目录
  • 文档 >
  • 使用 Tacotron2 进行文本转语音 >
  • 当前（稳定）
快捷方式
tutorials/tacotron2_pipeline_tutorial

注意

点击 这里 下载完整的示例代码

使用 Tacotron2 进行文本转语音

作者: 杨耀元, Moto Hira

概述

本教程展示了如何使用 torchaudio 中的预训练 Tacotron2 构建文本转语音管道。

文本转语音管道如下所示

  1. 文本预处理

    首先，将输入文本编码为符号列表。在本教程中，我们将使用英语字符和音素作为符号。

  2. 频谱图生成

    从编码的文本中，生成一个频谱图。我们使用 Tacotron2 模型来完成此操作。

  3. 时域转换

    最后一步是将频谱图转换为波形。从频谱图生成语音的过程也称为声码器。在本教程中，使用了三种不同的声码器，WaveRNN，GriffinLim 和 Nvidia 的 WaveGlow。

下图说明了整个过程。

https://download.pytorch.org/torchaudio/tutorial-assets/tacotron2_tts_pipeline.png

所有相关组件都捆绑在 torchaudio.pipelines.Tacotron2TTSBundle 中，但本教程也将涵盖幕后的过程。

准备

首先，我们安装必要的依赖项。除了 torchaudio 之外，还需要 DeepPhonemizer 来执行基于音素的编码。

%%bash
pip3 install deep_phonemizer

import torch
import torchaudio

torch.random.manual_seed(0)
device = "cuda" if torch.cuda.is_available() else "cpu"

print(torch.__version__)
print(torchaudio.__version__)
print(device)

2.3.0
2.3.0
cuda

import IPython
import matplotlib.pyplot as plt

文本处理

基于字符的编码

在本节中，我们将介绍基于字符的编码的工作原理。

由于预训练的 Tacotron2 模型期望特定的符号表，因此 torchaudio 中也提供了相同的功能。但是，为了便于理解，我们将首先手动实现编码。

首先，我们定义一组符号 '_-!\'(),.:;? abcdefghijklmnopqrstuvwxyz'。然后，我们将输入文本中的每个字符映射到表中对应符号的索引。不在表中的符号将被忽略。

symbols = "_-!'(),.:;? abcdefghijklmnopqrstuvwxyz"
look_up = {s: i for i, s in enumerate(symbols)}
symbols = set(symbols)


def text_to_sequence(text):
    text = text.lower()
    return [look_up[s] for s in text if s in symbols]


text = "Hello world! Text to speech!"
print(text_to_sequence(text))

[19, 16, 23, 23, 26, 11, 34, 26, 29, 23, 15, 2, 11, 31, 16, 35, 31, 11, 31, 26, 11, 30, 27, 16, 16, 14, 19, 2]

如上所述，符号表和索引必须与预训练的 Tacotron2 模型期望的一致。 torchaudio 提供了与预训练模型相同的转换。您可以按如下方式实例化和使用此类转换。

processor = torchaudio.pipelines.TACOTRON2_WAVERNN_CHAR_LJSPEECH.get_text_processor()

text = "Hello world! Text to speech!"
processed, lengths = processor(text)

print(processed)
print(lengths)

tensor([[19, 16, 23, 23, 26, 11, 34, 26, 29, 23, 15,  2, 11, 31, 16, 35, 31, 11,
         31, 26, 11, 30, 27, 16, 16, 14, 19,  2]])
tensor([28], dtype=torch.int32)

注意：我们的手动编码的输出和 torchaudio text_processor 的输出匹配（意味着我们正确地重新实现了库在内部执行的操作）。它接受文本或文本列表作为输入。当提供文本列表时，返回的 lengths 变量表示输出批次中每个处理过的标记的有效长度。

中间表示可以按如下方式检索

print([processor.tokens[i] for i in processed[0, : lengths[0]]])

['h', 'e', 'l', 'l', 'o', ' ', 'w', 'o', 'r', 'l', 'd', '!', ' ', 't', 'e', 'x', 't', ' ', 't', 'o', ' ', 's', 'p', 'e', 'e', 'c', 'h', '!']

基于音素的编码

基于音素的编码类似于基于字符的编码，但它使用基于音素的符号表和 G2P（音素-音素）模型。

G2P 模型的详细信息超出了本教程的范围，我们只看看转换的样子。

与基于字符的编码类似，编码过程应与预训练的 Tacotron2 模型的训练一致。 torchaudio 提供了一个接口来创建该过程。

以下代码演示了如何创建和使用该过程。在幕后，使用 DeepPhonemizer 包创建了一个 G2P 模型，并获取了 DeepPhonemizer 作者发布的预训练权重。

bundle = torchaudio.pipelines.TACOTRON2_WAVERNN_PHONE_LJSPEECH

processor = bundle.get_text_processor()

text = "Hello world! Text to speech!"
with torch.inference_mode():
    processed, lengths = processor(text)

print(processed)
print(lengths)

  0%|          | 0.00/63.6M [00:00<?, ?B/s]
  0%|          | 128k/63.6M [00:00<01:32, 722kB/s]
  1%|          | 384k/63.6M [00:00<00:57, 1.15MB/s]
  2%|1         | 1.25M/63.6M [00:00<00:22, 2.93MB/s]
  7%|7         | 4.75M/63.6M [00:00<00:05, 10.9MB/s]
 12%|#2        | 7.75M/63.6M [00:00<00:03, 14.7MB/s]
 21%|##        | 13.2M/63.6M [00:00<00:02, 23.5MB/s]
 26%|##6       | 16.6M/63.6M [00:01<00:02, 24.2MB/s]
 34%|###4      | 21.8M/63.6M [00:01<00:01, 31.4MB/s]
 39%|###9      | 25.1M/63.6M [00:01<00:01, 30.4MB/s]
 45%|####4     | 28.4M/63.6M [00:01<00:01, 28.6MB/s]
 50%|####9     | 31.8M/63.6M [00:01<00:01, 30.2MB/s]
 58%|#####7    | 36.6M/63.6M [00:01<00:00, 33.6MB/s]
 63%|######3   | 40.1M/63.6M [00:01<00:00, 33.5MB/s]
 68%|######8   | 43.5M/63.6M [00:01<00:00, 31.3MB/s]
 76%|#######5  | 48.2M/63.6M [00:02<00:00, 34.3MB/s]
 82%|########1 | 52.0M/63.6M [00:02<00:00, 32.4MB/s]
 87%|########6 | 55.2M/63.6M [00:02<00:00, 32.2MB/s]
 94%|#########4| 60.1M/63.6M [00:02<00:00, 35.4MB/s]
100%|##########| 63.6M/63.6M [00:02<00:00, 26.7MB/s]
/pytorch/audio/ci_env/lib/python3.10/site-packages/torch/nn/modules/transformer.py:306: UserWarning: enable_nested_tensor is True, but self.use_nested_tensor is False because encoder_layer.self_attn.batch_first was not True(use batch_first for better inference performance)
  warnings.warn(f"enable_nested_tensor is True, but self.use_nested_tensor is False because {why_not_sparsity_fast_path}")
tensor([[54, 20, 65, 69, 11, 92, 44, 65, 38,  2, 11, 81, 40, 64, 79, 81, 11, 81,
         20, 11, 79, 77, 59, 37,  2]])
tensor([25], dtype=torch.int32)

请注意，编码后的值与基于字符的编码示例不同。

中间表示如下所示。

print([processor.tokens[i] for i in processed[0, : lengths[0]]])

['HH', 'AH', 'L', 'OW', ' ', 'W', 'ER', 'L', 'D', '!', ' ', 'T', 'EH', 'K', 'S', 'T', ' ', 'T', 'AH', ' ', 'S', 'P', 'IY', 'CH', '!']

频谱图生成

Tacotron2 是我们用来从编码文本生成频谱图的模型。有关模型的详细信息，请参阅 论文。

使用预训练权重实例化 Tacotron2 模型很容易，但是请注意，Tacotron2 模型的输入需要由匹配的文本处理器进行处理。

torchaudio.pipelines.Tacotron2TTSBundle 将匹配的模型和处理器捆绑在一起，以便于创建管道。

有关可用捆绑包及其用法，请参阅 Tacotron2TTSBundle。

bundle = torchaudio.pipelines.TACOTRON2_WAVERNN_PHONE_LJSPEECH
processor = bundle.get_text_processor()
tacotron2 = bundle.get_tacotron2().to(device)

text = "Hello world! Text to speech!"

with torch.inference_mode():
    processed, lengths = processor(text)
    processed = processed.to(device)
    lengths = lengths.to(device)
    spec, _, _ = tacotron2.infer(processed, lengths)


_ = plt.imshow(spec[0].cpu().detach(), origin="lower", aspect="auto")
tacotron2 pipeline tutorial
/pytorch/audio/ci_env/lib/python3.10/site-packages/torch/nn/modules/transformer.py:306: UserWarning: enable_nested_tensor is True, but self.use_nested_tensor is False because encoder_layer.self_attn.batch_first was not True(use batch_first for better inference performance)
  warnings.warn(f"enable_nested_tensor is True, but self.use_nested_tensor is False because {why_not_sparsity_fast_path}")
Downloading: "https://download.pytorch.org/torchaudio/models/tacotron2_english_phonemes_1500_epochs_wavernn_ljspeech.pth" to /root/.cache/torch/hub/checkpoints/tacotron2_english_phonemes_1500_epochs_wavernn_ljspeech.pth

  0%|          | 0.00/107M [00:00<?, ?B/s]
 12%|#2        | 13.4M/107M [00:00<00:00, 140MB/s]
 25%|##4       | 26.8M/107M [00:00<00:01, 57.4MB/s]
 32%|###2      | 34.6M/107M [00:00<00:01, 60.0MB/s]
 45%|####4     | 48.0M/107M [00:00<00:00, 65.2MB/s]
 60%|#####9    | 64.0M/107M [00:00<00:00, 71.0MB/s]
 74%|#######4  | 80.0M/107M [00:01<00:00, 75.0MB/s]
 89%|########9 | 95.8M/107M [00:01<00:00, 86.4MB/s]
 97%|#########7| 105M/107M [00:01<00:00, 66.7MB/s]
100%|##########| 107M/107M [00:01<00:00, 67.7MB/s]

请注意，Tacotron2.infer 方法执行多项式采样，因此，生成频谱图的过程会产生随机性。

def plot():
    fig, ax = plt.subplots(3, 1)
    for i in range(3):
        with torch.inference_mode():
            spec, spec_lengths, _ = tacotron2.infer(processed, lengths)
        print(spec[0].shape)
        ax[i].imshow(spec[0].cpu().detach(), origin="lower", aspect="auto")


plot()
tacotron2 pipeline tutorial
torch.Size([80, 190])
torch.Size([80, 184])
torch.Size([80, 185])

波形生成

生成频谱图后，最后一个过程是使用声码器从频谱图中恢复波形。

torchaudio 提供基于 GriffinLim 和 WaveRNN 的声码器。

WaveRNN 声码器

从上一节继续，我们可以从同一个捆绑包中实例化匹配的 WaveRNN 模型。

bundle = torchaudio.pipelines.TACOTRON2_WAVERNN_PHONE_LJSPEECH

processor = bundle.get_text_processor()
tacotron2 = bundle.get_tacotron2().to(device)
vocoder = bundle.get_vocoder().to(device)

text = "Hello world! Text to speech!"

with torch.inference_mode():
    processed, lengths = processor(text)
    processed = processed.to(device)
    lengths = lengths.to(device)
    spec, spec_lengths, _ = tacotron2.infer(processed, lengths)
    waveforms, lengths = vocoder(spec, spec_lengths)

/pytorch/audio/ci_env/lib/python3.10/site-packages/torch/nn/modules/transformer.py:306: UserWarning: enable_nested_tensor is True, but self.use_nested_tensor is False because encoder_layer.self_attn.batch_first was not True(use batch_first for better inference performance)
  warnings.warn(f"enable_nested_tensor is True, but self.use_nested_tensor is False because {why_not_sparsity_fast_path}")
Downloading: "https://download.pytorch.org/torchaudio/models/wavernn_10k_epochs_8bits_ljspeech.pth" to /root/.cache/torch/hub/checkpoints/wavernn_10k_epochs_8bits_ljspeech.pth

  0%|          | 0.00/16.7M [00:00<?, ?B/s]
 89%|########9 | 14.9M/16.7M [00:00<00:00, 52.6MB/s]
100%|##########| 16.7M/16.7M [00:00<00:00, 45.5MB/s]

def plot(waveforms, spec, sample_rate):
    waveforms = waveforms.cpu().detach()

    fig, [ax1, ax2] = plt.subplots(2, 1)
    ax1.plot(waveforms[0])
    ax1.set_xlim(0, waveforms.size(-1))
    ax1.grid(True)
    ax2.imshow(spec[0].cpu().detach(), origin="lower", aspect="auto")
    return IPython.display.Audio(waveforms[0:1], rate=sample_rate)


plot(waveforms, spec, vocoder.sample_rate)
tacotron2 pipeline tutorial


Griffin-Lim 声码器

使用 Griffin-Lim 声码器与 WaveRNN 相同。您可以使用 get_vocoder() 方法实例化声码器对象，并传递频谱图。

bundle = torchaudio.pipelines.TACOTRON2_GRIFFINLIM_PHONE_LJSPEECH

processor = bundle.get_text_processor()
tacotron2 = bundle.get_tacotron2().to(device)
vocoder = bundle.get_vocoder().to(device)

with torch.inference_mode():
    processed, lengths = processor(text)
    processed = processed.to(device)
    lengths = lengths.to(device)
    spec, spec_lengths, _ = tacotron2.infer(processed, lengths)
waveforms, lengths = vocoder(spec, spec_lengths)

/pytorch/audio/ci_env/lib/python3.10/site-packages/torch/nn/modules/transformer.py:306: UserWarning: enable_nested_tensor is True, but self.use_nested_tensor is False because encoder_layer.self_attn.batch_first was not True(use batch_first for better inference performance)
  warnings.warn(f"enable_nested_tensor is True, but self.use_nested_tensor is False because {why_not_sparsity_fast_path}")
Downloading: "https://download.pytorch.org/torchaudio/models/tacotron2_english_phonemes_1500_epochs_ljspeech.pth" to /root/.cache/torch/hub/checkpoints/tacotron2_english_phonemes_1500_epochs_ljspeech.pth

  0%|          | 0.00/107M [00:00<?, ?B/s]
  0%|          | 256k/107M [00:00<00:42, 2.62MB/s]
 15%|#4        | 16.0M/107M [00:00<00:01, 77.4MB/s]
 25%|##5       | 27.0M/107M [00:00<00:00, 92.4MB/s]
 33%|###3      | 35.6M/107M [00:00<00:00, 81.2MB/s]
 44%|####3     | 46.9M/107M [00:00<00:00, 69.1MB/s]
 50%|#####     | 53.8M/107M [00:01<00:01, 44.2MB/s]
 60%|#####9    | 64.0M/107M [00:01<00:00, 46.8MB/s]
 74%|#######4  | 79.6M/107M [00:01<00:00, 67.8MB/s]
 82%|########1 | 88.1M/107M [00:01<00:00, 64.3MB/s]
 89%|########9 | 96.0M/107M [00:01<00:00, 66.0MB/s]
 99%|#########9| 107M/107M [00:01<00:00, 56.8MB/s]
100%|##########| 107M/107M [00:01<00:00, 60.6MB/s]

plot(waveforms, spec, vocoder.sample_rate)
tacotron2 pipeline tutorial


Waveglow 声码器

Waveglow 是由英伟达发布的一种声码器。预训练权重已发布在 Torch Hub 上。可以使用 torch.hub 模块实例化模型。

# Workaround to load model mapped on GPU
# https://stackoverflow.com/a/61840832
waveglow = torch.hub.load(
    "NVIDIA/DeepLearningExamples:torchhub",
    "nvidia_waveglow",
    model_math="fp32",
    pretrained=False,
)
checkpoint = torch.hub.load_state_dict_from_url(
    "https://api.ngc.nvidia.com/v2/models/nvidia/waveglowpyt_fp32/versions/1/files/nvidia_waveglowpyt_fp32_20190306.pth",  # noqa: E501
    progress=False,
    map_location=device,
)
state_dict = {key.replace("module.", ""): value for key, value in checkpoint["state_dict"].items()}

waveglow.load_state_dict(state_dict)
waveglow = waveglow.remove_weightnorm(waveglow)
waveglow = waveglow.to(device)
waveglow.eval()

with torch.no_grad():
    waveforms = waveglow.infer(spec)

/pytorch/audio/ci_env/lib/python3.10/site-packages/torch/hub.py:293: UserWarning: You are about to download and run code from an untrusted repository. In a future release, this won't be allowed. To add the repository to your trusted list, change the command to {calling_fn}(..., trust_repo=False) and a command prompt will appear asking for an explicit confirmation of trust, or load(..., trust_repo=True), which will assume that the prompt is to be answered with 'yes'. You can also use load(..., trust_repo='check') which will only prompt for confirmation if the repo is not already trusted. This will eventually be the default behaviour
  warnings.warn(
Downloading: "https://github.com/NVIDIA/DeepLearningExamples/zipball/torchhub" to /root/.cache/torch/hub/torchhub.zip
/root/.cache/torch/hub/NVIDIA_DeepLearningExamples_torchhub/PyTorch/Classification/ConvNets/image_classification/models/common.py:13: UserWarning: pytorch_quantization module not found, quantization will not be available
  warnings.warn(
/root/.cache/torch/hub/NVIDIA_DeepLearningExamples_torchhub/PyTorch/Classification/ConvNets/image_classification/models/efficientnet.py:17: UserWarning: pytorch_quantization module not found, quantization will not be available
  warnings.warn(
/pytorch/audio/ci_env/lib/python3.10/site-packages/torch/nn/utils/weight_norm.py:28: UserWarning: torch.nn.utils.weight_norm is deprecated in favor of torch.nn.utils.parametrizations.weight_norm.
  warnings.warn("torch.nn.utils.weight_norm is deprecated in favor of torch.nn.utils.parametrizations.weight_norm.")
Downloading: "https://api.ngc.nvidia.com/v2/models/nvidia/waveglowpyt_fp32/versions/1/files/nvidia_waveglowpyt_fp32_20190306.pth" to /root/.cache/torch/hub/checkpoints/nvidia_waveglowpyt_fp32_20190306.pth

plot(waveforms, spec, 22050)
tacotron2 pipeline tutorial


脚本总运行时间：（1 分钟 44.859 秒）

由 Sphinx-Gallery 生成的图库

文档

访问 PyTorch 的全面开发者文档

查看文档

教程

获取针对初学者和高级开发人员的深入教程

查看教程

资源

查找开发资源并获得问题的解答

查看资源

为了分析流量并优化您的体验，我们在本网站上提供 cookie。点击或导航，即表示您同意我们使用 cookie。作为本网站的当前维护者，Facebook 的 Cookie 政策适用。了解更多信息，包括有关可用控制的信息：Cookie 政策。