from malaya_speech.supervised import vocoder
from herpetologist import check_type
_melgan_availability = {
'male': {
'Size (MB)': 17.3,
'Quantized Size (MB)': 4.53,
'Mel loss': 0.4443,
},
'female': {
'Size (MB)': 17.3,
'Quantized Size (MB)': 4.53,
'Mel loss': 0.4434,
},
'husein': {
'Size (MB)': 17.3,
'Quantized Size (MB)': 4.53,
'Mel loss': 0.4442,
},
'haqkiem': {
'Size (MB)': 17.3,
'Quantized Size (MB)': 4.53,
'Mel loss': 0.4819,
},
'yasmin': {
'Size (MB)': 17.3,
'Quantized Size (MB)': 4.53,
'Mel loss': 0.4867,
},
'osman': {
'Size (MB)': 17.3,
'Quantized Size (MB)': 4.53,
'Mel loss': 0.4819,
},
'universal': {
'Size (MB)': 309,
'Quantized Size (MB)': 77.5,
'Mel loss': 0.4463,
},
'universal-1024': {
'Size (MB)': 78.4,
'Quantized Size (MB)': 19.9,
'Mel loss': 0.4591,
},
'universal-384': {
'Size (MB)': 11.3,
'Quantized Size (MB)': 3.06,
'Mel loss': 0.4445,
},
}
_mbmelgan_availability = {
'female': {
'Size (MB)': 10.4,
'Quantized Size (MB)': 2.82,
'Mel loss': 0.4356,
},
'male': {
'Size (MB)': 10.4,
'Quantized Size (MB)': 2.82,
'Mel loss': 0.3735,
},
'husein': {
'Size (MB)': 10.4,
'Quantized Size (MB)': 2.82,
'Mel loss': 0.4356,
},
'haqkiem': {
'Size (MB)': 10.4,
'Quantized Size (MB)': 2.82,
'Mel loss': 0.4192,
},
}
_hifigan_availability = {
'male': {
'Size (MB)': 8.8,
'Quantized Size (MB)': 2.49,
'Mel loss': 0.465,
},
'female': {
'Size (MB)': 8.8,
'Quantized Size (MB)': 2.49,
'Mel loss': 0.5547,
},
'universal-1024': {
'Size (MB)': 170,
'Quantized Size (MB)': 42.9,
'Mel loss': 0.3346,
},
'universal-768': {
'Size (MB)': 72.8,
'Quantized Size (MB)': 18.5,
'Mel loss': 0.3617,
},
'universal-512': {
'Size (MB)': 32.6,
'Quantized Size (MB)': 8.6,
'Mel loss': 0.3253,
},
}
[docs]def available_melgan():
"""
List available MelGAN Mel-to-Speech models.
"""
from malaya_speech.utils import describe_availability
return describe_availability(_melgan_availability)
[docs]def available_mbmelgan():
"""
List available Multiband MelGAN Mel-to-Speech models.
"""
from malaya_speech.utils import describe_availability
return describe_availability(_mbmelgan_availability)
[docs]def available_hifigan():
"""
List available HiFiGAN Mel-to-Speech models.
"""
from malaya_speech.utils import describe_availability
return describe_availability(_hifigan_availability)
[docs]@check_type
def melgan(model: str = 'universal-1024', quantized: bool = False, **kwargs):
"""
Load MelGAN Vocoder model.
Parameters
----------
model : str, optional (default='universal-1024')
Model architecture supported. Allowed values:
* ``'female'`` - MelGAN trained on female voice.
* ``'male'`` - MelGAN trained on male voice.
* ``'husein'`` - MelGAN trained on Husein voice, https://www.linkedin.com/in/husein-zolkepli/
* ``'haqkiem'`` - MelGAN trained on Haqkiem voice, https://www.linkedin.com/in/haqkiem-daim/
* ``'yasmin'`` - MelGAN trained on female Yasmin voice.
* ``'osman'`` - MelGAN trained on male Osman voice.
* ``'female-singlish'`` - MelGAN trained on Female Singlish voice, https://www.imda.gov.sg/programme-listing/digital-services-lab/national-speech-corpus
* ``'universal'`` - Universal MelGAN trained on multiple speakers.
* ``'universal-1024'`` - Universal MelGAN with 1024 filters trained on multiple speakers.
* ``'universal-384'`` - Universal MelGAN with 384 filters trained on multiple speakers.
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.Vocoder class
"""
model = model.lower()
if model not in _melgan_availability:
raise ValueError(
'model not supported, please check supported models from `malaya_speech.vocoder.available_melgan()`.'
)
return vocoder.load(
model=model,
module='vocoder-melgan',
quantized=quantized,
**kwargs
)
[docs]@check_type
def mbmelgan(model: str = 'female', quantized: bool = False, **kwargs):
"""
Load Multiband MelGAN Vocoder model.
Parameters
----------
model : str, optional (default='female')
Model architecture supported. Allowed values:
* ``'female'`` - MBMelGAN trained on female voice.
* ``'male'`` - MBMelGAN trained on male voice.
* ``'husein'`` - MBMelGAN trained on Husein voice, https://www.linkedin.com/in/husein-zolkepli/
* ``'haqkiem'`` - MBMelGAN trained on Haqkiem voice, https://www.linkedin.com/in/haqkiem-daim/
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.Vocoder class
"""
model = model.lower()
if model not in _mbmelgan_availability:
raise ValueError(
'model not supported, please check supported models from `malaya_speech.vocoder.available_mbmelgan()`.'
)
return vocoder.load(
model=model,
module='vocoder-mbmelgan',
quantized=quantized,
**kwargs
)
[docs]@check_type
def hifigan(model: str = 'universal-768', quantized: bool = False, **kwargs):
"""
Load HiFiGAN Vocoder model.
Parameters
----------
model : str, optional (default='universal-768')
Model architecture supported. Allowed values:
* ``'female'`` - HiFiGAN trained on female voice.
* ``'male'`` - HiFiGAN trained on male voice.
* ``'universal-1024'`` - Universal HiFiGAN with 1024 filters trained on multiple speakers.
* ``'universal-768'`` - Universal HiFiGAN with 768 filters trained on multiple speakers.
* ``'universal-512'`` - Universal HiFiGAN with 512 filters trained on multiple speakers.
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.Vocoder class
"""
model = model.lower()
if model not in _hifigan_availability:
raise ValueError(
'model not supported, please check supported models from `malaya_speech.vocoder.available_hifigan()`.'
)
return vocoder.load(
model=model,
module='vocoder-hifigan',
quantized=quantized,
**kwargs
)