Title: Birdsong Species Identification (5-class)

Problem statement
Build a robust audio classification model that identifies the bird species from a 3-second monophonic WAV clip. Each clip belongs to exactly one of five species.

Public files and directories
- public/train.csv: Training annotations with two columns: id,label. The id is the WAV filename in public/train_audio/; label is the species common name.
- public/test.csv: Test identifiers with a single column: id. The id is the WAV filename in public/test_audio/.
- public/sample_submission.csv: A properly formatted sample submission with columns: id,label.
- public/train_audio/: Folder of training WAV clips named by anonymized ids.
- public/test_audio/: Folder of test WAV clips named by anonymized ids.

Target
- label: The common species name. Possible classes include examples such as American Robin, Bewick's Wren, Northern Cardinal, Northern Mockingbird, Song Sparrow (exact classes depend on the provided training set).

Evaluation
- Metric: Macro-averaged F1 score across the classes.
- Submission format: CSV with exactly two columns id,label. Rows must cover all ids from public/test.csv exactly once. Labels must be one of the class names present in the training data.

Notes
- No external data is required. You may use any feature extraction techniques or models that comply with platform rules.
- Do not rely on filenames to infer labels; ids are anonymized.
- Use the CSVs to enumerate the data; do not assume a fixed number of files.
