Spaces:
Running
on
T4
Running
on
T4
File size: 5,870 Bytes
1ce5e18 |
1 2 3 4 5 6 7 8 9 10 11 12 13 14 15 16 17 18 19 20 21 22 23 24 25 26 27 28 29 30 31 32 33 34 35 36 37 38 39 40 41 42 43 44 45 46 47 48 49 50 51 52 53 54 55 56 57 58 59 60 61 62 63 64 65 66 67 68 69 70 71 72 73 74 75 76 77 78 79 80 81 82 83 84 85 86 87 88 89 90 91 92 93 94 95 96 97 98 99 100 101 102 103 104 105 106 107 108 109 110 111 112 113 114 115 116 117 118 119 120 121 122 123 124 125 126 127 128 129 130 131 132 133 134 135 136 137 138 139 140 141 142 143 144 145 146 147 148 149 150 151 152 153 154 155 156 157 158 159 160 |
# Copyright 2023 The HuggingFace Team. All rights reserved.
#
# Licensed under the Apache License, Version 2.0 (the "License");
# you may not use this file except in compliance with the License.
# You may obtain a copy of the License at
#
# http://www.apache.org/licenses/LICENSE-2.0
#
# Unless required by applicable law or agreed to in writing, software
# distributed under the License is distributed on an "AS IS" BASIS,
# WITHOUT WARRANTIES OR CONDITIONS OF ANY KIND, either express or implied.
# See the License for the specific language governing permissions and
# limitations under the License.from typing import List, Union
from typing import List, Union
from ..utils import is_torch_available
from .base import Pipeline
if is_torch_available():
from ..models.auto.modeling_auto import MODEL_FOR_TEXT_TO_SPECTROGRAM_MAPPING
from ..models.speecht5.modeling_speecht5 import SpeechT5HifiGan
DEFAULT_VOCODER_ID = "microsoft/speecht5_hifigan"
class TextToAudioPipeline(Pipeline):
"""
Text-to-audio generation pipeline using any `AutoModelForTextToWaveform` or `AutoModelForTextToSpectrogram`. This
pipeline generates an audio file from an input text and optional other conditional inputs.
Example:
```python
>>> from transformers import pipeline
>>> pipe = pipeline(model="suno/bark-small")
>>> output = pipe("Hey it's HuggingFace on the phone!")
>>> audio = output["audio"]
>>> sampling_rate = output["sampling_rate"]
```
Learn more about the basics of using a pipeline in the [pipeline tutorial](../pipeline_tutorial)
This pipeline can currently be loaded from [`pipeline`] using the following task identifiers: `"text-to-speech"` or
`"text-to-audio"`.
See the list of available models on [huggingface.co/models](https://huggingface.co/models?filter=text-to-speech).
"""
def __init__(self, *args, vocoder=None, sampling_rate=None, **kwargs):
super().__init__(*args, **kwargs)
if self.framework == "tf":
raise ValueError("The TextToAudioPipeline is only available in PyTorch.")
self.vocoder = None
if self.model.__class__ in MODEL_FOR_TEXT_TO_SPECTROGRAM_MAPPING.values():
self.vocoder = (
SpeechT5HifiGan.from_pretrained(DEFAULT_VOCODER_ID).to(self.model.device)
if vocoder is None
else vocoder
)
self.sampling_rate = sampling_rate
if self.vocoder is not None:
self.sampling_rate = self.vocoder.config.sampling_rate
if self.sampling_rate is None:
# get sampling_rate from config and generation config
config = self.model.config
gen_config = self.model.__dict__.get("generation_config", None)
if gen_config is not None:
config.update(gen_config.to_dict())
for sampling_rate_name in ["sample_rate", "sampling_rate"]:
sampling_rate = getattr(config, sampling_rate_name, None)
if sampling_rate is not None:
self.sampling_rate = sampling_rate
def preprocess(self, text, **kwargs):
if isinstance(text, str):
text = [text]
if self.model.config.model_type == "bark":
# bark Tokenizer is called with BarkProcessor which uses those kwargs
new_kwargs = {
"max_length": self.model.generation_config.semantic_config.get("max_input_semantic_length", 256),
"add_special_tokens": False,
"return_attention_mask": True,
"return_token_type_ids": False,
"padding": "max_length",
}
# priority is given to kwargs
new_kwargs.update(kwargs)
kwargs = new_kwargs
output = self.tokenizer(text, **kwargs, return_tensors="pt")
return output
def _forward(self, model_inputs, **kwargs):
# we expect some kwargs to be additional tensors which need to be on the right device
kwargs = self._ensure_tensor_on_device(kwargs, device=self.device)
if self.model.can_generate():
output = self.model.generate(**model_inputs, **kwargs)
else:
output = self.model(**model_inputs, **kwargs)[0]
if self.vocoder is not None:
# in that case, the output is a spectrogram that needs to be converted into a waveform
output = self.vocoder(output)
return output
def __call__(self, text_inputs: Union[str, List[str]], **forward_params):
"""
Generates speech/audio from the inputs. See the [`TextToAudioPipeline`] documentation for more information.
Args:
text_inputs (`str` or `List[str]`):
The text(s) to generate.
forward_params (*optional*):
Parameters passed to the model generation/forward method.
Return:
A `dict` or a list of `dict`: The dictionaries have two keys:
- **audio** (`np.ndarray` of shape `(nb_channels, audio_length)`) -- The generated audio waveform.
- **sampling_rate** (`int`) -- The sampling rate of the generated audio waveform.
"""
return super().__call__(text_inputs, **forward_params)
def _sanitize_parameters(
self,
preprocess_params=None,
forward_params=None,
):
if preprocess_params is None:
preprocess_params = {}
if forward_params is None:
forward_params = {}
postprocess_params = {}
return preprocess_params, forward_params, postprocess_params
def postprocess(self, waveform):
output_dict = {}
output_dict["audio"] = waveform.cpu().float().numpy()
output_dict["sampling_rate"] = self.sampling_rate
return output_dict
|