from herpetologist import check_type
from malaya_speech.supervised import voice_conversion
from malaya_speech.utils import describe_availability
_availability = {
'fastvc-32-vggvox-v2': {
'Size (MB)': 190,
'Quantized Size (MB)': 54.1,
'Total loss': 0.2851,
},
'fastvc-64-vggvox-v2': {
'Size (MB)': 194,
'Quantized Size (MB)': 55.7,
'Total loss': 0.2764,
},
}
[docs]def available_deep_conversion():
"""
List available Voice Conversion models.
"""
return describe_availability(_availability)
[docs]def deep_conversion(
model: str = 'fastvc-32-vggvox-v2', quantized: bool = False, **kwargs
):
"""
Load Voice Conversion model.
Parameters
----------
model : str, optional (default='fastvc-32-vggvox-v2')
Check available models at `malaya_speech.voice_conversion.available_deep_conversion()`.
quantized : bool, optional (default=False)
if True, will load 8-bit quantized model.
Quantized model not necessary faster, totally depends on the machine.
Returns
-------
result : malaya_speech.model.synthesis.FastVC class
"""
model = model.lower()
if model not in _availability:
raise ValueError(
'model not supported, please check supported models from `malaya_speech.voice_conversion.available_deep_conversion()`.'
)
return voice_conversion.load(
model=model,
module='voice-conversion',
quantized=quantized,
**kwargs
)
