ge2e

Speaker Encoder

This experiment trains a speaker encoder with speaker verification as to its task. It is done as a part of the experiment of transfer learning from speaker verification to multispeaker text-to-speech synthesis, which can be found at examples/aishell3/vc0. The trained speaker encoder is used to extract utterance embeddings from utterances.

Model

The model used in this experiment is the speaker encoder with text-independent speaker verification task in GENERALIZED END-TO-END LOSS FOR SPEAKER VERIFICATION. GE2E-softmax loss is used.

Download Datasets

Currently supported datasets are Librispeech-other-500, VoxCeleb, VoxCeleb2,ai-datatang-200zh, magicdata, which can be downloaded from the corresponding webpage.

1. Librispeech/train-other-500 An English multispeaker dataset，URL，only the train-other-500 subset is used.
2. VoxCeleb1 An English multispeaker dataset，URL, Audio Files from Dev A to Dev D should be downloaded, combined, and extracted.
3. VoxCeleb2 An English multispeaker dataset，URL, Audio Files from Dev A to Dev H should be downloaded, combined, and extracted.
4. Aidatatang-200zh A Mandarin Chinese multispeaker dataset ，URL.
5. magicdata A Mandarin Chinese multispeaker dataset ，URL.

If you want to use other datasets, you can also download and preprocess them as long as they meet the requirements described below.

Get Started

./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.

./run.sh --stage 0 --stop-stage 0

Data Preprocessing

./local/preprocess.sh calls ${BIN_DIR}/preprocess.py.

./local/preprocess.sh ${datasets_root} ${preprocess_path} ${dataset_names}

Assume datasets_root is ~/datasets/GE2E, and it has the following structure（We only use train-other-500 for simplicity）:

GE2E
├── LibriSpeech
└── (other datasets)

Multispeaker datasets are used as training data, though the transcriptions are not used. To enlarge the amount of data used for training, several multispeaker datasets are combined. The preprocessed datasets are organized in a file structure described below. The mel spectrogram of each utterance is saved in .npy format. The dataset is 2-stratified (speaker-utterance). Since multiple datasets are combined, to avoid conflict in speaker id, the dataset name is prepended to the speaker ids.

dataset_root
├── dataset01_speaker01/
│   ├── utterance01.npy
│   ├── utterance02.npy
│   └── utterance03.npy
├── dataset01_speaker02/
│   ├── utterance01.npy
│   ├── utterance02.npy
│   └── utterance03.npy
├── dataset02_speaker01/
│   ├── utterance01.npy
│   ├── utterance02.npy
│   └── utterance03.npy
└── dataset02_speaker02/
    ├── utterance01.npy
    ├── utterance02.npy
    └── utterance03.npy

In ${BIN_DIR}/preprocess.py:

1. --datasets_root is the directory that contains several extracted dataset
2. --output_dir is the directory to save the preprocessed dataset
3. --dataset_names is the dataset to preprocess. If there are multiple datasets in --datasets_root to preprocess, the names can be joined with a comma. Currently supported dataset names are librispeech_other, voxceleb1, voxceleb2, aidatatang_200zh, and magicdata.

Model Training

./local/train.sh calls ${BIN_DIR}/train.py.

CUDA_VISIBLE_DEVICES=${gpus} ./local/train.sh ${preprocess_path} ${train_output_path}

In ${BIN_DIR}/train.py:

1. --data is the path to the preprocessed dataset.
2. --output is the directory to save results，usually a subdirectory of runs. It contains visualdl log files, text log files, config files, and a checkpoints directory, which contains parameter files and optimizer state files. If --output already has some training results in it, the most recent parameter file and optimizer state file are loaded before training.
3. --ngpu is the number of gpus to use, if ngpu == 0, use cpu.
4. CUDA_VISIBLE_DEVICES can be used to specify visible devices with cuda.

Other options are described below.

• --config is a .yaml config file used to override the default config(which is coded in config.py).
• --opts is a command-line option to further override config files. It should be the last command-line options passed with multiple key-value pairs separated by spaces.
• --checkpoint_path specifies the checkpoint to load before training, extension is not included. A parameter file ( .pdparams) and an optimizer state file ( .pdopt) with the same name is used. This option has a higher priority than auto-resuming from the --output directory.

Inferencing

When training is done, run the command below to generate utterance embedding for each utterance in a dataset. ./local/inference.sh calls ${BIN_DIR}/inference.py.

CUDA_VISIBLE_DEVICES=${gpus} ./local/inference.sh ${infer_input} ${infer_output} ${train_output_path} ${ckpt_name}

In ${BIN_DIR}/inference.py:

1. --input is the path of the dataset used for inference.
2. --output is the directory to save the processed results. It has the same file structure as the input dataset. Each utterance in the dataset has a corresponding utterance embedding file in the *.npy format.
3. --checkpoint_path is the path of the checkpoint to use, extension not included.
4. --pattern is the wildcard pattern to filter audio files for inference, defaults to *.wav.
5. --ngpu is the number of gpus to use, if ngpu == 0, use cpu.

Pretrained Model

The pretrained model is first trained to 1560k steps at Librispeech-other-500 and voxceleb1. Then trained at aidatatang_200h and magic_data to 3000k steps.

Download URL ge2e_ckpt_0.3.zip.

References

1. Generalized End-to-end Loss for Speaker Verification
2. Transfer Learning from Speaker Verification to Multispeaker Text-To-Speech Synthesis