_21-224-5054_NNP_Synevyr_RB_18.jpg filter=lfs diff=lfs merge=lfs -text
+data/wikimedia/images/wm_020_nature_80_-_Machu_Picchu_-_Juin_2009_-_edit.jpg filter=lfs diff=lfs merge=lfs -text
+data/wikimedia/images/wm_021_nature_2014_Park_w_Kłodzku.jpg filter=lfs diff=lfs merge=lfs -text
+data/wikimedia/images/wm_022_nature_Andiast-Breil-Brigels._24-09-2025._(d.j.b.)_08.jpg filter=lfs diff=lfs merge=lfs -text
+data/wikimedia/images/wm_023_nature_Ansberg_Blickrichtung_Süden_120324.jpg filter=lfs diff=lfs merge=lfs -text
+data/wikimedia/images/wm_024_nature_Baumsilhouetten_am_Uetliberg_mit_Novembernebel_und_Sonnenstrahlen.jpg filter=lfs diff=lfs merge=lfs -text
+data/wikimedia/images/wm_025_nature_Beech_and_ferns_in_Gullmarsskogen.jpg filter=lfs diff=lfs merge=lfs -text
+data/wikimedia/images/wm_026_urban_128_Balconies_of_1390_Market_Street,_San_Francisco.jpg filter=lfs diff=lfs merge=lfs -text
+data/wikimedia/images/wm_027_urban_14-02-02-straszburg-RalfR-113.jpg filter=lfs diff=lfs merge=lfs -text
+data/wikimedia/images/wm_028_urban_140626_Tierser_Alpl_Rosszähne.jpg filter=lfs diff=lfs merge=lfs -text
+data/wikimedia/images/wm_029_urban_Wroclaw_-_Hala_Stulecia_03.jpg filter=lfs diff=lfs merge=lfs -text
+data/wikimedia/images/wm_030_urban_2014_Bożków,_pałac_01.jpg filter=lfs diff=lfs merge=lfs -text
+data/wikimedia/images/wm_031_urban_2015_Browar_w_Radkowie.jpg filter=lfs diff=lfs merge=lfs -text
+data/wikimedia/images/wm_032_urban_2015_Wieża_mieszkalna_w_Żelaźnie_01.jpg filter=lfs diff=lfs merge=lfs -text
+data/wikimedia/images/wm_033_urban_2016_Pałac_w_Łomnicy_2.jpg filter=lfs diff=lfs merge=lfs -text
+data/wikimedia/images/wm_034_urban_2016_Pałac_w_Żelaźnie_1.jpg filter=lfs diff=lfs merge=lfs -text
+data/wikimedia/images/wm_035_urban_2016_Phnom_Penh,_Pałac_Królewski,_Preah_Tineang_Phhochani_(14).jpg filter=lfs diff=lfs merge=lfs -text
+data/wikimedia/images/wm_036_urban_2016_Phnom_Penh,_Pałac_Królewski,_Srebrna_Pagoda_(02).jpg filter=lfs diff=lfs merge=lfs -text
+data/wikimedia/images/wm_038_urban_15-10-28-Pont_Bac_de_Roda_Barcelona-RalfR-WMA_3105.jpg filter=lfs diff=lfs merge=lfs -text
+data/wikimedia/images/wm_040_urban_2014_Kłodzko,_most_gotycki.jpg filter=lfs diff=lfs merge=lfs -text
+data/wikimedia/images/wm_041_urban_2017_-_Київ_-_Світанок_над_Дніпром.jpg filter=lfs diff=lfs merge=lfs -text
+data/wikimedia/images/wm_042_urban_2024_Most_Żelazny_w_Kłodzku_(8),_powódź.jpg filter=lfs diff=lfs merge=lfs -text
+data/wikimedia/images/wm_043_urban_2024_Wiadukt_kolejowy_w_Lewinie_Kłodzkim_(11).jpg filter=lfs diff=lfs merge=lfs -text
+data/wikimedia/images/wm_044_urban_2Fi02473_Pont_national_US_Army_retouchée.jpg filter=lfs diff=lfs merge=lfs -text
+data/wikimedia/images/wm_045_urban_A_pier_at_a_campsite_during_sunset,_Sidney_Spit_(part_of_Gulf_Islands_National_Park_Reserve),_Sidney_Island,_British_Columbia,_Canada_20.jpg filter=lfs diff=lfs merge=lfs -text
+data/wikimedia/images/wm_046_urban_A5_Aarebruecke.jpg filter=lfs diff=lfs merge=lfs -text
+data/wikimedia/images/wm_047_urban_Anjarle_Bridge_and_Cows-fix.jpg filter=lfs diff=lfs merge=lfs -text
+data/wikimedia/images/wm_048_water_Acapulco_fishermen.jpg filter=lfs diff=lfs merge=lfs -text
+data/wikimedia/images/wm_049_water_Accès_plage,_Sainte-Marie,_Ré_island,_august_2015.jpg filter=lfs diff=lfs merge=lfs -text
+data/wikimedia/images/wm_050_water_Amphitheatre_Bay_after_a_storm,_Akamas_Peninsula,_Cyprus.jpg filter=lfs diff=lfs merge=lfs -text

README.md

@@ -1,19 +1,33 @@
 ---
-title: MultiModal Coherence AI
-emoji: 🚀
-colorFrom: red
-colorTo: red
-sdk: docker
-app_port: 8501
-tags:
-- streamlit
+title: Multimodal Coherence AI
+emoji: "\U0001f3a8"
+colorFrom: purple
+colorTo: pink
+sdk: streamlit
+sdk_version: "1.41.0"
+app_file: app.py
 pinned: false
-short_description: Streamlit template space
+license: mit
+short_description: Coherent text + image + audio with MSCI
 ---
 
-# Welcome to Streamlit!
+# Multimodal Coherence AI
 
-Edit `/src/streamlit_app.py` to customize this app to your heart's desire. :heart:
+Generate semantically coherent **text + image + audio** bundles and evaluate
+cross-modal alignment using the **Multimodal Semantic Coherence Index (MSCI)**.
 
-If you have any questions, checkout our [documentation](https://docs.streamlit.io) and [community
-forums](https://discuss.streamlit.io).
+## How it works
+
+1. **Text** — generated via HF Inference API
+2. **Image** — retrieved from a curated index using CLIP (ViT-B/32) embeddings
+3. **Audio** — retrieved from a curated index using CLAP (HTSAT-unfused) embeddings
+4. **MSCI** — computed as `0.45 * cos_sim(text, image) + 0.45 * cos_sim(text, audio)`
+
+## Research
+
+This demo accompanies a study evaluating multimodal semantic coherence across
+three research questions:
+
+- **RQ1**: Is MSCI sensitive to controlled semantic perturbations? (Supported, d > 2.0)
+- **RQ2**: Does structured planning improve cross-modal alignment? (Not supported)
+- **RQ3**: Does MSCI correlate with human coherence judgments? (Supported, rho = 0.379)

app.py

@@ -0,0 +1,584 @@
+"""
+Multimodal Coherence AI — Hugging Face Spaces Demo
+
+Live demonstration of multimodal generation + coherence evaluation.
+Enter a scene description and the system produces coherent text, image,
+and audio with real-time MSCI scoring.
+
+Pipeline: HF Inference API (text) + CLIP retrieval (image) + CLAP retrieval (audio)
+"""
+
+from __future__ import annotations
+
+import logging
+import os
+import sys
+import time
+from pathlib import Path
+from typing import Optional
+
+import streamlit as st
+
+# ---------------------------------------------------------------------------
+# Paths
+# ---------------------------------------------------------------------------
+PROJECT_ROOT = Path(__file__).resolve().parent
+sys.path.insert(0, str(PROJECT_ROOT))
+
+logger = logging.getLogger(__name__)
+
+IMAGE_SIM_THRESHOLD = 0.20
+AUDIO_SIM_THRESHOLD = 0.10
+
+# ---------------------------------------------------------------------------
+# Custom CSS
+# ---------------------------------------------------------------------------
+CUSTOM_CSS = """
+<style>
+@import url('https://fonts.googleapis.com/css2?family=Inter:wght@400;500;600;700;800&family=JetBrains+Mono:wght@500;700&display=swap');
+
+.block-container { padding-top: 1.2rem !important; max-width: 1200px; }
+html, body, [class*="css"] { font-family: 'Inter', -apple-system, sans-serif; }
+
+.hero-wrap { text-align: center; padding: 1.5rem 0 1rem; }
+.hero-title {
+    font-size: 2.6rem; font-weight: 800; letter-spacing: -0.03em;
+    background: linear-gradient(135deg, #818cf8 0%, #c084fc 50%, #f472b6 100%);
+    -webkit-background-clip: text; -webkit-text-fill-color: transparent;
+    margin-bottom: 0.35rem;
+}
+.hero-sub {
+    font-size: 1rem; color: #94a3b8; max-width: 600px;
+    margin: 0 auto; line-height: 1.6;
+}
+.hero-sub b { color: #c4b5fd; }
+
+.stTextArea textarea {
+    border-radius: 14px !important;
+    border: 1.5px solid rgba(129,140,248,0.25) !important;
+    font-size: 0.95rem !important; padding: 0.9rem 1rem !important;
+    transition: border-color 0.2s;
+}
+.stTextArea textarea:focus {
+    border-color: rgba(129,140,248,0.6) !important;
+    box-shadow: 0 0 0 3px rgba(129,140,248,0.1) !important;
+}
+
+.chip-row { display: flex; gap: 0.4rem; flex-wrap: wrap; align-items: center; padding-top: 0.3rem; }
+.chip {
+    display: inline-flex; align-items: center; gap: 0.3rem;
+    padding: 0.22rem 0.7rem; border-radius: 20px;
+    font-size: 0.7rem; font-weight: 600; letter-spacing: 0.03em;
+}
+.chip-purple { background: rgba(129,140,248,0.14); color: #a5b4fc; }
+.chip-green  { background: rgba(52,211,153,0.14);  color: #6ee7b7; }
+.chip-dot { width: 6px; height: 6px; border-radius: 50%; }
+.chip-dot-purple { background: #818cf8; }
+.chip-dot-green  { background: #34d399; }
+
+.scores-grid {
+    display: grid; grid-template-columns: repeat(4, 1fr);
+    gap: 0.75rem; margin: 0.5rem 0 0.3rem;
+}
+@media (max-width: 768px) { .scores-grid { grid-template-columns: repeat(2, 1fr); } }
+.sc {
+    border-radius: 16px; padding: 1.1rem 0.8rem; text-align: center;
+    border: 1px solid rgba(255,255,255,0.06);
+    background: rgba(255,255,255,0.02);
+    backdrop-filter: blur(10px);
+    position: relative; overflow: hidden;
+}
+.sc::before {
+    content: ''; position: absolute; top: 0; left: 0; right: 0; height: 3px;
+    border-radius: 16px 16px 0 0;
+}
+.sc-high::before { background: linear-gradient(90deg, #10b981, #34d399); }
+.sc-mid::before  { background: linear-gradient(90deg, #f59e0b, #fbbf24); }
+.sc-low::before  { background: linear-gradient(90deg, #ef4444, #fb7185); }
+.sc-class::before { background: linear-gradient(90deg, #818cf8, #c084fc); }
+.sc-lbl {
+    font-size: 0.65rem; text-transform: uppercase; letter-spacing: 0.1em;
+    color: #64748b; margin-bottom: 0.4rem; font-weight: 600;
+}
+.sc-val {
+    font-size: 1.9rem; font-weight: 700; line-height: 1.1;
+    font-family: 'JetBrains Mono', monospace;
+}
+.sc-high .sc-val { color: #34d399; }
+.sc-mid  .sc-val { color: #fbbf24; }
+.sc-low  .sc-val { color: #fb7185; }
+.sc-class .sc-val { font-size: 1.15rem; font-family: 'Inter', sans-serif; color: #c4b5fd; }
+.sc-badge {
+    display: inline-block; margin-top: 0.35rem; padding: 0.15rem 0.55rem;
+    border-radius: 20px; font-size: 0.6rem; font-weight: 700;
+    text-transform: uppercase; letter-spacing: 0.07em;
+}
+.sc-high .sc-badge { background: rgba(52,211,153,0.12); color: #34d399; }
+.sc-mid  .sc-badge { background: rgba(251,191,36,0.12); color: #fbbf24; }
+.sc-low  .sc-badge { background: rgba(251,113,133,0.12); color: #fb7185; }
+.sc-class .sc-badge { background: rgba(196,181,253,0.12); color: #c4b5fd; }
+
+.sec-label {
+    font-size: 0.7rem; text-transform: uppercase; letter-spacing: 0.12em;
+    font-weight: 700; margin-bottom: 0.6rem; padding-bottom: 0.35rem;
+    border-bottom: 2px solid rgba(129,140,248,0.15); color: #818cf8;
+}
+.text-card {
+    border-radius: 14px; padding: 1.1rem 1.2rem;
+    background: rgba(255,255,255,0.02); border: 1px solid rgba(255,255,255,0.06);
+    font-size: 0.9rem; line-height: 1.75; color: #cbd5e1;
+}
+.timing {
+    display: flex; gap: 0.5rem; flex-wrap: wrap; align-items: center;
+    padding: 0.4rem 0.8rem; border-radius: 10px;
+    background: rgba(255,255,255,0.02); border: 1px solid rgba(255,255,255,0.04);
+    font-size: 0.72rem; color: #64748b; margin: 0.4rem 0;
+}
+.timing span { white-space: nowrap; }
+.timing .t-total { color: #a5b4fc; font-weight: 700; }
+.timing .t-sep { color: rgba(255,255,255,0.08); }
+
+.warn-banner {
+    border-radius: 12px; padding: 0.7rem 1rem; margin-bottom: 0.6rem;
+    border-left: 3px solid #fbbf24; font-size: 0.82rem; color: #fcd34d;
+    background: rgba(251,191,36,0.05);
+}
+.warn-banner b { color: #fde68a; }
+
+.sb { margin: 0.35rem 0; }
+.sb-top { display: flex; justify-content: space-between; font-size: 0.68rem; color: #64748b; margin-bottom: 0.15rem; }
+.sb-top .sb-v { font-family: 'JetBrains Mono', monospace; font-weight: 600; }
+.sb-track { height: 5px; border-radius: 3px; background: rgba(255,255,255,0.05); overflow: hidden; }
+.sb-fill { height: 100%; border-radius: 3px; }
+.sbf-g { background: linear-gradient(90deg, #10b981, #34d399); }
+.sbf-y { background: linear-gradient(90deg, #f59e0b, #fbbf24); }
+.sbf-r { background: linear-gradient(90deg, #ef4444, #fb7185); }
+
+.welcome { text-align: center; padding: 4rem 2rem; color: #475569; }
+.welcome-icons { font-size: 3.5rem; margin-bottom: 0.8rem; letter-spacing: 0.3rem; }
+.welcome-text { font-size: 1.05rem; color: #64748b; }
+.welcome-hint { font-size: 0.82rem; color: #475569; margin-top: 0.3rem; }
+
+section[data-testid="stSidebar"] > div:first-child { padding-top: 1.2rem; }
+.sidebar-info {
+    font-size: 0.72rem; color: #64748b; line-height: 1.6;
+    padding: 0.8rem; border-radius: 10px;
+    background: rgba(255,255,255,0.02); border: 1px solid rgba(255,255,255,0.04);
+}
+.sidebar-info b { color: #94a3b8; }
+</style>
+"""
+
+# ---------------------------------------------------------------------------
+# Example prompts
+# ---------------------------------------------------------------------------
+EXAMPLE_PROMPTS = {
+    "Nature": [
+        "A peaceful forest at dawn with birdsong and morning mist",
+        "A field of golden wheat under a warm summer sunset",
+        "A dense jungle with exotic birds calling from the canopy",
+    ],
+    "Urban": [
+        "A bustling city street at night with neon lights and traffic",
+        "A quiet alley in an old town with distant footsteps echoing",
+        "A cafe terrace on a busy boulevard with clinking glasses",
+    ],
+    "Water": [
+        "Ocean waves crashing on a sandy beach at sunset",
+        "Rain falling on a pond with ripples spreading across the surface",
+        "A mountain stream flowing over rocks through a pine forest",
+    ],
+    "Mixed": [
+        "A lighthouse on a cliff during a thunderstorm at night",
+        "A bonfire on a beach with waves and guitar music at night",
+        "A train passing through countryside with distant church bells",
+    ],
+}
+DOMAIN_ICONS = {"nature": "\U0001f33f", "urban": "\U0001f3d9\ufe0f", "water": "\U0001f30a", "mixed": "\U0001f310", "other": "\U0001f4cd"}
+
+
+# ---------------------------------------------------------------------------
+# Cached model loading
+# ---------------------------------------------------------------------------
+
+@st.cache_resource
+def load_coherence_engine():
+    from src.coherence.coherence_engine import CoherenceEngine
+    return CoherenceEngine(target_dim=512)
+
+@st.cache_resource
+def load_image_retriever():
+    from src.generators.image.generator_improved import ImprovedImageRetrievalGenerator
+    return ImprovedImageRetrievalGenerator(index_path="data/embeddings/image_index.npz", min_similarity=0.20)
+
+@st.cache_resource
+def load_audio_retriever():
+    from src.generators.audio.retrieval import AudioRetrievalGenerator
+    return AudioRetrievalGenerator(index_path="data/embeddings/audio_index.npz", min_similarity=0.10)
+
+@st.cache_resource
+def get_inference_client():
+    from huggingface_hub import InferenceClient
+    token = os.environ.get("HF_TOKEN")
+    return InferenceClient(token=token)
+
+
+# ---------------------------------------------------------------------------
+# Generation / retrieval functions
+# ---------------------------------------------------------------------------
+
+def gen_text_hf(prompt: str) -> dict:
+    """Generate descriptive text using HF Inference API."""
+    system_prompt = (
+        "You are a concise descriptive writer. "
+        "Write a literal description of the scene in 3 to 5 natural sentences. "
+        "No bullet points, no numbered lists, no meta commentary. "
+        "Focus on concrete visual details AND the likely audio ambience."
+    )
+    try:
+        client = get_inference_client()
+        response = client.chat_completion(
+            messages=[
+                {"role": "system", "content": system_prompt},
+                {"role": "user", "content": f"Describe this scene: {prompt}"},
+            ],
+            max_tokens=250,
+        )
+        text = response.choices[0].message.content.strip()
+        if not text:
+            raise ValueError("Empty response")
+        return {"text": text, "image_prompt": prompt, "audio_prompt": prompt, "plan": None}
+    except Exception as e:
+        logger.warning("HF Inference API failed: %s — using prompt as text", e)
+        return {"text": prompt, "image_prompt": prompt, "audio_prompt": prompt, "plan": None}
+
+
+def retrieve_image(prompt: str) -> dict:
+    r = load_image_retriever().retrieve(prompt)
+    return {
+        "path": r.image_path, "similarity": r.similarity, "domain": r.domain,
+        "failed": r.retrieval_failed, "top_5": r.top_5, "backend": "retrieval",
+    }
+
+
+def retrieve_audio(prompt: str) -> dict:
+    r = load_audio_retriever().retrieve(prompt)
+    return {
+        "path": r.audio_path, "similarity": r.similarity,
+        "failed": r.retrieval_failed, "top_5": r.top_5, "backend": "retrieval",
+    }
+
+
+def eval_coherence(text: str, image_path: str, audio_path: str) -> dict:
+    return load_coherence_engine().evaluate(text=text, image_path=image_path, audio_path=audio_path)
+
+
+# ---------------------------------------------------------------------------
+# HTML helpers
+# ---------------------------------------------------------------------------
+
+def _sc_cls(v: Optional[float]) -> str:
+    if v is None: return ""
+    if v >= 0.45: return "sc-high"
+    if v >= 0.30: return "sc-mid"
+    return "sc-low"
+
+def _sc_badge(v: Optional[float]) -> str:
+    if v is None: return ""
+    if v >= 0.45: return "High"
+    if v >= 0.30: return "Moderate"
+    return "Low"
+
+def score_card_html(label: str, value: Optional[float], is_class: bool = False) -> str:
+    if is_class:
+        badge_text = _sc_badge(value) or "N/A"
+        val_display = f"{badge_text} Coherence"
+        badge_html = f'<div class="sc-badge">MSCI {value:.3f}</div>' if value is not None else ""
+        return (f'<div class="sc sc-class"><div class="sc-lbl">{label}</div>'
+                f'<div class="sc-val">{val_display}</div>{badge_html}</div>')
+    cls = _sc_cls(value)
+    val_str = f"{value:.4f}" if value is not None else "\u2014"
+    badge = _sc_badge(value)
+    badge_html = f'<div class="sc-badge">{badge}</div>' if badge else ""
+    return (f'<div class="sc {cls}"><div class="sc-lbl">{label}</div>'
+            f'<div class="sc-val">{val_str}</div>{badge_html}</div>')
+
+def sim_bar_html(name: str, val: float, mx: float = 0.6) -> str:
+    pct = min(val / mx * 100, 100)
+    cls = "sbf-g" if val >= 0.35 else ("sbf-y" if val >= 0.20 else "sbf-r")
+    return (f'<div class="sb"><div class="sb-top"><span>{name}</span>'
+            f'<span class="sb-v">{val:.4f}</span></div>'
+            f'<div class="sb-track"><div class="sb-fill {cls}" style="width:{pct}%"></div></div></div>')
+
+
+# ---------------------------------------------------------------------------
+# Main
+# ---------------------------------------------------------------------------
+
+def main():
+    st.set_page_config(
+        page_title="Multimodal Coherence AI",
+        page_icon="\U0001f3a8",
+        layout="wide",
+        initial_sidebar_state="expanded",
+    )
+    st.markdown(CUSTOM_CSS, unsafe_allow_html=True)
+
+    # Hero
+    st.markdown(
+        '<div class="hero-wrap">'
+        '<div class="hero-title">Multimodal Coherence AI</div>'
+        '<div class="hero-sub">Generate semantically coherent <b>text + image + audio</b> bundles '
+        'and evaluate cross-modal alignment with the <b>MSCI</b> metric.</div>'
+        '</div>', unsafe_allow_html=True)
+
+    # Sidebar
+    with st.sidebar:
+        st.markdown("#### Examples")
+        for dname, prompts in EXAMPLE_PROMPTS.items():
+            icon = DOMAIN_ICONS.get(dname.lower(), "\U0001f4cd")
+            with st.expander(f"{icon} {dname}"):
+                for p in prompts:
+                    if st.button(p, key=f"ex_{hash(p)}", use_container_width=True):
+                        st.session_state["prompt_input"] = p
+
+        st.divider()
+        st.markdown(
+            '<div class="sidebar-info">'
+            '<b>Text</b> HF Inference API<br>'
+            '<b>Image</b> CLIP retrieval (57 images)<br>'
+            '<b>Audio</b> CLAP retrieval (104 clips)<br><br>'
+            '<b>Metric</b> MSCI = 0.45 &times; s<sub>t,i</sub> + 0.45 &times; s<sub>t,a</sub><br><br>'
+            '<b>Models</b><br>'
+            'CLIP ViT-B/32 (text-image)<br>'
+            'CLAP HTSAT-unfused (text-audio)'
+            '</div>', unsafe_allow_html=True)
+
+    # Prompt input
+    default_prompt = st.session_state.get("prompt_input", "")
+    prompt = st.text_area(
+        "Scene", value=default_prompt, height=80,
+        placeholder="Describe a scene... e.g., 'A peaceful forest at dawn with birdsong and morning mist'",
+        label_visibility="collapsed",
+    )
+
+    # Button + chips
+    bc1, bc2 = st.columns([1, 3])
+    with bc1:
+        go = st.button("Generate Bundle", type="primary", use_container_width=True, disabled=not prompt.strip())
+    with bc2:
+        st.markdown(
+            '<div class="chip-row">'
+            '<span class="chip chip-purple"><span class="chip-dot chip-dot-purple"></span>Retrieval</span>'
+            '<span class="chip chip-green"><span class="chip-dot chip-dot-green"></span>CLIP + CLAP</span>'
+            '</div>', unsafe_allow_html=True)
+
+    # Welcome state
+    if not go and "last_result" not in st.session_state:
+        st.markdown(
+            '<div class="welcome">'
+            '<div class="welcome-icons">\U0001f3a8  \U0001f5bc\ufe0f  \U0001f50a</div>'
+            '<div class="welcome-text">Enter a scene description and click <b>Generate Bundle</b></div>'
+            '<div class="welcome-hint">or pick an example from the sidebar</div>'
+            '</div>', unsafe_allow_html=True)
+        return
+
+    if go and prompt.strip():
+        st.session_state["last_result"] = run_pipeline(prompt.strip())
+
+    if "last_result" in st.session_state:
+        show_results(st.session_state["last_result"])
+
+
+# ---------------------------------------------------------------------------
+# Pipeline
+# ---------------------------------------------------------------------------
+
+def run_pipeline(prompt: str) -> dict:
+    R: dict = {}
+    t_all = time.time()
+
+    # 1) Text
+    with st.status("Generating text...", expanded=True) as s:
+        t0 = time.time()
+        try:
+            R["text"] = gen_text_hf(prompt)
+            R["t_text"] = time.time() - t0
+            s.update(label=f"Text ready ({R['t_text']:.1f}s)", state="complete")
+        except Exception as e:
+            s.update(label=f"Text failed: {e}", state="error")
+            R["text"] = {"text": prompt, "image_prompt": prompt, "audio_prompt": prompt}
+            R["t_text"] = time.time() - t0
+
+    ip = R["text"].get("image_prompt", prompt)
+    ap = R["text"].get("audio_prompt", prompt)
+
+    # 2) Image retrieval
+    with st.status("Retrieving image...", expanded=True) as s:
+        t0 = time.time()
+        try:
+            R["image"] = retrieve_image(ip)
+            R["t_img"] = time.time() - t0
+            f = R["image"].get("failed", False)
+            lbl = f"Image retrieved (sim={R['image']['similarity']:.3f}, {R['t_img']:.1f}s)"
+            if f:
+                lbl += " \u2014 below threshold"
+            s.update(label=lbl, state="complete" if not f else "error")
+        except Exception as e:
+            s.update(label=f"Image failed: {e}", state="error")
+            R["image"] = None
+            R["t_img"] = time.time() - t0
+
+    # 3) Audio retrieval
+    with st.status("Retrieving audio...", expanded=True) as s:
+        t0 = time.time()
+        try:
+            R["audio"] = retrieve_audio(ap)
+            R["t_aud"] = time.time() - t0
+            f = R["audio"].get("failed", False)
+            lbl = f"Audio retrieved (sim={R['audio']['similarity']:.3f}, {R['t_aud']:.1f}s)"
+            if f:
+                lbl += " \u2014 below threshold"
+            s.update(label=lbl, state="complete" if not f else "error")
+        except Exception as e:
+            s.update(label=f"Audio failed: {e}", state="error")
+            R["audio"] = None
+            R["t_aud"] = time.time() - t0
+
+    # 4) Coherence evaluation
+    with st.status("Evaluating coherence...", expanded=True) as s:
+        t0 = time.time()
+        try:
+            imgp = R.get("image", {}).get("path") if R.get("image") else None
+            audp = R.get("audio", {}).get("path") if R.get("audio") else None
+            R["coherence"] = eval_coherence(R["text"]["text"], imgp, audp)
+            R["t_eval"] = time.time() - t0
+            msci = R["coherence"].get("scores", {}).get("msci")
+            s.update(label=f"MSCI = {msci:.4f} ({R['t_eval']:.1f}s)", state="complete")
+        except Exception as e:
+            s.update(label=f"Eval failed: {e}", state="error")
+            R["coherence"] = None
+            R["t_eval"] = time.time() - t0
+
+    R["t_total"] = time.time() - t_all
+    R["prompt"] = prompt
+    return R
+
+
+# ---------------------------------------------------------------------------
+# Results display
+# ---------------------------------------------------------------------------
+
+def show_results(R: dict):
+    coh = R.get("coherence")
+    sc = coh.get("scores", {}) if coh else {}
+    msci = sc.get("msci")
+    st_i = sc.get("st_i")
+    st_a = sc.get("st_a")
+
+    # Score cards
+    st.markdown('<div class="sec-label">Coherence Scores</div>', unsafe_allow_html=True)
+    cards = (
+        score_card_html("MSCI (Overall)", msci)
+        + score_card_html("Text \u2192 Image", st_i)
+        + score_card_html("Text \u2192 Audio", st_a)
+        + score_card_html("Classification", msci, is_class=True)
+    )
+    st.markdown(f'<div class="scores-grid">{cards}</div>', unsafe_allow_html=True)
+
+    # Timing strip
+    tt = R.get("t_total", 0)
+    sep = '<span class="t-sep">|</span>'
+    st.markdown(
+        f'<div class="timing">'
+        f'<span class="t-total">Total {tt:.1f}s</span>{sep}'
+        f'<span>Text {R.get("t_text", 0):.1f}s</span>{sep}'
+        f'<span>Image {R.get("t_img", 0):.1f}s</span>{sep}'
+        f'<span>Audio {R.get("t_aud", 0):.1f}s</span>{sep}'
+        f'<span>Eval {R.get("t_eval", 0):.1f}s</span>'
+        f'</div>', unsafe_allow_html=True)
+
+    st.markdown("---")
+
+    # Three columns: text | image | audio
+    ct, ci, ca = st.columns([1.15, 1, 0.85])
+
+    with ct:
+        st.markdown('<div class="sec-label">Generated Text</div>', unsafe_allow_html=True)
+        txt = R.get("text", {}).get("text", "")
+        st.markdown(f'<div class="text-card">{txt}</div>', unsafe_allow_html=True)
+
+    with ci:
+        st.markdown('<div class="sec-label">Image</div>', unsafe_allow_html=True)
+        ii = R.get("image")
+        if ii and ii.get("path"):
+            ip = Path(ii["path"])
+            failed = ii.get("failed", False)
+            sim = ii.get("similarity")
+
+            if failed:
+                st.markdown(
+                    f'<div class="warn-banner"><b>Below threshold</b> '
+                    f'(sim={sim:.3f} &lt; {IMAGE_SIM_THRESHOLD}) '
+                    f'\u2014 best match from index.</div>',
+                    unsafe_allow_html=True)
+
+            if ip.exists():
+                st.image(str(ip), use_container_width=True)
+                dom = ii.get("domain", "other")
+                ic = DOMAIN_ICONS.get(dom, "\U0001f4cd")
+                st.caption(f"{ic} {dom} \u00b7 sim **{sim:.3f}** \u00b7 {ip.name}")
+        else:
+            st.info("No image.")
+
+    with ca:
+        st.markdown('<div class="sec-label">Audio</div>', unsafe_allow_html=True)
+        ai = R.get("audio")
+        if ai and ai.get("path"):
+            ap = Path(ai["path"])
+            sim = ai.get("similarity")
+            failed = ai.get("failed", False)
+
+            if failed:
+                st.markdown(
+                    f'<div class="warn-banner"><b>Below threshold</b> '
+                    f'(sim={sim:.3f} &lt; {AUDIO_SIM_THRESHOLD}).</div>',
+                    unsafe_allow_html=True)
+
+            if ap.exists():
+                st.audio(str(ap))
+                st.caption(f"sim **{sim:.3f}** \u00b7 {ap.name}")
+        else:
+            st.info("No audio.")
+
+    st.markdown("---")
+
+    # Expandable details
+    with st.expander("Retrieval Details"):
+        r1, r2 = st.columns(2)
+        with r1:
+            ii = R.get("image")
+            if ii and ii.get("top_5"):
+                st.markdown("**Image \u2014 Top 5 candidates**")
+                bars = "".join(sim_bar_html(n, s) for n, s in ii["top_5"])
+                st.markdown(bars, unsafe_allow_html=True)
+            else:
+                st.write("No image data.")
+        with r2:
+            ai = R.get("audio")
+            if ai and ai.get("top_5"):
+                st.markdown("**Audio \u2014 Top 5 candidates**")
+                bars = "".join(sim_bar_html(n, s) for n, s in ai["top_5"])
+                st.markdown(bars, unsafe_allow_html=True)
+            else:
+                st.write("No audio data.")
+
+    with st.expander("Full Coherence Report"):
+        if coh:
+            st.json(coh)
+        else:
+            st.write("No data.")
+
+
+if __name__ == "__main__":
+    main()

data/embeddings/audio_index.npz

@@ -0,0 +1,3 @@
+version https://git-lfs.github.com/spec/v1
+oid sha256:5d3f42fa5191e4f226284ea44ded2088e2d536e866c5ceca72f3a97b92084b02
+size 201463

data/embeddings/image_index.npz

@@ -0,0 +1,3 @@
+version https://git-lfs.github.com/spec/v1
+oid sha256:5d1f400311b4c7a1f23767d1e4e589bc1734db734e261c3f88235bfcdede72c1
+size 111366

data/freesound/audio/fs_001_nature_Autumn_leaves_falling_on_forest_floor_(close,_loopable).wav

@@ -0,0 +1,3 @@
+version https://git-lfs.github.com/spec/v1
+oid sha256:8a19139048193129622a4a260ac867ce54bf6faa3cbea57b2156864a5b05e0e3
+size 904034

data/freesound/audio/fs_002_nature_woodpeckers_Haanwijk_1206_PM_250306_1056.wav

@@ -0,0 +1,3 @@
+version https://git-lfs.github.com/spec/v1
+oid sha256:e94c8382bde74a377c8ee9eda21c6e30efd0ea88085dd6e291b188c252b3749a
+size 1965922

data/freesound/audio/fs_003_nature_Bird_-_Tawny_owl,_female.wav

@@ -0,0 +1,3 @@
+version https://git-lfs.github.com/spec/v1
+oid sha256:2a74ec22281db411cbaca31c06e218e4521121ab2235d2ed8ed0898e2aba9d6a
+size 1239808

data/freesound/audio/fs_004_nature_Bird_Chirps.wav

@@ -0,0 +1,3 @@
+version https://git-lfs.github.com/spec/v1
+oid sha256:67e4db8b8f5019aa421247c28dca4d07fd6ca49825486dc0e43f9199db5532fc
+size 1992088

data/freesound/audio/fs_005_nature_Birds_Chirping_Busy.wav.wav

@@ -0,0 +1,3 @@
+version https://git-lfs.github.com/spec/v1
+oid sha256:7cb9914242fbb4cfc6b1f15f5684b4d6c2a065dd07c8988db28dc57dab00e87e
+size 2196300

data/freesound/audio/fs_006_nature_bird_in_oostduinen_17.wav.wav

@@ -0,0 +1,3 @@
+version https://git-lfs.github.com/spec/v1
+oid sha256:ab66fccbe0a5c5bbb8909fcc89875598b8e1267cb4886e8f1ebcdbe12145e6c6
+size 2496044

data/freesound/audio/fs_007_nature_wind_bushes_coast_903_PM_240311_0680.wav

@@ -0,0 +1,3 @@
+version https://git-lfs.github.com/spec/v1
+oid sha256:7b17b1c51903e9e8f6c302eb6a3e5b661314e5bce366f16f977a33ec0cbfc3d0
+size 2497080

data/freesound/audio/fs_008_nature_estate_farm_noise_1153AM3_220509_0343.wav.wav

@@ -0,0 +1,3 @@
+version https://git-lfs.github.com/spec/v1
+oid sha256:c27d632f13199874d3a7bcd239faf615cd74bfccd307cc30f00230ba9a92880e
+size 1290268

data/freesound/audio/fs_009_nature_Howler_monkey_howling_in_the_evening_jungle.wav

@@ -0,0 +1,3 @@
+version https://git-lfs.github.com/spec/v1
+oid sha256:3f7d541ad4f512b08a7890e8b6c54802a79058360476e38a5b6569f35465edfd
+size 2823188

data/freesound/audio/fs_010_nature_forest_heavy_rain_loop.wav.wav

@@ -0,0 +1,3 @@
+version https://git-lfs.github.com/spec/v1
+oid sha256:10d2681d774b21b0b6b5d74ffa46546afd6d927e3670e4f82bb9592ec28fa8ed
+size 2016046

data/freesound/audio/fs_011_nature_CardinalChirpingParabolicMP3VersionMay62013.mp3.wav

@@ -0,0 +1,3 @@
+version https://git-lfs.github.com/spec/v1
+oid sha256:21bac3a5ac767cd7f3b03108756dbf8a8c63605b86dda71db30a209a479fdc42
+size 2066826

data/freesound/audio/fs_012_nature_Superb_Lyrebird_II.wav

@@ -0,0 +1,3 @@
+version https://git-lfs.github.com/spec/v1
+oid sha256:fa8f7970bda68d45a2dc8307f2f2c0feb789573bcdbcb1699db59eb18d736fdd
+size 2731094

data/freesound/audio/fs_013_nature_Night_Wind_Chimes.wav

@@ -0,0 +1,3 @@
+version https://git-lfs.github.com/spec/v1
+oid sha256:dcf6ed33ddd7283df586f8c39b03bd5a4e27fc2323be6bf72e796c80b4a59be7
+size 1556490

data/freesound/audio/fs_014_nature_Southern_Summer_Evening_Ambience_with_Crickets_4.wav

@@ -0,0 +1,3 @@
+version https://git-lfs.github.com/spec/v1
+oid sha256:98035c3df9658cc73fac5214e29ff73dfe4a56ff93f89dd8d7b6fa41e74799b4
+size 2810960

data/freesound/audio/fs_015_nature_Li_River._Crickets_&_Frogs.wav

@@ -0,0 +1,3 @@
+version https://git-lfs.github.com/spec/v1
+oid sha256:225df4253c469d91504e23fa4e365d09893d5e9fcb5cb9dec7a03ecc7ee1f7be
+size 2758824

data/freesound/audio/fs_016_nature_Thunder_sound_effect_2.wav

@@ -0,0 +1,3 @@
+version https://git-lfs.github.com/spec/v1
+oid sha256:7eae669115404f55dac50814b543aa96665a91ffee618eabe999698170f93fa6
+size 1750612

data/freesound/audio/fs_017_nature_Rain_medium_-_falling_on_soft_surface.wav

@@ -0,0 +1,3 @@
+version https://git-lfs.github.com/spec/v1
+oid sha256:6cb55480eaefd15af572af917f0689d9b236b9317c4c5fb52d84ab800722a3dc
+size 2595922

data/freesound/audio/fs_018_nature_Thunder_6.wav

@@ -0,0 +1,3 @@
+version https://git-lfs.github.com/spec/v1
+oid sha256:20f51911074f08f93697d5bbcd7dbb38f6e2930316b074d9c3835c4f49046a8b
+size 1317046

data/freesound/audio/fs_019_nature_Water_Flowing_into_Underground_Sinkhole.wav

@@ -0,0 +1,3 @@
+version https://git-lfs.github.com/spec/v1
+oid sha256:04ca65aa8f703a36700c89ae090dd778c8a66a8a33fe66cd3e0abcbf150f046d
+size 2192044

data/freesound/audio/fs_020_nature_Water_Flowing_Through_Close_Rapids_6.wav

@@ -0,0 +1,3 @@
+version https://git-lfs.github.com/spec/v1
+oid sha256:9f46d5db77e80d22b8d547e71367d95ee36b2ffa526321d6eb49507ad60f8952
+size 2214176

data/freesound/audio/fs_021_nature_Water_Flowing_Through_Close_Rapids_5.wav

@@ -0,0 +1,3 @@
+version https://git-lfs.github.com/spec/v1
+oid sha256:d010c690c23be21d4d46b130f4442b8b5f1da88c9f202e569dea739b7cd499f1
+size 2616404

data/freesound/audio/fs_022_urban_CITY_AMBIENCE_for_CLEAN_LOOP.wav.wav

@@ -0,0 +1,3 @@
+version https://git-lfs.github.com/spec/v1
+oid sha256:f42d5388f879b79e682ac9807c935312353d014b205437bddee53c0dba559b1b
+size 2146684

data/freesound/audio/fs_023_urban_Kitchen_Ambience_During_Daytime_.wav.wav

@@ -0,0 +1,3 @@
+version https://git-lfs.github.com/spec/v1
+oid sha256:e5758edd039b63d52cb9e1b25f03c5eefd9de9ec777aee18e785d459d70cd7d7
+size 2774062

data/freesound/audio/fs_024_urban_police_in_Paris.wav.wav

@@ -0,0 +1,3 @@
+version https://git-lfs.github.com/spec/v1
+oid sha256:13b6de6a41e3b27bb04a38122750cb73827795b2e9b4dbc064f7059c60ba9ad3
+size 2859688

data/freesound/audio/fs_025_urban_BerlinGameScene.com_Crowd_Cheer_4.wav

@@ -0,0 +1,3 @@
+version https://git-lfs.github.com/spec/v1
+oid sha256:ca553ee2ac8517e773e9b3c6c544763a934e8059388c4a1ff64469f6c6ac282f
+size 682308

data/freesound/audio/fs_026_urban_Да_ладно.wav

@@ -0,0 +1,3 @@
+version https://git-lfs.github.com/spec/v1
+oid sha256:e915163fa6094eb25107cc4255e716c39b954f7816b8af31d4379f1a84433570
+size 1029580

data/freesound/audio/fs_027_urban_Restaurant_5.mp3.wav

@@ -0,0 +1,3 @@
+version https://git-lfs.github.com/spec/v1
+oid sha256:46956eee584e391c06b083a55a7e301e7d90b7c21810b35f5c33b6dcfd0c6ecf
+size 2727330

data/freesound/audio/fs_028_urban_Industrial_grinder.wav

@@ -0,0 +1,3 @@
+version https://git-lfs.github.com/spec/v1
+oid sha256:32aa2939ede7f9b5b573976e2bddd69cafb989b84912f41c7c92e68312b81078
+size 2343072

data/freesound/audio/fs_029_urban_Continuous_mechanical_whir_.wav

@@ -0,0 +1,3 @@
+version https://git-lfs.github.com/spec/v1
+oid sha256:f34cb93921b80702b4a895f978640864d16b80d105c68cdd92e5832020368fde
+size 2569814

data/freesound/audio/fs_030_urban_Electric_saw.wav

@@ -0,0 +1,3 @@
+version https://git-lfs.github.com/spec/v1
+oid sha256:9eebafed83e3904b2607bbe55013740f966f806ab607346a4a783f4bb4fee159
+size 2581360

data/freesound/audio/fs_031_urban_Renault_Master_F3500_dCi135_Foley_Horn_Outside_Mono.wav.wav

@@ -0,0 +1,3 @@
+version https://git-lfs.github.com/spec/v1
+oid sha256:627b86f6d4d094686ef69849ba01156e0dd59ce3524134d518eaab6010d35650
+size 742252

data/freesound/audio/fs_032_urban_Street_atmosphere_and_sounds_in_Rome.wav

@@ -0,0 +1,3 @@
+version https://git-lfs.github.com/spec/v1
+oid sha256:50615b73fe56ddf1a2a182b3800dcbd6e43c731c51d47d62794e6e6043eaa775
+size 2296034

data/freesound/audio/fs_033_urban_Distant_horn_beeps_09.flac.wav

@@ -0,0 +1,3 @@
+version https://git-lfs.github.com/spec/v1
+oid sha256:3d6be04f78e509ecb0311405f0e47d22ca1c6de01a9c9af7b1001796851d7069
+size 1588044

data/freesound/audio/fs_034_urban_Underground_Train_Station_Ambience.wav

@@ -0,0 +1,3 @@
+version https://git-lfs.github.com/spec/v1
+oid sha256:dad7da1daf478f4b52fb81f69642fec07e92e015073966fe8aceeb1a8124cc51
+size 2078834

data/freesound/audio/fs_035_urban_subwaystartsequence.wav.wav

@@ -0,0 +1,3 @@
+version https://git-lfs.github.com/spec/v1
+oid sha256:20b0d9dcaaa7d811f77aebf37b15cbbb932c2b91cfa7d05e7ff5d7fba293c746
+size 1399148

data/freesound/audio/fs_036_urban_Train,_Subway,_Pass_By,_Under_Bridge_Perspective.wav

@@ -0,0 +1,3 @@
+version https://git-lfs.github.com/spec/v1
+oid sha256:eb5a75bb293b4e43675f3ac73ba6d43dabe507f531452776b5e34aeda5f90c8f
+size 1990804

data/freesound/audio/fs_037_urban_SFX_Donkey_Cart.MP3.wav

@@ -0,0 +1,3 @@
+version https://git-lfs.github.com/spec/v1
+oid sha256:76547ae043e0ac913767dd5c4726cfc68067048f7c8ac2059aad917257c61887
+size 854038

data/freesound/audio/fs_038_urban_people_talking_at_street_market.wav

@@ -0,0 +1,3 @@
+version https://git-lfs.github.com/spec/v1
+oid sha256:f6b4025514d31795ceca082446fd2dc85159e8b3dbc1c2606dcc73163b2f8351
+size 1176878

data/freesound/audio/fs_039_urban_Japan_Tokyo_Shinjuku_Street_Promoter_Yelling_City.wav.wav

@@ -0,0 +1,3 @@
+version https://git-lfs.github.com/spec/v1
+oid sha256:38c4b41f9b1b0a894f55fce69058fdfa830d3e1c90c735fdeb47d409a0bbe257
+size 2448622

data/freesound/audio/fs_040_water_waves_close_coast_small_004.wav.wav

@@ -0,0 +1,3 @@
+version https://git-lfs.github.com/spec/v1
+oid sha256:0dd695890eb84143ffa372aae09270c6bd4c69c80572ecdf31b69e438ae0d192
+size 1213436

data/freesound/audio/fs_041_water_wave_sand_beach_012.wav.wav

@@ -0,0 +1,3 @@
+version https://git-lfs.github.com/spec/v1
+oid sha256:c585c1a0d81502962a3e964b6bd472534d520a952a4ca9d33798bc2c88683b2a
+size 572044

data/freesound/audio/fs_042_water_waves_over_mall_shells_01_150515_00.wav.wav

@@ -0,0 +1,3 @@
+version https://git-lfs.github.com/spec/v1
+oid sha256:41327238d54282c6fe7c7e6187d688f3019737985d4931dab8ec964db51ae37d
+size 1374572

data/freesound/audio/fs_043_water_Rain_on_a_Plastic_Roof.wav

@@ -0,0 +1,3 @@
+version https://git-lfs.github.com/spec/v1
+oid sha256:0ec74947189f6e2e2b66c3f6f99556dd05264291c8a1cc67eb2fe9a66e130027
+size 2136578

data/freesound/audio/fs_044_water_Ambiance_Waterfall_Big_Skogafoss_Far_Loop_Stereo_02.wav.wav

@@ -0,0 +1,3 @@
+version https://git-lfs.github.com/spec/v1
+oid sha256:048b7779e05420063cb4c2aefc0c4a480e0fec9cca3227d79a95f2b35692f9ac
+size 2880044

data/freesound/audio/fs_045_water_Water_Stream_1.wav

@@ -0,0 +1,3 @@
+version https://git-lfs.github.com/spec/v1
+oid sha256:fe4d4a8c6655ef169541954da8db29a889da1c12135e28638389f8c4a59fe879
+size 2496044