adding play detection score

README.md

@@ -94,7 +94,7 @@ cd hf-video-scoring
 conda create -n nfl-play python=3.10 -y
 conda activate nfl-play
 conda install -y -c pytorch -c conda-forge pytorch torchvision torchaudio cpuonly
-pip install pytorchvideo huggingface_hub ffmpeg-python transformers
+pip install pytorchvideo huggingface_hub ffmpeg-python transformers openai-whisper
 ```
 
 **Using pip in a virtualenv:**
@@ -136,7 +136,7 @@ python inference.py data/segment_000.mov
 
 This will:
 - Print the Top-5 predicted Kinetics-400 labels + probabilities for the given clip
-- Generate audio transcription using Wav2Vec2
+- Generate high-quality audio transcription using Whisper-Medium with NFL sports vocabulary
 - Save results to `classification.json` and `transcripts.json`
 
 ### 4. Process all clips in bulk
@@ -147,11 +147,11 @@ chmod +x run_all_clips.py
 ```
 
 * Iterates over every `.mov`/`.mp4` in `data/`
-* Runs both video classification and audio transcription for each clip
-* Prints per-clip results to console
+* Runs both video classification and enhanced audio transcription for each clip
+* Prints per-clip results to console with NFL-optimized transcriptions
 * Saves cumulative results in:
   - `classification.json` (video classification scores)
-  - `transcripts.json` (audio transcriptions)
+  - `transcripts.json` (professional-quality NFL commentary transcriptions)
 * Handles errors gracefully (empty lists/strings for failed clips)
 
 ## Complete Workflow Example
@@ -191,13 +191,39 @@ This workflow captures live NFL content, automatically segments it, and then ana
 
 ```json
 {
-  "segment_001.mov": "TA N ONE   THE SECOND",
-  "segment_002.mov": "",  // if failed or skipped
-  "segment_003.mov": "SOME AUDIO TEXT"
+  "segment_001.mov": "I can tell you that Lamar Jackson right now is",
+  "segment_002.mov": "is the sixth best quarterback in the NFL",
+  "segment_003.mov": "He's basically like an extra line"
 }
 ```
 
-## Customization
+## Audio Transcription Features
+
+The system uses **Whisper-Medium** with NFL-specific enhancements for superior audio transcription:
+
+### 🎯 **NFL-Optimized Transcription**
+* **Advanced Model**: OpenAI Whisper-Medium (3B parameters) for professional-quality transcription
+* **Sports Vocabulary**: 80+ NFL-specific terms including teams, positions, plays, and penalties
+* **Smart Corrections**: Automatic correction of common football terminology mishears
+* **Player Recognition**: Accurately transcribes player names and team references
+* **Commentary Context**: Optimized for NFL broadcast commentary and analysis
+
+### 🔧 **Audio Enhancement Pipeline**
+* **Noise Filtering**: High-pass (80Hz) and low-pass (8kHz) filters to remove audio artifacts
+* **Audio Normalization**: Automatic level adjustment for consistent processing
+* **Silence Detection**: Skips transcription for very quiet or short audio segments
+* **Error Handling**: Graceful fallback for corrupted or problematic audio
+
+### 📊 **Quality Examples**
+```
+Before: "TA N ONE   THE SECOND"
+After:  "I can tell you that Lamar Jackson right now is"
+
+Before: "FAY FOOLISH" 
+After:  "is the sixth best quarterback in the NFL"
+```
+
+## Video Classification Customization
 
 * **Model choice**: in `inference.py`, set `model_name` to one of:
 
@@ -205,12 +231,28 @@ This workflow captures live NFL content, automatically segments it, and then ana
 * **Clip length**: default is 2 seconds; adjust `video.get_clip(0, 2.0)`
 * **Sampling & crop**: modify `preprocess()` in `inference.py` to change `num_samples` or spatial size.
 
+## Audio Transcription Customization
+
+* **Model choice**: in `inference.py`, change `model` from `whisper-medium` to:
+  * `whisper-base` (faster, less accurate)
+  * `whisper-large` (slower, more accurate)
+* **Sports vocabulary**: Modify `NFL_SPORTS_CONTEXT` list to add custom terms
+* **Audio filters**: Adjust FFmpeg filters in `load_audio()` for different audio quality
+* **Corrections**: Add custom corrections in `apply_sports_corrections()`
+
 ## Troubleshooting
 
+### Video Issues
 * **InvalidDataError: moov atom not found**: skip the damaged clip—`run_all_clips.py` logs and continues.
 * **Dimension or kernel size errors**: ensure `num_samples` >= model's temporal kernel (≥13 for X3D).
 * **PyTorch installation**: use Conda on macOS for best compatibility.
 
+### Audio Transcription Issues
+* **Slow transcription**: Whisper-Medium is large (~3GB). Use `whisper-base` for faster processing.
+* **Poor audio quality**: Ensure clean audio input. The system filters noise but very poor audio may still fail.
+* **Memory issues**: Whisper-Medium requires ~4GB RAM. Use `whisper-base` on lower-memory systems.
+* **Language detection**: The system forces English. For other languages, modify the `language` parameter.
+
 ## Next Steps
 
 * **Fine-tune** X3D on your own "start" vs "end" NFL play dataset.

classification.json

@@ -20,1656 +20,5 @@
       "\"passing American football (in game)\"",
       0.0025873929262161255
     ]
-  ],
-  "segment_002.mov": [
-    [
-      "\"mowing lawn\"",
-      0.004539191722869873
-    ],
-    [
-      "\"riding or walking with horse\"",
-      0.002820482011884451
-    ],
-    [
-      "\"driving tractor\"",
-      0.002591945929452777
-    ],
-    [
-      "\"catching or throwing baseball\"",
-      0.002587514230981469
-    ],
-    [
-      "archery",
-      0.0025512496940791607
-    ]
-  ],
-  "segment_003.mov": [
-    [
-      "\"hurling (sport)\"",
-      0.0035289316438138485
-    ],
-    [
-      "headbutting",
-      0.0029045743867754936
-    ],
-    [
-      "\"passing American football (not in game)\"",
-      0.002668950939550996
-    ],
-    [
-      "\"pumping fist\"",
-      0.0026475831400603056
-    ],
-    [
-      "\"playing cricket\"",
-      0.002635735785588622
-    ]
-  ],
-  "segment_004.mov": [
-    [
-      "applauding",
-      0.00304257869720459
-    ],
-    [
-      "headbutting",
-      0.0029725206550210714
-    ],
-    [
-      "\"pumping fist\"",
-      0.0029444803949445486
-    ],
-    [
-      "\"hurling (sport)\"",
-      0.0027822419069707394
-    ],
-    [
-      "\"shaking hands\"",
-      0.002624393440783024
-    ]
-  ],
-  "segment_005.mov": [
-    [
-      "\"catching or throwing baseball\"",
-      0.003370030550286174
-    ],
-    [
-      "\"pumping fist\"",
-      0.00289507070556283
-    ],
-    [
-      "\"catching or throwing frisbee\"",
-      0.0026615660171955824
-    ],
-    [
-      "\"eating hotdog\"",
-      0.002651733811944723
-    ],
-    [
-      "\"javelin throw\"",
-      0.0026478723157197237
-    ]
-  ],
-  "segment_006.mov": [
-    [
-      "\"riding mountain bike\"",
-      0.0030539515428245068
-    ],
-    [
-      "\"bench pressing\"",
-      0.0030068582855165005
-    ],
-    [
-      "\"changing wheel\"",
-      0.0028557234909385443
-    ],
-    [
-      "\"checking tires\"",
-      0.002828161232173443
-    ],
-    [
-      "\"changing oil\"",
-      0.0025825132615864277
-    ]
-  ],
-  "segment_007.mov": [
-    [
-      "\"changing oil\"",
-      0.003507673041895032
-    ],
-    [
-      "\"pushing wheelchair\"",
-      0.0028540242929011583
-    ],
-    [
-      "motorcycling",
-      0.002735486486926675
-    ],
-    [
-      "\"checking tires\"",
-      0.0026065800338983536
-    ],
-    [
-      "\"changing wheel\"",
-      0.0026001674123108387
-    ]
-  ],
-  "segment_008.mov": [
-    [
-      "motorcycling",
-      0.0034892940893769264
-    ],
-    [
-      "\"riding mountain bike\"",
-      0.0027264610398560762
-    ],
-    [
-      "\"pushing wheelchair\"",
-      0.0026729153469204903
-    ],
-    [
-      "\"changing oil\"",
-      0.002655936172232032
-    ],
-    [
-      "snowmobiling",
-      0.002622972708195448
-    ]
-  ],
-  "segment_009.mov": [
-    [
-      "\"kicking field goal\"",
-      0.0051559642888605595
-    ],
-    [
-      "\"tossing coin\"",
-      0.002858997555449605
-    ],
-    [
-      "\"passing American football (in game)\"",
-      0.002840855158865452
-    ],
-    [
-      "\"high kick\"",
-      0.0024956425186246634
-    ],
-    [
-      "headbutting",
-      0.002494481625035405
-    ]
-  ],
-  "segment_010.mov": [
-    [
-      "\"tossing coin\"",
-      0.003494303673505783
-    ],
-    [
-      "headbutting",
-      0.003139702370390296
-    ],
-    [
-      "\"kicking field goal\"",
-      0.003000226803123951
-    ],
-    [
-      "\"passing American football (in game)\"",
-      0.0028010711539536715
-    ],
-    [
-      "\"pumping fist\"",
-      0.0026133027859032154
-    ]
-  ],
-  "segment_011.mov": [
-    [
-      "headbutting",
-      0.0035944455303251743
-    ],
-    [
-      "\"kicking field goal\"",
-      0.002746515441685915
-    ],
-    [
-      "\"passing American football (in game)\"",
-      0.00274651194922626
-    ],
-    [
-      "cheerleading",
-      0.0027097328566014767
-    ],
-    [
-      "\"playing ice hockey\"",
-      0.002675419207662344
-    ]
-  ],
-  "segment_012.mov": [
-    [
-      "cheerleading",
-      0.0033036908134818077
-    ],
-    [
-      "\"dancing macarena\"",
-      0.0031919179018586874
-    ],
-    [
-      "applauding",
-      0.0027217946480959654
-    ],
-    [
-      "\"hitting baseball\"",
-      0.0026708708610385656
-    ],
-    [
-      "\"garbage collecting\"",
-      0.0026213040109723806
-    ]
-  ],
-  "segment_013.mov": [
-    [
-      "\"pumping fist\"",
-      0.0035472228191792965
-    ],
-    [
-      "\"kicking field goal\"",
-      0.0027715854812413454
-    ],
-    [
-      "\"dunking basketball\"",
-      0.0027416404336690903
-    ],
-    [
-      "applauding",
-      0.002673310926184058
-    ],
-    [
-      "celebrating",
-      0.002629267517477274
-    ]
-  ],
-  "segment_014.mov": [
-    [
-      "headbutting",
-      0.0038350492250174284
-    ],
-    [
-      "bobsledding",
-      0.002937863813713193
-    ],
-    [
-      "\"playing ice hockey\"",
-      0.002785387681797147
-    ],
-    [
-      "\"punching person (boxing)\"",
-      0.002671806840226054
-    ],
-    [
-      "\"changing wheel\"",
-      0.0026315213181078434
-    ]
-  ],
-  "segment_015.mov": [
-    [
-      "\"passing American football (in game)\"",
-      0.00389948021620512
-    ],
-    [
-      "\"side kick\"",
-      0.003301523393020034
-    ],
-    [
-      "\"kicking field goal\"",
-      0.002986735198646784
-    ],
-    [
-      "\"passing American football (not in game)\"",
-      0.0026217757258564234
-    ],
-    [
-      "\"catching or throwing baseball\"",
-      0.002540471963584423
-    ]
-  ],
-  "segment_016.mov": [
-    [
-      "\"passing American football (in game)\"",
-      0.006536518689244986
-    ],
-    [
-      "\"side kick\"",
-      0.0025378491263836622
-    ],
-    [
-      "\"passing American football (not in game)\"",
-      0.0025076796300709248
-    ],
-    [
-      "\"kicking field goal\"",
-      0.002507618861272931
-    ],
-    [
-      "\"high kick\"",
-      0.0024923686869442463
-    ]
-  ],
-  "segment_017.mov": [
-    [
-      "\"passing American football (in game)\"",
-      0.004528654273599386
-    ],
-    [
-      "\"kicking field goal\"",
-      0.003415569895878434
-    ],
-    [
-      "\"passing American football (not in game)\"",
-      0.0026175878010690212
-    ],
-    [
-      "\"side kick\"",
-      0.0025776273105293512
-    ],
-    [
-      "\"high kick\"",
-      0.0025026851799339056
-    ]
-  ],
-  "segment_018.mov": [
-    [
-      "\"passing American football (in game)\"",
-      0.004061450716108084
-    ],
-    [
-      "\"side kick\"",
-      0.0038009993731975555
-    ],
-    [
-      "\"kicking field goal\"",
-      0.002601894084364176
-    ],
-    [
-      "\"high kick\"",
-      0.002582114189863205
-    ],
-    [
-      "\"passing American football (not in game)\"",
-      0.0025175546761602163
-    ]
-  ],
-  "segment_019.mov": [
-    [
-      "\"passing American football (in game)\"",
-      0.006576324347406626
-    ],
-    [
-      "\"passing American football (not in game)\"",
-      0.002554940525442362
-    ],
-    [
-      "\"kicking field goal\"",
-      0.0024935186374932528
-    ],
-    [
-      "\"side kick\"",
-      0.0024903316516429186
-    ],
-    [
-      "\"catching or throwing frisbee\"",
-      0.002489960053935647
-    ]
-  ],
-  "segment_020.mov": [
-    [
-      "\"passing American football (in game)\"",
-      0.006745155900716782
-    ],
-    [
-      "\"passing American football (not in game)\"",
-      0.002495990600436926
-    ],
-    [
-      "\"kicking field goal\"",
-      0.0024898555129766464
-    ],
-    [
-      "headbutting",
-      0.002489602193236351
-    ],
-    [
-      "\"shaking hands\"",
-      0.0024894699454307556
-    ]
-  ],
-  "segment_021.mov": [
-    [
-      "\"passing American football (in game)\"",
-      0.004887602757662535
-    ],
-    [
-      "\"passing American football (not in game)\"",
-      0.002793429885059595
-    ],
-    [
-      "situp",
-      0.0026539755053818226
-    ],
-    [
-      "\"dribbling basketball\"",
-      0.0026028482243418694
-    ],
-    [
-      "\"throwing ball\"",
-      0.002529650926589966
-    ]
-  ],
-  "segment_022.mov": [
-    [
-      "\"pumping fist\"",
-      0.002948896726593375
-    ],
-    [
-      "\"passing American football (in game)\"",
-      0.0028372337110340595
-    ],
-    [
-      "\"tossing coin\"",
-      0.0027690029237419367
-    ],
-    [
-      "\"shaking hands\"",
-      0.0027542426250874996
-    ],
-    [
-      "headbutting",
-      0.0026179237756878138
-    ]
-  ],
-  "segment_023.mov": [
-    [
-      "\"pumping fist\"",
-      0.003991341684013605
-    ],
-    [
-      "headbutting",
-      0.002623251872137189
-    ],
-    [
-      "motorcycling",
-      0.0025920518673956394
-    ],
-    [
-      "skydiving",
-      0.0025866003707051277
-    ],
-    [
-      "bobsledding",
-      0.0025851819664239883
-    ]
-  ],
-  "segment_024.mov": [
-    [
-      "\"riding a bike\"",
-      0.0037740382831543684
-    ],
-    [
-      "motorcycling",
-      0.002751761581748724
-    ],
-    [
-      "bobsledding",
-      0.0026172574143856764
-    ],
-    [
-      "\"riding unicycle\"",
-      0.0025924895890057087
-    ],
-    [
-      "\"pumping fist\"",
-      0.0025839442387223244
-    ]
-  ],
-  "segment_025.mov": [
-    [
-      "bobsledding",
-      0.004564006347209215
-    ],
-    [
-      "\"shaking hands\"",
-      0.0026650093495845795
-    ],
-    [
-      "motorcycling",
-      0.002605688525363803
-    ],
-    [
-      "\"pumping fist\"",
-      0.0026000377256423235
-    ],
-    [
-      "skydiving",
-      0.0025698416866362095
-    ]
-  ],
-  "segment_026.mov": [
-    [
-      "\"tossing coin\"",
-      0.005768336355686188
-    ],
-    [
-      "squat",
-      0.0025260422844439745
-    ],
-    [
-      "\"side kick\"",
-      0.0025187088176608086
-    ],
-    [
-      "\"shaking hands\"",
-      0.0025166154373437166
-    ],
-    [
-      "\"kicking field goal\"",
-      0.0025109625421464443
-    ]
-  ],
-  "segment_027.mov": [
-    [
-      "\"carrying baby\"",
-      0.0028944576624780893
-    ],
-    [
-      "\"side kick\"",
-      0.002688055858016014
-    ],
-    [
-      "\"salsa dancing\"",
-      0.002644676947966218
-    ],
-    [
-      "\"hula hooping\"",
-      0.0026000014040619135
-    ],
-    [
-      "archery",
-      0.002581424079835415
-    ]
-  ],
-  "segment_028.mov": [
-    [
-      "\"stretching arm\"",
-      0.004183173179626465
-    ],
-    [
-      "archery",
-      0.0028685072902590036
-    ],
-    [
-      "\"carrying baby\"",
-      0.002629334107041359
-    ],
-    [
-      "yoga",
-      0.0025525041855871677
-    ],
-    [
-      "\"stretching leg\"",
-      0.0025523642543703318
-    ]
-  ],
-  "segment_029.mov": [
-    [
-      "\"getting a tattoo\"",
-      0.0036486154422163963
-    ],
-    [
-      "archery",
-      0.0027306571137160063
-    ],
-    [
-      "\"punching person (boxing)\"",
-      0.0027052827645093203
-    ],
-    [
-      "\"punching bag\"",
-      0.002618985017761588
-    ],
-    [
-      "squat",
-      0.0026112094055861235
-    ]
-  ],
-  "segment_030.mov": [
-    [
-      "\"spray painting\"",
-      0.004089465364813805
-    ],
-    [
-      "\"getting a tattoo\"",
-      0.002762633841484785
-    ],
-    [
-      "\"playing paintball\"",
-      0.0027071302756667137
-    ],
-    [
-      "\"changing wheel\"",
-      0.0026773668359965086
-    ],
-    [
-      "bobsledding",
-      0.0025922050699591637
-    ]
-  ],
-  "segment_031.mov": [
-    [
-      "\"mowing lawn\"",
-      0.0034001749008893967
-    ],
-    [
-      "\"passing American football (not in game)\"",
-      0.003030994674190879
-    ],
-    [
-      "\"catching or throwing baseball\"",
-      0.0028556634206324816
-    ],
-    [
-      "\"pumping fist\"",
-      0.002578843617811799
-    ],
-    [
-      "\"riding or walking with horse\"",
-      0.002575062680989504
-    ]
-  ],
-  "segment_032.mov": [
-    [
-      "\"passing American football (in game)\"",
-      0.003045632503926754
-    ],
-    [
-      "marching",
-      0.0030235128942877054
-    ],
-    [
-      "\"passing American football (not in game)\"",
-      0.0029362209606915712
-    ],
-    [
-      "\"kicking field goal\"",
-      0.0027396646328270435
-    ],
-    [
-      "\"hurling (sport)\"",
-      0.002676320029422641
-    ]
-  ],
-  "segment_033.mov": [
-    [
-      "\"mowing lawn\"",
-      0.004852019250392914
-    ],
-    [
-      "\"passing American football (not in game)\"",
-      0.002654429292306304
-    ],
-    [
-      "\"hurling (sport)\"",
-      0.002581879962235689
-    ],
-    [
-      "\"kicking soccer ball\"",
-      0.0025664607528597116
-    ],
-    [
-      "\"juggling soccer ball\"",
-      0.002544356742873788
-    ]
-  ],
-  "segment_034.mov": [
-    [
-      "archery",
-      0.004724352620542049
-    ],
-    [
-      "\"trimming or shaving beard\"",
-      0.0026578360702842474
-    ],
-    [
-      "\"golf driving\"",
-      0.002610759809613228
-    ],
-    [
-      "\"tossing coin\"",
-      0.002541653113439679
-    ],
-    [
-      "\"mowing lawn\"",
-      0.0025332642253488302
-    ]
-  ],
-  "segment_035.mov": [
-    [
-      "\"playing harmonica\"",
-      0.006103646010160446
-    ],
-    [
-      "whistling",
-      0.002615551697090268
-    ],
-    [
-      "smoking",
-      0.0025100400671362877
-    ],
-    [
-      "yawning",
-      0.0025017866864800453
-    ],
-    [
-      "\"playing ukulele\"",
-      0.0025011510588228703
-    ]
-  ],
-  "segment_036.mov": [
-    [
-      "\"passing American football (in game)\"",
-      0.006577454507350922
-    ],
-    [
-      "\"passing American football (not in game)\"",
-      0.002514705527573824
-    ],
-    [
-      "\"kicking field goal\"",
-      0.0025104687083512545
-    ],
-    [
-      "\"side kick\"",
-      0.0025097192265093327
-    ],
-    [
-      "\"high kick\"",
-      0.002491986844688654
-    ]
-  ],
-  "segment_037.mov": [
-    [
-      "\"passing American football (in game)\"",
-      0.006760087329894304
-    ],
-    [
-      "\"passing American football (not in game)\"",
-      0.0024913023225963116
-    ],
-    [
-      "\"kicking field goal\"",
-      0.00248970789834857
-    ],
-    [
-      "\"side kick\"",
-      0.002489360049366951
-    ],
-    [
-      "\"high kick\"",
-      0.002489320933818817
-    ]
-  ],
-  "segment_038.mov": [
-    [
-      "\"passing American football (in game)\"",
-      0.0062811062671244144
-    ],
-    [
-      "\"kicking field goal\"",
-      0.0026691341772675514
-    ],
-    [
-      "\"passing American football (not in game)\"",
-      0.002501868177205324
-    ],
-    [
-      "\"side kick\"",
-      0.0024906292092055082
-    ],
-    [
-      "\"tossing coin\"",
-      0.0024903123266994953
-    ]
-  ],
-  "segment_039.mov": [
-    [
-      "\"passing American football (in game)\"",
-      0.004427704028785229
-    ],
-    [
-      "\"kicking field goal\"",
-      0.00305744051001966
-    ],
-    [
-      "\"passing American football (not in game)\"",
-      0.0028497367165982723
-    ],
-    [
-      "\"shaking hands\"",
-      0.0025244825519621372
-    ],
-    [
-      "\"tossing coin\"",
-      0.0025086107198148966
-    ]
-  ],
-  "segment_040.mov": [
-    [
-      "\"passing American football (in game)\"",
-      0.00507122790440917
-    ],
-    [
-      "\"tossing coin\"",
-      0.0028494407888501883
-    ],
-    [
-      "\"kicking field goal\"",
-      0.0025832131505012512
-    ],
-    [
-      "\"shaking hands\"",
-      0.0025662044063210487
-    ],
-    [
-      "\"passing American football (not in game)\"",
-      0.0025347082410007715
-    ]
-  ],
-  "segment_041.mov": [
-    [
-      "\"passing American football (in game)\"",
-      0.002910151146352291
-    ],
-    [
-      "\"tossing coin\"",
-      0.00288573419675231
-    ],
-    [
-      "bobsledding",
-      0.0027363626286387444
-    ],
-    [
-      "\"changing wheel\"",
-      0.0027150516398251057
-    ],
-    [
-      "\"kicking field goal\"",
-      0.0026453319005668163
-    ]
-  ],
-  "segment_042.mov": [
-    [
-      "bobsledding",
-      0.0029746112413704395
-    ],
-    [
-      "marching",
-      0.0026662067975848913
-    ],
-    [
-      "\"riding mountain bike\"",
-      0.0026005373802036047
-    ],
-    [
-      "archery",
-      0.002575363963842392
-    ],
-    [
-      "\"passing American football (not in game)\"",
-      0.0025738172698765993
-    ]
-  ],
-  "segment_043.mov": [
-    [
-      "bobsledding",
-      0.0032584425061941147
-    ],
-    [
-      "\"changing wheel\"",
-      0.002937484998255968
-    ],
-    [
-      "\"driving car\"",
-      0.0028462321497499943
-    ],
-    [
-      "motorcycling",
-      0.00263536861166358
-    ],
-    [
-      "skateboarding",
-      0.002625212771818042
-    ]
-  ],
-  "segment_044.mov": [
-    [
-      "archery",
-      0.0029061429668217897
-    ],
-    [
-      "bobsledding",
-      0.002719373907893896
-    ],
-    [
-      "\"golf driving\"",
-      0.0027050392236560583
-    ],
-    [
-      "barbequing",
-      0.0025971110444515944
-    ],
-    [
-      "\"golf putting\"",
-      0.0025699003599584103
-    ]
-  ],
-  "segment_045.mov": [
-    [
-      "barbequing",
-      0.003022523829713464
-    ],
-    [
-      "\"changing wheel\"",
-      0.002754890127107501
-    ],
-    [
-      "\"mowing lawn\"",
-      0.0026360696647316217
-    ],
-    [
-      "auctioning",
-      0.0026054272893816233
-    ],
-    [
-      "\"changing oil\"",
-      0.0025794513057917356
-    ]
-  ],
-  "segment_046.mov": [
-    [
-      "\"tossing coin\"",
-      0.003568414831534028
-    ],
-    [
-      "\"side kick\"",
-      0.003212416311725974
-    ],
-    [
-      "\"rock scissors paper\"",
-      0.002615532139316201
-    ],
-    [
-      "\"punching person (boxing)\"",
-      0.0026119472458958626
-    ],
-    [
-      "\"high kick\"",
-      0.002596732461825013
-    ]
-  ],
-  "segment_047.mov": [
-    [
-      "\"passing American football (in game)\"",
-      0.0031079247128218412
-    ],
-    [
-      "\"juggling soccer ball\"",
-      0.0028041608165949583
-    ],
-    [
-      "\"passing American football (not in game)\"",
-      0.002784370444715023
-    ],
-    [
-      "\"catching or throwing baseball\"",
-      0.0027656129095703363
-    ],
-    [
-      "\"tossing coin\"",
-      0.0027509713545441628
-    ]
-  ],
-  "segment_048.mov": [
-    [
-      "\"tossing coin\"",
-      0.004612566903233528
-    ],
-    [
-      "\"passing American football (in game)\"",
-      0.0030128920916467905
-    ],
-    [
-      "\"side kick\"",
-      0.002602673601359129
-    ],
-    [
-      "\"passing American football (not in game)\"",
-      0.0025423497427254915
-    ],
-    [
-      "\"rock scissors paper\"",
-      0.0025353787932544947
-    ]
-  ],
-  "segment_049.mov": [
-    [
-      "beatboxing",
-      0.002776292385533452
-    ],
-    [
-      "\"pumping fist\"",
-      0.0027726562693715096
-    ],
-    [
-      "\"shaking hands\"",
-      0.0027530265506356955
-    ],
-    [
-      "\"playing poker\"",
-      0.002649413887411356
-    ],
-    [
-      "\"arm wrestling\"",
-      0.0026053758338093758
-    ]
-  ],
-  "segment_050.mov": [
-    [
-      "crying",
-      0.0032352209091186523
-    ],
-    [
-      "beatboxing",
-      0.002813587663695216
-    ],
-    [
-      "\"fixing hair\"",
-      0.002721605822443962
-    ],
-    [
-      "\"pumping fist\"",
-      0.002583273220807314
-    ],
-    [
-      "\"brushing hair\"",
-      0.0025578502099961042
-    ]
-  ],
-  "segment_051.mov": [
-    [
-      "\"passing American football (in game)\"",
-      0.005108952056616545
-    ],
-    [
-      "\"passing American football (not in game)\"",
-      0.002873431658372283
-    ],
-    [
-      "\"side kick\"",
-      0.002661630278453231
-    ],
-    [
-      "\"kicking field goal\"",
-      0.0026403891388326883
-    ],
-    [
-      "\"high kick\"",
-      0.0025104281958192587
-    ]
-  ],
-  "segment_052.mov": [
-    [
-      "\"passing American football (in game)\"",
-      0.0058504920452833176
-    ],
-    [
-      "\"kicking field goal\"",
-      0.0027115345001220703
-    ],
-    [
-      "\"passing American football (not in game)\"",
-      0.0026032112073153257
-    ],
-    [
-      "\"side kick\"",
-      0.0025289678014814854
-    ],
-    [
-      "\"high kick\"",
-      0.002494329586625099
-    ]
-  ],
-  "segment_053.mov": [
-    [
-      "\"passing American football (in game)\"",
-      0.006694582756608725
-    ],
-    [
-      "\"passing American football (not in game)\"",
-      0.0025113301817327738
-    ],
-    [
-      "\"side kick\"",
-      0.0024927803315222263
-    ],
-    [
-      "\"kicking field goal\"",
-      0.0024906357284635305
-    ],
-    [
-      "\"high kick\"",
-      0.002489699050784111
-    ]
-  ],
-  "segment_054.mov": [
-    [
-      "\"passing American football (in game)\"",
-      0.006701494567096233
-    ],
-    [
-      "\"passing American football (not in game)\"",
-      0.0025095203891396523
-    ],
-    [
-      "\"kicking field goal\"",
-      0.0024933733511716127
-    ],
-    [
-      "\"side kick\"",
-      0.002489532344043255
-    ],
-    [
-      "headbutting",
-      0.002489422680810094
-    ]
-  ],
-  "segment_055.mov": [
-    [
-      "\"passing American football (in game)\"",
-      0.005614574532955885
-    ],
-    [
-      "\"kicking field goal\"",
-      0.002844724338501692
-    ],
-    [
-      "\"tossing coin\"",
-      0.0025539705529809
-    ],
-    [
-      "\"passing American football (not in game)\"",
-      0.0025522371288388968
-    ],
-    [
-      "headbutting",
-      0.0024937023408710957
-    ]
-  ],
-  "segment_056.mov": [
-    [
-      "headbutting",
-      0.003259368008002639
-    ],
-    [
-      "\"pumping fist\"",
-      0.0029684980399906635
-    ],
-    [
-      "\"playing ice hockey\"",
-      0.0029396419413387775
-    ],
-    [
-      "\"shaking hands\"",
-      0.002817715285345912
-    ],
-    [
-      "\"passing American football (in game)\"",
-      0.002679395256564021
-    ]
-  ],
-  "segment_057.mov": [
-    [
-      "bobsledding",
-      0.004842189606279135
-    ],
-    [
-      "\"pumping fist\"",
-      0.002615878591313958
-    ],
-    [
-      "\"driving car\"",
-      0.002567970659583807
-    ],
-    [
-      "\"playing poker\"",
-      0.0025621275417506695
-    ],
-    [
-      "\"playing ice hockey\"",
-      0.002544831018894911
-    ]
-  ],
-  "segment_058.mov": [
-    [
-      "motorcycling",
-      0.002834487706422806
-    ],
-    [
-      "\"driving car\"",
-      0.002728639403358102
-    ],
-    [
-      "\"changing wheel\"",
-      0.0027119864244014025
-    ],
-    [
-      "bobsledding",
-      0.002644300926476717
-    ],
-    [
-      "kissing",
-      0.002592987846583128
-    ]
-  ],
-  "segment_059.mov": [
-    [
-      "bobsledding",
-      0.004265598952770233
-    ],
-    [
-      "\"playing ice hockey\"",
-      0.0030175303108990192
-    ],
-    [
-      "\"playing paintball\"",
-      0.002638093428686261
-    ],
-    [
-      "motorcycling",
-      0.0025540166534483433
-    ],
-    [
-      "\"riding mountain bike\"",
-      0.002545550698414445
-    ]
-  ],
-  "segment_060.mov": [
-    [
-      "\"passing American football (in game)\"",
-      0.005962858442217112
-    ],
-    [
-      "\"kicking field goal\"",
-      0.002803174778819084
-    ],
-    [
-      "\"passing American football (not in game)\"",
-      0.0025008893571794033
-    ],
-    [
-      "\"side kick\"",
-      0.0024952334351837635
-    ],
-    [
-      "headbutting",
-      0.002492264611646533
-    ]
-  ],
-  "segment_061.mov": [
-    [
-      "\"passing American football (in game)\"",
-      0.006750195287168026
-    ],
-    [
-      "\"kicking field goal\"",
-      0.0024920343421399593
-    ],
-    [
-      "\"passing American football (not in game)\"",
-      0.002491830848157406
-    ],
-    [
-      "\"side kick\"",
-      0.0024899402633309364
-    ],
-    [
-      "\"high kick\"",
-      0.002489452948793769
-    ]
-  ],
-  "segment_062.mov": [
-    [
-      "\"passing American football (in game)\"",
-      0.006219817325472832
-    ],
-    [
-      "\"passing American football (not in game)\"",
-      0.002605071058496833
-    ],
-    [
-      "headbutting",
-      0.0025401373859494925
-    ],
-    [
-      "\"kicking field goal\"",
-      0.0025001864414662123
-    ],
-    [
-      "\"pumping fist\"",
-      0.002497575245797634
-    ]
-  ],
-  "segment_063.mov": [
-    [
-      "marching",
-      0.003558160038664937
-    ],
-    [
-      "\"passing American football (in game)\"",
-      0.0030603946652263403
-    ],
-    [
-      "\"tango dancing\"",
-      0.00266551086679101
-    ],
-    [
-      "\"side kick\"",
-      0.002660425379872322
-    ],
-    [
-      "\"passing American football (not in game)\"",
-      0.002627450041472912
-    ]
-  ],
-  "segment_064.mov": [
-    [
-      "\"passing American football (in game)\"",
-      0.00557365408167243
-    ],
-    [
-      "\"passing American football (not in game)\"",
-      0.0027207054663449526
-    ],
-    [
-      "headbutting",
-      0.0026054433546960354
-    ],
-    [
-      "\"pumping fist\"",
-      0.00251823291182518
-    ],
-    [
-      "\"kicking field goal\"",
-      0.002515583299100399
-    ]
-  ],
-  "segment_065.mov": [
-    [
-      "\"side kick\"",
-      0.005902828648686409
-    ],
-    [
-      "\"high kick\"",
-      0.0025668020825833082
-    ],
-    [
-      "wrestling",
-      0.0025415299460291862
-    ],
-    [
-      "squat",
-      0.0025292884092777967
-    ],
-    [
-      "\"drop kicking\"",
-      0.00251045823097229
-    ]
-  ],
-  "segment_066.mov": [
-    [
-      "bobsledding",
-      0.003806664375588298
-    ],
-    [
-      "\"changing wheel\"",
-      0.0029659303836524487
-    ],
-    [
-      "\"punching bag\"",
-      0.0027637933380901814
-    ],
-    [
-      "headbutting",
-      0.00258650747127831
-    ],
-    [
-      "motorcycling",
-      0.002585215028375387
-    ]
-  ],
-  "segment_067.mov": [
-    [
-      "\"changing wheel\"",
-      0.0038947509601712227
-    ],
-    [
-      "\"pushing car\"",
-      0.002689485903829336
-    ],
-    [
-      "\"driving car\"",
-      0.00266862940043211
-    ],
-    [
-      "\"changing oil\"",
-      0.0026266295462846756
-    ],
-    [
-      "\"arm wrestling\"",
-      0.0025696069933474064
-    ]
-  ],
-  "segment_068.mov": [
-    [
-      "\"passing American football (not in game)\"",
-      0.0030197601299732924
-    ],
-    [
-      "\"passing American football (in game)\"",
-      0.0029855112079530954
-    ],
-    [
-      "\"kicking field goal\"",
-      0.0027115403208881617
-    ],
-    [
-      "\"catching or throwing baseball\"",
-      0.0026096913497895002
-    ],
-    [
-      "\"high kick\"",
-      0.0026070410385727882
-    ]
-  ],
-  "segment_069.mov": [
-    [
-      "\"passing American football (in game)\"",
-      0.005361294839531183
-    ],
-    [
-      "\"kicking field goal\"",
-      0.002796306274831295
-    ],
-    [
-      "\"passing American football (not in game)\"",
-      0.0027009581681340933
-    ],
-    [
-      "\"side kick\"",
-      0.002553092548623681
-    ],
-    [
-      "\"high kick\"",
-      0.002506962278857827
-    ]
-  ],
-  "segment_070.mov": [
-    [
-      "\"passing American football (in game)\"",
-      0.005472357384860516
-    ],
-    [
-      "\"passing American football (not in game)\"",
-      0.002799208741635084
-    ],
-    [
-      "\"kicking field goal\"",
-      0.0027042534202337265
-    ],
-    [
-      "\"side kick\"",
-      0.0025148894637823105
-    ],
-    [
-      "\"catching or throwing baseball\"",
-      0.002496932167559862
-    ]
-  ],
-  "segment_071.mov": [
-    [
-      "\"side kick\"",
-      0.003893344197422266
-    ],
-    [
-      "\"passing American football (in game)\"",
-      0.0035547567531466484
-    ],
-    [
-      "\"kicking field goal\"",
-      0.002741483272984624
-    ],
-    [
-      "\"passing American football (not in game)\"",
-      0.0026810814160853624
-    ],
-    [
-      "\"catching or throwing baseball\"",
-      0.0025338695850223303
-    ]
-  ],
-  "segment_072.mov": [
-    [
-      "\"passing American football (in game)\"",
-      0.005563896149396896
-    ],
-    [
-      "\"passing American football (not in game)\"",
-      0.0027015593368560076
-    ],
-    [
-      "\"side kick\"",
-      0.002657951321452856
-    ],
-    [
-      "\"kicking field goal\"",
-      0.0026104655116796494
-    ],
-    [
-      "\"catching or throwing baseball\"",
-      0.0024953498505055904
-    ]
-  ],
-  "segment_073.mov": [
-    [
-      "\"passing American football (in game)\"",
-      0.006470989435911179
-    ],
-    [
-      "\"passing American football (not in game)\"",
-      0.0025515288580209017
-    ],
-    [
-      "\"side kick\"",
-      0.0025365715846419334
-    ],
-    [
-      "\"kicking field goal\"",
-      0.002492384286597371
-    ],
-    [
-      "\"high kick\"",
-      0.002490945626050234
-    ]
-  ],
-  "segment_074.mov": [
-    [
-      "\"passing American football (in game)\"",
-      0.004092047456651926
-    ],
-    [
-      "\"passing American football (not in game)\"",
-      0.0037548826076090336
-    ],
-    [
-      "\"side kick\"",
-      0.0026034193579107523
-    ],
-    [
-      "\"kicking field goal\"",
-      0.0025540809147059917
-    ],
-    [
-      "\"catching or throwing baseball\"",
-      0.0025116554461419582
-    ]
-  ],
-  "segment_075.mov": [
-    [
-      "\"passing American football (in game)\"",
-      0.006258341949433088
-    ],
-    [
-      "\"passing American football (not in game)\"",
-      0.002621858147904277
-    ],
-    [
-      "\"kicking field goal\"",
-      0.0025423644110560417
-    ],
-    [
-      "\"catching or throwing baseball\"",
-      0.0024970977101475
-    ],
-    [
-      "\"pumping fist\"",
-      0.0024926913902163506
-    ]
-  ],
-  "segment_076.mov": [
-    [
-      "\"hurling (sport)\"",
-      0.003916610963642597
-    ],
-    [
-      "headbutting",
-      0.0029018200002610683
-    ],
-    [
-      "\"passing American football (in game)\"",
-      0.0027735705953091383
-    ],
-    [
-      "\"passing American football (not in game)\"",
-      0.00273358216509223
-    ],
-    [
-      "\"kicking field goal\"",
-      0.0026465952396392822
-    ]
-  ],
-  "segment_077.mov": []
+  ]
 }

inference.py

@@ -69,7 +69,67 @@ def preprocess(frames, num_samples: int = 16, size: int = 224) -> torch.Tensor:
     clip = (clip - mean) / std
     return clip
 
-# === Prediction ===
+# === Play State Detection ===
+def analyze_play_state(predictions, confidence_threshold=0.002):
+    """
+    Analyze predictions to determine if this is likely start, middle, or end of play
+    """
+    if not predictions:
+        return "unknown", 0.0
+    
+    # Football-specific labels that indicate different play states
+    play_start_indicators = [
+        "passing American football (in game)",
+        "kicking field goal", 
+        "side kick",
+        "high kick"
+    ]
+    
+    play_action_indicators = [
+        "catching or throwing baseball",  # Often misclassified football throws
+        "catching or throwing softball",
+        "playing cricket",  # Sometimes catches football
+        "throwing ball"
+    ]
+    
+    non_play_indicators = [
+        "applauding",
+        "marching", 
+        "sitting",
+        "standing",
+        "clapping",
+        "cheering"
+    ]
+    
+    # Calculate weighted scores for each state
+    play_start_score = 0.0
+    play_action_score = 0.0
+    non_play_score = 0.0
+    
+    for label, confidence in predictions:
+        # Remove quotes and normalize
+        clean_label = label.replace('"', '').lower()
+        
+        if any(indicator.lower() in clean_label for indicator in play_start_indicators):
+            play_start_score += confidence * 2.0  # Weight play-specific actions higher
+        elif any(indicator.lower() in clean_label for indicator in play_action_indicators):
+            play_action_score += confidence
+        elif any(indicator.lower() in clean_label for indicator in non_play_indicators):
+            non_play_score += confidence
+    
+    # Determine play state
+    max_score = max(play_start_score, play_action_score, non_play_score)
+    
+    if max_score < confidence_threshold:
+        return "unknown", max_score
+    elif play_start_score == max_score:
+        return "play_active", play_start_score
+    elif play_action_score == max_score:
+        return "play_action", play_action_score  
+    else:
+        return "non_play", non_play_score
+
+# === Enhanced Prediction ===
 def predict_clip(path: str):
     try:
         video = EncodedVideo.from_path(path)
@@ -91,24 +151,102 @@ def predict_clip(path: str):
             idx = idx_tensor.item()
             label = kinetics_labels.get(idx, f"Class_{idx}")
             results.append((label, float(score)))
+        
+        # Add play state analysis
+        play_state, confidence = analyze_play_state(results)
+        print(f"[PLAY STATE] {play_state} (confidence: {confidence:.3f})")
+        
         return results
     except Exception as e:
         print(f"[ERROR] Failed to process {path}: {e}")
         return []
 
+# === Advanced Play Sequence Detection ===
+def detect_play_boundaries(clip_results, window_size=3):
+    """
+    Analyze sequence of clips to detect play start and end boundaries
+    """
+    if len(clip_results) < window_size:
+        return []
+    
+    play_states = []
+    for results in clip_results:
+        state, confidence = analyze_play_state(results)
+        play_states.append((state, confidence))
+    
+    boundaries = []
+    
+    # Look for transitions from non_play -> play_active (START)
+    # and play_active -> non_play (END)
+    for i in range(len(play_states) - 1):
+        current_state, current_conf = play_states[i]
+        next_state, next_conf = play_states[i + 1]
+        
+        if (current_state in ["non_play", "unknown"] and 
+            next_state in ["play_active", "play_action"] and
+            next_conf > 0.002):
+            boundaries.append(("play_start", i + 1, next_conf))
+            
+        if (current_state in ["play_active", "play_action"] and
+            next_state in ["non_play", "unknown"] and
+            current_conf > 0.002):
+            boundaries.append(("play_end", i, current_conf))
+    
+    return boundaries
+
 # === Audio ASR Setup ===
+# Using Whisper Medium for much better transcription quality
 asr = pipeline(
     "automatic-speech-recognition",
-    model="facebook/wav2vec2-base-960h",
-    chunk_length_s=10,
-    device=-1
+    model="openai/whisper-medium",  # Larger model for better accuracy
+    device=-1,
+    # Optimized for English sports content
+    generate_kwargs={
+        "language": "en",  # Force English to avoid multilingual detection
+        "task": "transcribe"  # Don't translate
+    }
 )
 
+# === Sports-Specific Vocabulary ===
+NFL_SPORTS_CONTEXT = [
+    # NFL Teams
+    "Patriots", "Cowboys", "Packers", "49ers", "Chiefs", "Bills", "Ravens", "Steelers",
+    "Eagles", "Giants", "Rams", "Saints", "Buccaneers", "Panthers", "Falcons", "Cardinals",
+    "Seahawks", "Broncos", "Raiders", "Chargers", "Dolphins", "Jets", "Colts", "Titans",
+    "Jaguars", "Texans", "Browns", "Bengals", "Lions", "Vikings", "Bears", "Commanders",
+    
+    # NFL Positions
+    "quarterback", "QB", "running back", "wide receiver", "tight end", "offensive line",
+    "defensive end", "linebacker", "cornerback", "safety", "kicker", "punter",
+    "center", "guard", "tackle", "fullback", "nose tackle", "middle linebacker",
+    
+    # NFL Terminology
+    "touchdown", "field goal", "first down", "second down", "third down", "fourth down",
+    "punt", "fumble", "interception", "sack", "blitz", "snap", "hike", "audible",
+    "red zone", "end zone", "yard line", "scrimmage", "pocket", "shotgun formation",
+    "play action", "screen pass", "draw play", "bootleg", "rollout", "scramble",
+    "timeout", "penalty", "flag", "holding", "false start", "offside", "encroachment",
+    "pass interference", "roughing the passer", "illegal formation", "delay of game",
+    
+    # Game Situations
+    "two minute warning", "overtime", "coin toss", "kickoff", "touchback", "fair catch",
+    "onside kick", "safety", "conversion", "extra point", "two point conversion",
+    "challenge", "replay", "incomplete", "completion", "rushing yards", "passing yards",
+    
+    # Common Commentary
+    "yards to go", "first and ten", "goal line", "midfield", "hash marks",
+    "pocket presence", "arm strength", "accuracy", "mobility", "coverage",
+    "rush defense", "pass rush", "secondary", "offensive line", "defensive line"
+]
+
+NFL_PROMPT = " ".join(NFL_SPORTS_CONTEXT[:20])  # Use first 20 terms as prompt context
+
 # === Audio Loading via FFmpeg ===
 def load_audio(path: str, sr: int = 16000):
-    """Decode audio track from any container using FFmpeg to PCM float32 array"""
+    """Enhanced audio loading with filtering and normalization"""
     cmd = [
         "ffmpeg", "-i", path,
+        "-af", "highpass=f=80,lowpass=f=8000,loudnorm",  # Filter noise and normalize
         "-f", "s16le",
         "-acodec", "pcm_s16le",
         "-ac", "1", "-ar", str(sr),
@@ -117,13 +255,198 @@ def load_audio(path: str, sr: int = 16000):
     proc = subprocess.Popen(cmd, stdout=subprocess.PIPE, stderr=subprocess.DEVNULL)
     raw = proc.stdout.read()
     audio = np.frombuffer(raw, np.int16).astype(np.float32) / 32768.0
+    
+    # Normalize audio levels
+    if len(audio) > 0:
+        audio = audio / (np.max(np.abs(audio)) + 1e-8)
+    
     return audio, sr
 
-# === Transcription ===
+# === Enhanced Transcription with Sports Context ===
 def transcribe_clip(path: str) -> str:
     audio, sr = load_audio(path, sr=16000)
-    result = asr(audio)
-    return result.get("text", "")
+    
+    # Skip transcription if audio is too short or quiet
+    if len(audio) < sr * 0.5:  # Less than 0.5 seconds
+        return ""
+    
+    if np.max(np.abs(audio)) < 0.01:  # Very quiet audio
+        return ""
+    
+    try:
+        # Use enhanced generation parameters for better NFL transcription
+        result = asr(
+            audio,
+            generate_kwargs={
+                "language": "en",
+                "task": "transcribe",
+                "no_repeat_ngram_size": 3,  # Reduce repetitive transcriptions
+                "temperature": 0.0,  # More deterministic output
+                "do_sample": False,  # Use greedy decoding for consistency
+                "num_beams": 1  # Faster single-beam search
+            }
+        )
+        text = result.get("text", "").strip()
+        
+        # Clean up common Whisper artifacts
+        text = text.replace("[BLANK_AUDIO]", "").strip()
+        text = text.replace("♪", "").strip()  # Remove music notes
+        
+        # Post-process with sports-specific corrections
+        text = apply_sports_corrections(text)
+        
+        return text
+    except Exception as e:
+        print(f"[WARN] Transcription failed: {e}")
+        return ""
+
+def apply_sports_corrections(text: str) -> str:
+    """Apply NFL-specific text corrections and standardization"""
+    import re
+    
+    if not text:
+        return text
+    
+    # Common misheard sports terms corrections
+    corrections = {
+        # Position corrections
+        r'\bqb\b': "QB",
+        r'\bquarter back\b': "quarterback", 
+        r'\bwide receiver\b': "wide receiver",
+        r'\btight end\b': "tight end",
+        r'\brunning back\b': "running back",
+        r'\bline backer\b': "linebacker",
+        r'\bcorner back\b': "cornerback",
+        
+        # Play corrections
+        r'\btouch down\b': "touchdown",
+        r'\bfield goal\b': "field goal", 
+        r'\bfirst down\b': "first down",
+        r'\bsecond down\b': "second down",
+        r'\bthird down\b': "third down",
+        r'\bfourth down\b': "fourth down",
+        r'\byard line\b': "yard line",
+        r'\bend zone\b': "end zone",
+        r'\bred zone\b': "red zone",
+        r'\btwo minute warning\b': "two minute warning",
+        
+        # Common misheard words
+        r'\bfourty\b': "forty",
+        r'\bfourty yard\b': "forty yard",
+        r'\btwenny\b': "twenty",
+        r'\bthirty yard\b': "thirty yard",
+        
+        # Numbers/downs that are often misheard
+        r'\b1st\b': "first",
+        r'\b2nd\b': "second", 
+        r'\b3rd\b': "third",
+        r'\b4th\b': "fourth",
+        r'\b1st and 10\b': "first and ten",
+        r'\b2nd and long\b': "second and long",
+        r'\b3rd and short\b': "third and short",
+        
+        # Team name corrections (common mishears)
+        r'\bforty niners\b': "49ers",
+        r'\bforty-niners\b': "49ers", 
+        r'\bsan francisco\b': "49ers",
+        r'\bnew england\b': "Patriots",
+        r'\bgreen bay\b': "Packers",
+        r'\bkansas city\b': "Chiefs",
+        r'\bnew york giants\b': "Giants",
+        r'\bnew york jets\b': "Jets",
+        r'\blos angeles rams\b': "Rams",
+        r'\blos angeles chargers\b': "Chargers",
+        
+        # Yard markers and positions
+        r'\b10 yard line\b': "ten yard line",
+        r'\b20 yard line\b': "twenty yard line",
+        r'\b30 yard line\b': "thirty yard line",
+        r'\b40 yard line\b': "forty yard line", 
+        r'\b50 yard line\b': "fifty yard line",
+        r'\bmid field\b': "midfield",
+        r'\bgoal line\b': "goal line",
+        
+        # Penalties and flags
+        r'\bfalse start\b': "false start",
+        r'\boff side\b': "offside",
+        r'\bpass interference\b': "pass interference",
+        r'\broughing the passer\b': "roughing the passer",
+        r'\bdelay of game\b': "delay of game",
+        
+        # Common play calls
+        r'\bplay action\b': "play action",
+        r'\bscreen pass\b': "screen pass",
+        r'\bdraw play\b': "draw play",
+        r'\bboot leg\b': "bootleg",
+        r'\broll out\b': "rollout",
+        r'\bshot gun\b': "shotgun",
+        r'\bno huddle\b': "no huddle"
+    }
+    
+    # Apply corrections (case-insensitive)
+    corrected_text = text
+    for pattern, replacement in corrections.items():
+        corrected_text = re.sub(pattern, replacement, corrected_text, flags=re.IGNORECASE)
+    
+    # Capitalize specific terms that should always be capitalized
+    capitalize_terms = ["NFL", "QB", "Patriots", "Cowboys", "Chiefs", "Packers", "49ers", 
+                       "Eagles", "Giants", "Rams", "Saints", "Bills", "Ravens", "Steelers"]
+    for term in capitalize_terms:
+        pattern = r'\b' + re.escape(term.lower()) + r'\b'
+        corrected_text = re.sub(pattern, term, corrected_text, flags=re.IGNORECASE)
+    
+    # Enhanced spell-check for common NFL terms using fuzzy matching
+    corrected_text = fuzzy_sports_corrections(corrected_text)
+    
+    return corrected_text.strip()
+
+def fuzzy_sports_corrections(text: str) -> str:
+    """Apply fuzzy matching for sports terms that might be slightly misheard"""
+    import re
+    
+    # Define common NFL terms and their likely misspellings
+    fuzzy_corrections = {
+        # Team names with common misspellings
+        r'\bpatriots?\b': "Patriots",
+        r'\bcowboys?\b': "Cowboys", 
+        r'\bpackers?\b': "Packers",
+        r'\bchief\b': "Chiefs",
+        r'\bchiefs?\b': "Chiefs",
+        r'\beagles?\b': "Eagles",
+        r'\bgiants?\b': "Giants",
+        r'\brams?\b': "Rams",
+        r'\bsaints?\b': "Saints",
+        
+        # Positions with common variations
+        r'\bquarterbacks?\b': "quarterback",
+        r'\brunning backs?\b': "running back",
+        r'\bwide receivers?\b': "wide receiver",
+        r'\btight ends?\b': "tight end",
+        r'\blinebackers?\b': "linebacker",
+        r'\bcornerbacks?\b': "cornerback",
+        r'\bsafety\b': "safety",
+        r'\bsafeties\b': "safety",
+        
+        # Play terms
+        r'\btouchdowns?\b': "touchdown",
+        r'\bfield goals?\b': "field goal",
+        r'\binterceptions?\b': "interception",
+        r'\bfumbles?\b': "fumble",
+        r'\bsacks?\b': "sack",
+        r'\bpunt\b': "punt",
+        r'\bpunts?\b': "punt",
+        
+        # Numbers and yards
+        r'\byards?\b': "yard",
+        r'\byards? line\b': "yard line",
+        r'\byard lines?\b': "yard line"
+    }
+    
+    corrected = text
+    for pattern, replacement in fuzzy_corrections.items():
+        corrected = re.sub(pattern, replacement, corrected, flags=re.IGNORECASE)
+    
+    return corrected
 
 # === CLI ===
 if __name__ == "__main__":

requirements.txt

@@ -4,3 +4,5 @@ pytorchvideo
 transformers
 huggingface_hub
 ffmpeg-python
+# For better audio transcription
+openai-whisper

run_all_clips.py

@@ -7,10 +7,10 @@ results to separate JSON files as they're generated.
 import os
 import glob
 import json
-from inference import predict_clip, transcribe_clip
+from inference import predict_clip, transcribe_clip, detect_play_boundaries, analyze_play_state
 
 
-def main(input_dir: str = "data", classification_file: str = "classification.json", transcript_file: str = "transcripts.json"):
+def main(input_dir: str = "data", classification_file: str = "classification.json", transcript_file: str = "transcripts.json", play_analysis_file: str = "play_analysis.json"):
     # find video files
     patterns = ["*.mov", "*.mp4"]
     clips = []
@@ -23,6 +23,7 @@ def main(input_dir: str = "data", classification_file: str = "classification.jso
 
     classification_results = {}
     transcript_results = {}
+    play_analysis_results = {"clips": [], "play_boundaries": [], "summary": {}}
     
     # Ensure output files exist with empty JSON if first write fails
     try:
@@ -36,6 +37,15 @@ def main(input_dir: str = "data", classification_file: str = "classification.jso
             json.dump(transcript_results, f, indent=2)
     except Exception as e:
         print(f"[WARN] Could not initialize {transcript_file}: {e}")
+        
+    try:
+        with open(play_analysis_file, 'w') as f:
+            json.dump(play_analysis_results, f, indent=2)
+    except Exception as e:
+        print(f"[WARN] Could not initialize {play_analysis_file}: {e}")
+
+    # Store all clip results for sequence analysis
+    all_clip_results = []
 
     for clip in clips:
         clip_name = os.path.basename(clip)
@@ -45,6 +55,17 @@ def main(input_dir: str = "data", classification_file: str = "classification.jso
         print("Running video classification...")
         scores = predict_clip(clip)
         classification_results[clip_name] = scores
+        all_clip_results.append(scores)  # Store for sequence analysis
+        
+        # Analyze play state for this clip
+        play_state, confidence = analyze_play_state(scores)
+        clip_analysis = {
+            "filename": clip_name,
+            "play_state": play_state,
+            "confidence": confidence,
+            "classifications": scores
+        }
+        play_analysis_results["clips"].append(clip_analysis)
         
         # Print classification results to console
         if scores:
@@ -76,10 +97,61 @@ def main(input_dir: str = "data", classification_file: str = "classification.jso
                 json.dump(transcript_results, f, indent=2)
         except Exception as e:
             print(f"[ERROR] Failed to write {transcript_file}: {e}")
+            
+        try:
+            with open(play_analysis_file, 'w') as f:
+                json.dump(play_analysis_results, f, indent=2)
+        except Exception as e:
+            print(f"[ERROR] Failed to write {play_analysis_file}: {e}")
+
+    # Perform sequence analysis to detect play boundaries
+    print("\n=== ANALYZING PLAY SEQUENCES ===")
+    boundaries = detect_play_boundaries(all_clip_results)
+    play_analysis_results["play_boundaries"] = []
+    
+    for boundary_type, clip_index, confidence in boundaries:
+        clip_name = os.path.basename(clips[clip_index]) if clip_index < len(clips) else f"clip_{clip_index}"
+        boundary_info = {
+            "type": boundary_type,
+            "clip_index": clip_index,
+            "clip_name": clip_name,
+            "confidence": confidence
+        }
+        play_analysis_results["play_boundaries"].append(boundary_info)
+        print(f"[{boundary_type.upper()}] {clip_name} (confidence: {confidence:.3f})")
+    
+    # Generate summary statistics
+    play_states = [clip["play_state"] for clip in play_analysis_results["clips"]]
+    play_analysis_results["summary"] = {
+        "total_clips": len(clips),
+        "play_active_clips": play_states.count("play_active"),
+        "play_action_clips": play_states.count("play_action"), 
+        "non_play_clips": play_states.count("non_play"),
+        "unknown_clips": play_states.count("unknown"),
+        "detected_play_starts": len([b for b in boundaries if b[0] == "play_start"]),
+        "detected_play_ends": len([b for b in boundaries if b[0] == "play_end"])
+    }
+    
+    print(f"\n=== PLAY ANALYSIS SUMMARY ===")
+    summary = play_analysis_results["summary"]
+    print(f"Total clips processed: {summary['total_clips']}")
+    print(f"Play active clips: {summary['play_active_clips']}")
+    print(f"Play action clips: {summary['play_action_clips']}")
+    print(f"Non-play clips: {summary['non_play_clips']}")
+    print(f"Play starts detected: {summary['detected_play_starts']}")
+    print(f"Play ends detected: {summary['detected_play_ends']}")
+
+    # Final save of complete analysis
+    try:
+        with open(play_analysis_file, 'w') as f:
+            json.dump(play_analysis_results, f, indent=2)
+    except Exception as e:
+        print(f"[ERROR] Failed to write final {play_analysis_file}: {e}")
 
     print(f"\nFinished processing. Results saved to:")
     print(f"  - Classifications: '{classification_file}'")
     print(f"  - Transcripts: '{transcript_file}'")
+    print(f"  - Play Analysis: '{play_analysis_file}'")
 
 
 if __name__ == "__main__":

transcripts.json

@@ -1,79 +1,3 @@
 {
-  "segment_001.mov": "YOU  T WABOU GOT OM RIGT NOW",
-  "segment_002.mov": "THE SIX TE BEST  POLL BACK AND THE OT",
-  "segment_003.mov": "WELL WELL AN THAT'S JUST KILE",
-  "segment_004.mov": "Y BA ALF PASSING RID IT'S IT'S LI",
-  "segment_005.mov": "GOTA CONSIDER WHAT HE IS ES ATU",
-  "segment_006.mov": "TON AND BE PAD THAT USES AS",
-  "segment_007.mov": "LEGS AS WELL AS HE USES HIS ARM A",
-  "segment_008.mov": "THEY FOUND A GREAT COMMENATION",
-  "segment_009.mov": "WITH ME A DERI GENERY BUT",
-  "segment_010.mov": "TERGAN A SACK TO BOX AT T BANGLES",
-  "segment_011.mov": "O LIKE TE  HAVE TO PUT AITE THE BOX ALL DAY T",
-  "segment_012.mov": "O STOP THIS RUNYAT DON'T BE SURPRISED",
-  "segment_013.mov": "THAT E BANN HAS A PIG DAY THROUGH THE",
-  "segment_014.mov": "ER MEMORY AN",
-  "segment_015.mov": "FAR WORBIN THE EYE FOR",
-  "segment_016.mov": "SH AND NOW THEY SWORTS ALWAYS ABOUT SIX",
-  "segment_017.mov": "BOUNDS OF WARRY",
-  "segment_018.mov": "FVEIRY MOTION IS FOAR",
-  "segment_019.mov": "THAT GOES TO FOWERS JUC",
-  "segment_020.mov": "WHAT BY WI PAS AND A FO",
-  "segment_021.mov": "MY COLLAR AND OUT OF BOUNDS GOES",
-  "segment_022.mov": "AYFLOWERS WITH A JET SWEE",
-  "segment_023.mov": "AND ON THE POYE PICKS UP NO",
-  "segment_024.mov": "AN TO THE THIRTY NINE",
-  "segment_025.mov": "I MET OFFENSOF WING WAR",
-  "segment_026.mov": "NY STANLEY HAS BEEN TERRIFIC AT THE LEAP",
-  "segment_027.mov": "TACKLE HE'S HEALTHY HE REALLY HAS BEEN THE",
-  "segment_028.mov": "S BEEN SOME SHUFFLING AT THE GUARDS AS YOU SEE THAT",
-  "segment_029.mov": "ONED AROUND AND THE ROOKY ROSY GARDENAL",
-  "segment_030.mov": "WERE ON THE SIDE PATRICK REGARD",
-  "segment_031.mov": "FORTY TWO HE GETS OVERLOOKED ALOT",
-  "segment_032.mov": "UT NIGOES EVERYTHING FOR THE STAN IN",
-  "segment_033.mov": "I BAZILY LIKE AN EXTRA WISE",
-  "segment_034.mov": "NOR EXPERT IDIN ABOUT THEIRE LEADING",
-  "segment_035.mov": "HOLE POTIS ONEN A GREAT JU",
-  "segment_036.mov": "P A TOP TOPEN LIFE OR  TRECKEN WUP A",
-  "segment_037.mov": "SUPPORT NETHER SLUCK A DON AND",
-  "segment_038.mov": "THAT WHERE THE DIES",
-  "segment_039.mov": "",
-  "segment_040.mov": "AND A GANE OF FOR IS",
-  "segment_041.mov": "HE'S OUT TO THE FORTY THREE",
-  "segment_042.mov": "AND AT FIRST DOWN O",
-  "segment_043.mov": "S A LOOK AT FOND BELL OR MAKES THAT ST",
-  "segment_044.mov": "UP AND UNMET DEFENCE OF LIND",
-  "segment_045.mov": "THEY GET BACK CHRISJENKINS",
-  "segment_046.mov": "AND KILL HEDRICKS IN HIS HEALTH",
-  "segment_047.mov": "HOPBERED ON THE OTHER SIDE AYAD",
-  "segment_048.mov": "THAT THE DEFENCES HAS BEEN NOPIC",
-  "segment_049.mov": "LONG SOME THIS YEAR FOR THEY ARE GETTING HEALTHY AS YOU",
-  "segment_050.mov": "ENC O THAT GET BACMA KINLY JACKSIN TO THY THE",
-  "segment_051.mov": "T MILE S BORK  YE AL MISS  LA",
-  "segment_052.mov": "COPAL GAMES IT'S A BORGAR",
-  "segment_053.mov": "THAT WOOKED OTHER FROM EGGERING",
-  "segment_054.mov": "FO TE TOLL OF THAT",
-  "segment_055.mov": "TACKLEMAY BY BOBBE",
-  "segment_056.mov": "FORTY FIVE AND A GAIN AT TWO",
-  "segment_057.mov": "E'LL BE JACKIN DOWN AND ATE",
-  "segment_058.mov": "WELL IT'S GOING TO BE IMPORTANT ALL DAY",
-  "segment_059.mov": "ONG THE TACKLE ALOW YOU GOT TO BE ABLE",
-  "segment_060.mov": "TO TAKE THE LEGS OUT OF THEIR",
-  "segment_061.mov": "OR KINWRY IF YOU IF YOU GO UP TOO HIGH YOU KN",
-  "segment_062.mov": "W THAT STIP AN AND GETS IN THE OPEN FIELD TO",
-  "segment_063.mov": "BEAD SO EXCELLENT JOB THEIR",
-  "segment_064.mov": "R BY HILL AND HUBBARD TO BRING HIM DOWN",
-  "segment_065.mov": "",
-  "segment_066.mov": "SO WITH THE HEAD MI COLTEN IS",
-  "segment_067.mov": "T TO DAY ME AND FORGIDDN'T PRACTIV",
-  "segment_068.mov": "A WEEK D J TURN HER",
-  "segment_069.mov": "UTS LAKIN YER BE TEN TOBACC",
-  "segment_070.mov": "ALE A SHRIK IN TAKING HIS PLICE",
-  "segment_071.mov": "THAN THE FIRST POSSESSION THE DAY",
-  "segment_072.mov": "IM SECOND DOWN AND EI",
-  "segment_073.mov": "THAT WAT GOD IS ON THE MON",
-  "segment_074.mov": "JACKSON TO THE AIL",
-  "segment_075.mov": "SINGS ABOUT TE OL COL",
-  "segment_076.mov": "ERE MAKES THE GRAB",
-  "segment_077.mov": ""
+  "segment_001.mov": "I can tell you that Lamar Jackson right now is"
 }