Update xtts training code

TTS/tts/datasets/formatters.py

@@ -657,10 +657,13 @@ def bel_tts_formatter(root_path, meta_file, **kwargs):  # pylint: disable=unused
 def afrotts(root_path, meta_file, **kwargs):
     csv_path = os.path.join(root_path, meta_file)
     csv_file = pd.read_csv(csv_path)
+    csv_file["char_count"] = csv_file.transcript.apply(lambda x: len(list(x)))
+    csv_file = csv_file[csv_file.char_count < 400].copy()
     csv_file["audio_paths"] = csv_file["audio_paths"].apply(
         lambda x: x.replace("/AfriSpeech-TTS-D/", root_path)
     )
     csv_file = csv_file.rename(columns={"transcript":"text", "audio_paths":"audio_file", "user_ids":"speaker_name"})
+    csv_file = csv_file[["audio_file", "text", "speaker_name"]]
     csv_file['root_path'] = root_path
     items = csv_file.to_dict('records')
     return items

recipes/ljspeech/xtts_v2/README.md

@@ -16,15 +16,21 @@ This repository contains the code for training a text-to-speech (TTS) model usin
    ```bash
    cd coqui-ai-TTS
    ```
+3. **Create a virtual environment with python version 3.10**
 
-3. **Install system dependencies and the code:**
+   ```bash
+   conda create -n xtts python==3.10
+   conda activate xtts
+   ```
+
+4. **Install system dependencies and the code:**
 
    ```bash
    make system-deps  # Intended to be used on Ubuntu (Debian). Let us know if you have a different OS.
    make install
    ```
 
-4. **Open the following file and redefine the specified variables:**
+5. **Open the following file and redefine the specified variables:**
 
    File: `recipes/ljspeech/xtts_v2/train_gpt_xtts.py`
 
@@ -50,4 +56,12 @@ From the repository root, run the following command:
 python3 recipes/ljspeech/xtts_v2/train_gpt_xtts.py
 ```
 
-You are now ready to train your TTS model using Coqui AI's framework. Enjoy!
+You are now ready to train your TTS model using Coqui AI's framework. Enjoy!
+
+## Optional: Resuming from a checkpoint
+
+File: `recipes/ljspeech/xtts_v2/train_gpt_xtts_resume.py`
+
+Update the parameters in the file for the models
+
+

recipes/ljspeech/xtts_v2/train_gpt_xtts.py

@@ -1,4 +1,6 @@
 import os
+os.environ["CUDA_DEVICE_ORDER"] = "PCI_BUS_ID"
+os.environ["CUDA_VISIBLE_DEVICES"] = "0"
 
 from trainer import Trainer, TrainerArgs
 

recipes/ljspeech/xtts_v2/train_gpt_xtts_resume_ft.py

@@ -0,0 +1,171 @@
+import os
+os.environ["CUDA_DEVICE_ORDER"] = "PCI_BUS_ID"
+os.environ["CUDA_VISIBLE_DEVICES"] = "0"
+
+from trainer import Trainer, TrainerArgs
+
+from TTS.config.shared_configs import BaseDatasetConfig
+from TTS.tts.datasets import load_tts_samples
+from TTS.tts.layers.xtts.trainer.gpt_trainer import GPTArgs, GPTTrainer, GPTTrainerConfig, XttsAudioConfig
+from TTS.utils.manage import ModelManager
+
+# Logging parameters
+RUN_NAME = "GPT_XTTS_v2.0_AfroTTS_FT"
+PROJECT_NAME = "XTTS_trainer"
+DASHBOARD_LOGGER = "tensorboard"
+LOGGER_URI = None
+
+# Set here the path that the checkpoints will be saved. Default: ./run/training/
+OUT_PATH = os.path.join(os.path.dirname(os.path.abspath(__file__)), "run", "training")
+
+# Training Parameters
+OPTIMIZER_WD_ONLY_ON_WEIGHTS = True  # for multi-gpu training please make it False
+START_WITH_EVAL = False  # if True it will star with evaluation
+BATCH_SIZE = 2  # set here the batch size
+GRAD_ACUMM_STEPS = 126  # set here the grad accumulation steps
+# Note: we recommend that BATCH_SIZE * GRAD_ACUMM_STEPS need to be at least 252 for more efficient training. You can increase/decrease BATCH_SIZE but then set GRAD_ACUMM_STEPS accordingly.
+
+afrotts_dir = "AfriSpeech-TTS-D" # add path to afrotts data here
+
+# Define here the dataset that you want to use for the fine-tuning on.
+config_dataset = BaseDatasetConfig(
+    formatter="afrotts",
+    dataset_name="afrotts",
+    path=afrotts_dir,
+    meta_file_train=os.path.join(afrotts_dir, "data/afritts-train-clean-upsamp.csv") #afritts-train-clean-upsamp.csv
+    meta_file_val=os.path.join(afrotts_dir,"data/afritts-dev-clean.csv"),
+    language="en",
+)
+
+# Add here the configs of the datasets
+DATASETS_CONFIG_LIST = [config_dataset]
+
+# Define the path where XTTS v2.0.1 files will be downloaded
+CHECKPOINTS_OUT_PATH = os.path.join(OUT_PATH, "XTTS_v2.0_original_model_files/")
+os.makedirs(CHECKPOINTS_OUT_PATH, exist_ok=True)
+
+# Set the path to the downloaded files
+DVAE_CHECKPOINT = "coqui-ai-TTS/recipes/ljspeech/xtts_v2/run/training/XTTS_v2.0_original_model_files/dvae.pth"
+MEL_NORM_FILE = "coqui-ai-TTS/recipes/ljspeech/xtts_v2/run/training/XTTS_v2.0_original_model_files/mel_stats.pth"
+
+# download DVAE files if needed
+if not os.path.isfile(DVAE_CHECKPOINT) or not os.path.isfile(MEL_NORM_FILE):
+    print(" > Downloading DVAE files!")
+    ModelManager._download_model_files([MEL_NORM_LINK, DVAE_CHECKPOINT_LINK], CHECKPOINTS_OUT_PATH, progress_bar=True)
+
+
+# XTTS transfer learning parameters: You we need to provide the paths of XTTS model checkpoint that you want to do the fine tuning.
+TOKENIZER_FILE = "coqui-ai-TTS/recipes/ljspeech/xtts_v2/run/training/XTTS_v2.0_original_model_files/vocab.json"
+XTTS_CHECKPOINT = "coqui-ai-TTS/recipes/ljspeech/xtts_v2/run/training/GPT_XTTS_v2.0_AfroTTS_FT-March-06-2024_06+36AM-581cf506/checkpoint_135000.pth"
+
+
+# download XTTS v2.0 files if needed
+if not os.path.isfile(TOKENIZER_FILE) or not os.path.isfile(XTTS_CHECKPOINT):
+    print(" > Downloading XTTS v2.0 files!")
+    ModelManager._download_model_files(
+        [TOKENIZER_FILE_LINK, XTTS_CHECKPOINT_LINK], CHECKPOINTS_OUT_PATH, progress_bar=True
+    )
+
+
+# Training sentences generations
+SPEAKER_REFERENCE = [
+    "/AfriSpeech-TTS/train/defc5e03-926c-4e0b-a639-c821e5e7db89/14f64f13c57f9a64a2a1521253934a0b_KYA8MaKS.wav"  # speaker reference to be used in training test sentences
+]
+LANGUAGE = config_dataset.language
+
+def main():
+    # init args and config
+    model_args = GPTArgs(
+        max_conditioning_length=132300,  # 6 secs
+        min_conditioning_length=66150,  # 3 secs
+        debug_loading_failures=False,
+        max_wav_length=255995,  # ~11.6 seconds  661500, #~ 30 seconds #
+        max_text_length=300,
+        mel_norm_file=MEL_NORM_FILE,
+        dvae_checkpoint=DVAE_CHECKPOINT,
+        xtts_checkpoint=XTTS_CHECKPOINT,  # checkpoint path of the model that you want to fine-tune
+        tokenizer_file=TOKENIZER_FILE,
+        gpt_num_audio_tokens=1026,
+        gpt_start_audio_token=1024,
+        gpt_stop_audio_token=1025,
+        gpt_use_masking_gt_prompt_approach=True,
+        gpt_use_perceiver_resampler=True,
+    )
+    # define audio config
+    audio_config = XttsAudioConfig(sample_rate=22050, dvae_sample_rate=22050, output_sample_rate=24000)
+    # training parameters config
+    config = GPTTrainerConfig(
+        output_path=OUT_PATH,
+        model_args=model_args,
+        run_name=RUN_NAME,
+        project_name=PROJECT_NAME,
+        run_description="""
+            GPT XTTS training
+            """,
+        dashboard_logger=DASHBOARD_LOGGER,
+        logger_uri=LOGGER_URI,
+        audio=audio_config,
+        batch_size=BATCH_SIZE,
+        batch_group_size=64,
+        eval_batch_size=BATCH_SIZE,
+        num_loader_workers=8,
+        eval_split_max_size=256,
+        print_step=1000,
+        plot_step=1000,
+        log_model_step=1000,
+        save_step=1000,
+        save_n_checkpoints=3,
+        save_checkpoints=True,
+        # target_loss="loss",
+        print_eval=True,
+        # Optimizer values like tortoise, pytorch implementation with modifications to not apply WD to non-weight parameters.
+        optimizer="AdamW",
+        optimizer_wd_only_on_weights=OPTIMIZER_WD_ONLY_ON_WEIGHTS,
+        optimizer_params={"betas": [0.9, 0.96], "eps": 1e-8, "weight_decay": 1e-2},
+        lr=5e-06,  # learning rate
+        lr_scheduler="MultiStepLR",
+        # it was adjusted accordly for the new step scheme
+        lr_scheduler_params={"milestones": [50000 * 18, 150000 * 18, 300000 * 18], "gamma": 0.5, "last_epoch": -1},
+        test_sentences=[
+            {
+                "text": "It took me quite a long time to develop a voice, and now that I have it I'm not going to be silent.",
+                "speaker_wav": SPEAKER_REFERENCE,
+                "language": LANGUAGE,
+            },
+            {
+                "text": "This cake is great. It's so delicious and moist.",
+                "speaker_wav": SPEAKER_REFERENCE,
+                "language": LANGUAGE,
+            },
+        ],
+    )
+
+    # init the model from config
+    model = GPTTrainer.init_from_config(config)
+
+    # load training samples
+    train_samples, eval_samples = load_tts_samples(
+        DATASETS_CONFIG_LIST,
+        eval_split=True,
+        eval_split_max_size=config.eval_split_max_size,
+        eval_split_size=config.eval_split_size,
+    )
+    # init the trainer and 🚀
+    trainer = Trainer(
+        TrainerArgs(
+            restore_path=None,  # xtts checkpoint is restored via xtts_checkpoint key so no need of restore it using Trainer restore_path parameter
+            skip_train_epoch=False,
+            start_with_eval=START_WITH_EVAL,
+            grad_accum_steps=GRAD_ACUMM_STEPS,
+        ),
+        config,
+        output_path=OUT_PATH,
+        model=model,
+        train_samples=train_samples,
+        eval_samples=eval_samples,
+    )
+    trainer.fit()
+
+
+if __name__ == "__main__":
+    main()