File size: 10,575 Bytes
b381930 |
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 99 100 101 102 103 104 105 106 107 108 109 110 111 112 113 114 115 116 117 118 119 120 121 122 123 124 125 126 127 128 129 130 131 132 133 134 135 136 137 138 139 140 141 142 143 144 145 146 147 148 149 150 151 152 153 154 155 156 157 158 159 160 161 162 163 164 165 166 167 168 169 170 171 172 173 174 175 176 177 178 179 180 181 182 183 184 185 186 187 188 189 190 191 192 193 194 195 196 197 198 199 200 201 202 203 204 205 206 207 208 209 210 211 212 213 214 215 216 217 218 219 220 221 222 223 224 225 226 227 228 229 230 231 232 233 234 235 236 237 238 239 240 241 242 243 244 245 246 247 248 249 250 251 252 253 254 255 256 257 258 259 260 261 262 263 264 265 266 267 268 269 270 271 272 273 274 275 276 277 278 279 280 281 282 |
# 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.
import unittest
import numpy as np
from transformers import (
MODEL_FOR_TEXT_TO_WAVEFORM_MAPPING,
AutoProcessor,
TextToAudioPipeline,
pipeline,
)
from transformers.testing_utils import (
is_pipeline_test,
require_torch,
require_torch_accelerator,
require_torch_or_tf,
slow,
torch_device,
)
from transformers.trainer_utils import set_seed
from .test_pipelines_common import ANY
@is_pipeline_test
@require_torch_or_tf
class TextToAudioPipelineTests(unittest.TestCase):
model_mapping = MODEL_FOR_TEXT_TO_WAVEFORM_MAPPING
# for now only test text_to_waveform and not text_to_spectrogram
@slow
@require_torch
def test_small_musicgen_pt(self):
music_generator = pipeline(task="text-to-audio", model="facebook/musicgen-small", framework="pt")
forward_params = {
"do_sample": False,
"max_new_tokens": 250,
}
outputs = music_generator("This is a test", forward_params=forward_params)
self.assertEqual({"audio": ANY(np.ndarray), "sampling_rate": 32000}, outputs)
# test two examples side-by-side
outputs = music_generator(["This is a test", "This is a second test"], forward_params=forward_params)
audio = [output["audio"] for output in outputs]
self.assertEqual([ANY(np.ndarray), ANY(np.ndarray)], audio)
# test batching
outputs = music_generator(
["This is a test", "This is a second test"], forward_params=forward_params, batch_size=2
)
audio = [output["audio"] for output in outputs]
self.assertEqual([ANY(np.ndarray), ANY(np.ndarray)], audio)
@slow
@require_torch
def test_medium_seamless_m4t_pt(self):
speech_generator = pipeline(task="text-to-audio", model="facebook/hf-seamless-m4t-medium", framework="pt")
for forward_params in [{"tgt_lang": "eng"}, {"return_intermediate_token_ids": True, "tgt_lang": "eng"}]:
outputs = speech_generator("This is a test", forward_params=forward_params)
self.assertEqual({"audio": ANY(np.ndarray), "sampling_rate": 16000}, outputs)
# test two examples side-by-side
outputs = speech_generator(["This is a test", "This is a second test"], forward_params=forward_params)
audio = [output["audio"] for output in outputs]
self.assertEqual([ANY(np.ndarray), ANY(np.ndarray)], audio)
# test batching
outputs = speech_generator(
["This is a test", "This is a second test"], forward_params=forward_params, batch_size=2
)
audio = [output["audio"] for output in outputs]
self.assertEqual([ANY(np.ndarray), ANY(np.ndarray)], audio)
@slow
@require_torch
def test_small_bark_pt(self):
speech_generator = pipeline(task="text-to-audio", model="suno/bark-small", framework="pt")
forward_params = {
# Using `do_sample=False` to force deterministic output
"do_sample": False,
"semantic_max_new_tokens": 100,
}
outputs = speech_generator("This is a test", forward_params=forward_params)
self.assertEqual(
{"audio": ANY(np.ndarray), "sampling_rate": 24000},
outputs,
)
# test two examples side-by-side
outputs = speech_generator(
["This is a test", "This is a second test"],
forward_params=forward_params,
)
audio = [output["audio"] for output in outputs]
self.assertEqual([ANY(np.ndarray), ANY(np.ndarray)], audio)
# test other generation strategy
forward_params = {
"do_sample": True,
"semantic_max_new_tokens": 100,
"semantic_num_return_sequences": 2,
}
outputs = speech_generator("This is a test", forward_params=forward_params)
audio = outputs["audio"]
self.assertEqual(ANY(np.ndarray), audio)
# test using a speaker embedding
processor = AutoProcessor.from_pretrained("suno/bark-small")
temp_inp = processor("hey, how are you?", voice_preset="v2/en_speaker_5")
history_prompt = temp_inp["history_prompt"]
forward_params["history_prompt"] = history_prompt
outputs = speech_generator(
["This is a test", "This is a second test"],
forward_params=forward_params,
batch_size=2,
)
audio = [output["audio"] for output in outputs]
self.assertEqual([ANY(np.ndarray), ANY(np.ndarray)], audio)
@slow
@require_torch_accelerator
def test_conversion_additional_tensor(self):
speech_generator = pipeline(task="text-to-audio", model="suno/bark-small", framework="pt", device=torch_device)
processor = AutoProcessor.from_pretrained("suno/bark-small")
forward_params = {
"do_sample": True,
"semantic_max_new_tokens": 100,
}
# atm, must do to stay coherent with BarkProcessor
preprocess_params = {
"max_length": 256,
"add_special_tokens": False,
"return_attention_mask": True,
"return_token_type_ids": False,
"padding": "max_length",
}
outputs = speech_generator(
"This is a test",
forward_params=forward_params,
preprocess_params=preprocess_params,
)
temp_inp = processor("hey, how are you?", voice_preset="v2/en_speaker_5")
history_prompt = temp_inp["history_prompt"]
forward_params["history_prompt"] = history_prompt
# history_prompt is a torch.Tensor passed as a forward_param
# if generation is successful, it means that it was passed to the right device
outputs = speech_generator(
"This is a test", forward_params=forward_params, preprocess_params=preprocess_params
)
self.assertEqual(
{"audio": ANY(np.ndarray), "sampling_rate": 24000},
outputs,
)
@slow
@require_torch
def test_vits_model_pt(self):
speech_generator = pipeline(task="text-to-audio", model="facebook/mms-tts-eng", framework="pt")
outputs = speech_generator("This is a test")
self.assertEqual(outputs["sampling_rate"], 16000)
audio = outputs["audio"]
self.assertEqual(ANY(np.ndarray), audio)
# test two examples side-by-side
outputs = speech_generator(["This is a test", "This is a second test"])
audio = [output["audio"] for output in outputs]
self.assertEqual([ANY(np.ndarray), ANY(np.ndarray)], audio)
# test batching
outputs = speech_generator(["This is a test", "This is a second test"], batch_size=2)
self.assertEqual(ANY(np.ndarray), outputs[0]["audio"])
@slow
@require_torch
def test_forward_model_kwargs(self):
# use vits - a forward model
speech_generator = pipeline(task="text-to-audio", model="kakao-enterprise/vits-vctk", framework="pt")
# for reproducibility
set_seed(555)
outputs = speech_generator("This is a test", forward_params={"speaker_id": 5})
audio = outputs["audio"]
with self.assertRaises(TypeError):
# assert error if generate parameter
outputs = speech_generator("This is a test", forward_params={"speaker_id": 5, "do_sample": True})
forward_params = {"speaker_id": 5}
generate_kwargs = {"do_sample": True}
with self.assertRaises(ValueError):
# assert error if generate_kwargs with forward-only models
outputs = speech_generator(
"This is a test", forward_params=forward_params, generate_kwargs=generate_kwargs
)
self.assertTrue(np.abs(outputs["audio"] - audio).max() < 1e-5)
@slow
@require_torch
def test_generative_model_kwargs(self):
# use musicgen - a generative model
music_generator = pipeline(task="text-to-audio", model="facebook/musicgen-small", framework="pt")
forward_params = {
"do_sample": True,
"max_new_tokens": 250,
}
# for reproducibility
set_seed(555)
outputs = music_generator("This is a test", forward_params=forward_params)
audio = outputs["audio"]
self.assertEqual(ANY(np.ndarray), audio)
# make sure generate kwargs get priority over forward params
forward_params = {
"do_sample": False,
"max_new_tokens": 250,
}
generate_kwargs = {"do_sample": True}
# for reproducibility
set_seed(555)
outputs = music_generator("This is a test", forward_params=forward_params, generate_kwargs=generate_kwargs)
self.assertListEqual(outputs["audio"].tolist(), audio.tolist())
def get_test_pipeline(
self,
model,
tokenizer=None,
image_processor=None,
feature_extractor=None,
processor=None,
torch_dtype="float32",
):
speech_generator = TextToAudioPipeline(
model=model,
tokenizer=tokenizer,
feature_extractor=feature_extractor,
image_processor=image_processor,
processor=processor,
torch_dtype=torch_dtype,
)
return speech_generator, ["This is a test", "Another test"]
def run_pipeline_test(self, speech_generator, _):
outputs = speech_generator("This is a test")
self.assertEqual(ANY(np.ndarray), outputs["audio"])
forward_params = (
{"num_return_sequences": 2, "do_sample": True} if speech_generator.model.can_generate() else {}
)
outputs = speech_generator(["This is great !", "Something else"], forward_params=forward_params)
audio = [output["audio"] for output in outputs]
self.assertEqual([ANY(np.ndarray), ANY(np.ndarray)], audio)
|