FastSpeech2

FastSpeech2 Recipe

In this recipe, we will show how to train FastSpeech2 using Amphion's infrastructure. FastSpeech2 is a non-autoregressive TTS architecture that utilizes feed-forward Transformer blocks.

There are four stages in total:

1. Data preparation
2. Features extraction
3. Training
4. Inference

NOTE: You need to run every command of this recipe in the Amphion root path:

cd Amphion

1. Data Preparation

Dataset Download

You can use the commonly used TTS dataset to train TTS model, e.g., LJSpeech, VCTK, LibriTTS, etc. We strongly recommend you use LJSpeech to train TTS model for the first time. How to download dataset is detailed here.

Configuration

After downloading the dataset, you can set the dataset paths in exp_config.json. Note that you can change the dataset list to use your preferred datasets.

    "dataset": [
        "LJSpeech",
    ],
    "dataset_path": {
        // TODO: Fill in your dataset path
        "LJSpeech": "[LJSpeech dataset path]",
    },

2. Features Extraction

Configuration

Specify the processed_dir and the log_dir and for saving the processed data and the checkpoints in exp_config.json:

    // TODO: Fill in the output log path
    "log_dir": "ckpts/tts",
    "preprocess": {
        // TODO: Fill in the output data path
        "processed_dir": "data",
        ...
    },

Run

Run the run.sh as the preproces stage (set --stage 1):

sh egs/tts/FastSpeech2/run.sh --stage 1

3. Training

Configuration

We provide the default hyparameters in the exp_config.json. They can work on single NVIDIA-24g GPU. You can adjust them based on your GPU machines.

"train": {
        "batch_size": 16,
    }

Run

Run the run.sh as the training stage (set --stage 2). Specify a experimental name to run the following command. The tensorboard logs and checkpoints will be saved in ckpts/tts/[YourExptName].

sh egs/tts/FastSpeech2/run.sh --stage 2 --name [YourExptName]

NOTE: The CUDA_VISIBLE_DEVICES is set as "0" in default. You can change it when running run.sh by specifying such as --gpu "0,1,2,3".

4. Inference

Pre-trained Fastspeech 2 and HiFi-GAN Download

We released a pre-trained Amphion Fastspeech 2 model and HiFi-GAN trained on LJSpeech. So you can download the them and generate speech according to the following inference instruction.

Configuration

For inference, you need to specify the following configurations when running run.sh:

Parameters	Description	Example
--infer_expt_dir	The experimental directory which contains checkpoint	ckpts/tts/[YourExptName]
--infer_output_dir	The output directory to save inferred audios.	ckpts/tts/[YourExptName]/result
--infer_mode	The inference mode, e.g., "single", "batch".	"single" to generate a clip of speech, "batch" to generate a batch of speech at a time.
--infer_dataset	The dataset used for inference.	For LJSpeech dataset, the inference dataset would be LJSpeech.
--infer_testing_set	The subset of the inference dataset used for inference, e.g., train, test, golden_test	For LJSpeech dataset, the testing set would be  "test" split from LJSpeech at the feature extraction, or "golden_test" cherry-picked from test set as template testing set.
--infer_text	The text to be synthesized.	"This is a clip of generated speech with the given text from a TTS model."
--vocoder_dir	The directory for the vocoder.	"ckpts/vocoder/hifigan_ljspeech"

Run

For example, if you want to generate speech of all testing set split from LJSpeech, just run:

sh egs/tts/FastSpeech2/run.sh --stage 3 \
    --infer_expt_dir ckpts/tts/[YourExptName] \
    --infer_output_dir ckpts/tts/[YourExptName]/result \
    --infer_mode "batch" \
    --infer_dataset "LJSpeech" \
    --infer_testing_set "test" \
    --vocoder_dir ckpts/vocoder/hifigan_ljspeech/checkpoints

Or, if you want to generate a single clip of speech from a given text, just run:

sh egs/tts/FastSpeech2/run.sh --stage 3 \
    --infer_expt_dir ckpts/tts/[YourExptName] \
    --infer_output_dir ckpts/tts/[YourExptName]/result \
    --infer_mode "single" \
    --infer_text "This is a clip of generated speech with the given text from a TTS model." \
    --vocoder_dir ckpts/vocoder/hifigan_ljspeech

ISSUES and Solutions

NotImplementedError: Using RTX 3090 or 4000 series doesn't support faster communication broadband via P2P or IB. Please set `NCCL_P2P_DISABLE="1"` and `NCCL_IB_DISABLE="1" or use `accelerate launch` which will do this automatically.
2024-02-24 10:57:49 | INFO | torch.distributed.distributed_c10d | Rank 0: Completed store-based barrier for key:store_based_barrier_key:1 with 2 nodes.

The error message is related to an incompatibility issue with the NVIDIA RTX 3090 or 4000 series GPUs when trying to use peer-to-peer (P2P) communication or InfiniBand (IB) for faster communication. This incompatibility arises within the PyTorch accelerate library, which facilitates distributed training and inference.

To fix this issue, before running your script, you can set the environment variables in your terminal:

export NCCL_P2P_DISABLE=1
export NCCL_IB_DISABLE=1

Noted

Extensive logging messages related to torch._subclasses.fake_tensor and torch._dynamo.output_graph may be observed during inference. Despite attempts to ignore these logs, no effective solution has been found. However, it does not impact the inference process.

@inproceedings{ren2020fastspeech,
  title={FastSpeech 2: Fast and High-Quality End-to-End Text to Speech},
  author={Ren, Yi and Hu, Chenxu and Tan, Xu and Qin, Tao and Zhao, Sheng and Zhao, Zhou and Liu, Tie-Yan},
  booktitle={International Conference on Learning Representations},
  year={2020}
}