You can not select more than 25 topics
Topics must start with a letter or number, can include dashes ('-') and can be up to 35 characters long.
142 lines
6.2 KiB
142 lines
6.2 KiB
# Parallel WaveGAN with the LJSpeech-1.1
|
|
This example contains code used to train a [parallel wavegan](http://arxiv.org/abs/1910.11480) model with [LJSpeech-1.1](https://keithito.com/LJ-Speech-Dataset/).
|
|
## Dataset
|
|
### Download and Extract
|
|
Download LJSpeech-1.1 from the [official website](https://keithito.com/LJ-Speech-Dataset/).
|
|
### Get MFA Result and Extract
|
|
We use [MFA](https://github.com/MontrealCorpusTools/Montreal-Forced-Aligner) results to cut the silence in the edge of audio.
|
|
You can download from here [ljspeech_alignment.tar.gz](https://paddlespeech.bj.bcebos.com/MFA/LJSpeech-1.1/ljspeech_alignment.tar.gz), or train your MFA model reference to [mfa example](https://github.com/PaddlePaddle/PaddleSpeech/tree/develop/examples/other/mfa) of our repo.
|
|
|
|
## Get Started
|
|
Assume the path to the dataset is `~/datasets/LJSpeech-1.1`.
|
|
Assume the path to the MFA result of LJSpeech-1.1 is `./ljspeech_alignment`.
|
|
Run the command below to
|
|
1. **source path**.
|
|
2. preprocess the dataset.
|
|
3. train the model.
|
|
4. synthesize wavs.
|
|
- synthesize waveform from `metadata.jsonl`.
|
|
```bash
|
|
./run.sh
|
|
```
|
|
You can choose a range of stages you want to run, or set `stage` equal to `stop-stage` to use only one stage, for example, running the following command will only preprocess the dataset.
|
|
```bash
|
|
./run.sh --stage 0 --stop-stage 0
|
|
```
|
|
### Data Preprocessing
|
|
```bash
|
|
./local/preprocess.sh ${conf_path}
|
|
```
|
|
When it is done. A `dump` folder is created in the current directory. The structure of the dump folder is listed below.
|
|
|
|
```text
|
|
dump
|
|
├── dev
|
|
│ ├── norm
|
|
│ └── raw
|
|
├── test
|
|
│ ├── norm
|
|
│ └── raw
|
|
└── train
|
|
├── norm
|
|
├── raw
|
|
└── feats_stats.npy
|
|
```
|
|
|
|
The dataset is split into 3 parts, namely `train`, `dev`, and `test`, each of which contains a `norm` and `raw` subfolder. The `raw` folder contains the log magnitude of the mel spectrogram of each utterance, while the norm folder contains the normalized spectrogram. The statistics used to normalize the spectrogram are computed from the training set, which is located in `dump/train/feats_stats.npy`.
|
|
|
|
Also, there is a `metadata.jsonl` in each subfolder. It is a table-like file that contains id and paths to the spectrogram of each utterance.
|
|
|
|
### Model Training
|
|
`./local/train.sh` calls `${BIN_DIR}/train.py`.
|
|
```bash
|
|
CUDA_VISIBLE_DEVICES=${gpus} ./local/train.sh ${conf_path} ${train_output_path}
|
|
```
|
|
Here's the complete help message.
|
|
|
|
```text
|
|
usage: train.py [-h] [--config CONFIG] [--train-metadata TRAIN_METADATA]
|
|
[--dev-metadata DEV_METADATA] [--output-dir OUTPUT_DIR]
|
|
[--ngpu NGPU] [--batch-size BATCH_SIZE] [--max-iter MAX_ITER]
|
|
[--run-benchmark RUN_BENCHMARK]
|
|
[--profiler_options PROFILER_OPTIONS]
|
|
|
|
Train a ParallelWaveGAN model.
|
|
|
|
optional arguments:
|
|
-h, --help show this help message and exit
|
|
--config CONFIG config file to overwrite default config.
|
|
--train-metadata TRAIN_METADATA
|
|
training data.
|
|
--dev-metadata DEV_METADATA
|
|
dev data.
|
|
--output-dir OUTPUT_DIR
|
|
output dir.
|
|
--ngpu NGPU if ngpu == 0, use cpu.
|
|
|
|
benchmark:
|
|
arguments related to benchmark.
|
|
|
|
--batch-size BATCH_SIZE
|
|
batch size.
|
|
--max-iter MAX_ITER train max steps.
|
|
--run-benchmark RUN_BENCHMARK
|
|
runing benchmark or not, if True, use the --batch-size
|
|
and --max-iter.
|
|
--profiler_options PROFILER_OPTIONS
|
|
The option of profiler, which should be in format
|
|
"key1=value1;key2=value2;key3=value3".
|
|
```
|
|
|
|
1. `--config` is a config file in yaml format to overwrite the default config, which can be found at `conf/default.yaml`.
|
|
2. `--train-metadata` and `--dev-metadata` should be the metadata file in the normalized subfolder of `train` and `dev` in the `dump` folder.
|
|
3. `--output-dir` is the directory to save the results of the experiment. Checkpoints are saved in `checkpoints/` inside this directory.
|
|
4. `--ngpu` is the number of gpus to use, if ngpu == 0, use cpu.
|
|
|
|
### Synthesizing
|
|
`./local/synthesize.sh` calls `${BIN_DIR}/../synthesize.py`, which can synthesize waveform from `metadata.jsonl`.
|
|
```bash
|
|
CUDA_VISIBLE_DEVICES=${gpus} ./local/synthesize.sh ${conf_path} ${train_output_path} ${ckpt_name}
|
|
```
|
|
```text
|
|
usage: synthesize.py [-h] [--generator-type GENERATOR_TYPE] [--config CONFIG]
|
|
[--checkpoint CHECKPOINT] [--test-metadata TEST_METADATA]
|
|
[--output-dir OUTPUT_DIR] [--ngpu NGPU]
|
|
|
|
Synthesize with GANVocoder.
|
|
|
|
optional arguments:
|
|
-h, --help show this help message and exit
|
|
--generator-type GENERATOR_TYPE
|
|
type of GANVocoder, should in {pwgan, mb_melgan,
|
|
style_melgan, } now
|
|
--config CONFIG GANVocoder config file.
|
|
--checkpoint CHECKPOINT
|
|
snapshot to load.
|
|
--test-metadata TEST_METADATA
|
|
dev data.
|
|
--output-dir OUTPUT_DIR
|
|
output dir.
|
|
--ngpu NGPU if ngpu == 0, use cpu.
|
|
```
|
|
|
|
1. `--config` parallel wavegan config file. You should use the same config with which the model is trained.
|
|
2. `--checkpoint` is the checkpoint to load. Pick one of the checkpoints from `checkpoints` inside the training output directory.
|
|
3. `--test-metadata` is the metadata of the test dataset. Use the `metadata.jsonl` in the `dev/norm` subfolder from the processed directory.
|
|
4. `--output-dir` is the directory to save the synthesized audio files.
|
|
5. `--ngpu` is the number of gpus to use, if ngpu == 0, use cpu.
|
|
|
|
## Pretrained Model
|
|
Pretrained models can be downloaded here. [pwg_ljspeech_ckpt_0.5.zip](https://paddlespeech.bj.bcebos.com/Parakeet/released_models/pwgan/pwg_ljspeech_ckpt_0.5.zip)
|
|
|
|
Parallel WaveGAN checkpoint contains files listed below.
|
|
|
|
```text
|
|
pwg_ljspeech_ckpt_0.5
|
|
├── pwg_default.yaml # default config used to train parallel wavegan
|
|
├── pwg_snapshot_iter_400000.pdz # generator parameters of parallel wavegan
|
|
└── pwg_stats.npy # statistics used to normalize spectrogram when training parallel wavegan
|
|
```
|
|
## Acknowledgement
|
|
We adapted some code from https://github.com/kan-bayashi/ParallelWaveGAN.
|