Index
birdnet_stm32.models
¶
Model architectures, audio frontend layer, magnitude scaling, and inference runners.
Use :func:build_model to create a model by name::
from birdnet_stm32.models import build_model
model = build_model("dscnn", num_mels=64, spec_width=256, ...)
register_model(name)
¶
Decorator to register a model builder function.
The decorated function must accept keyword arguments and return an
uncompiled tf.keras.Model.
Parameters:
| Name | Type | Description | Default |
|---|---|---|---|
name
|
str
|
Canonical model name (e.g. "dscnn"). |
required |
Source code in birdnet_stm32/models/__init__.py
build_model(name, **kwargs)
¶
Build a model by registered name.
Parameters:
| Name | Type | Description | Default |
|---|---|---|---|
name
|
str
|
Model architecture name (e.g. "dscnn"). |
required |
**kwargs
|
Any
|
Forwarded to the model builder. |
{}
|
Returns:
| Type | Description |
|---|---|
Model
|
Uncompiled Keras model. |
Raises:
| Type | Description |
|---|---|
KeyError
|
If no model with the given name is registered. |