VCTK recipes (finally 🚀)

recipes/ljspeech/align_tts/train_aligntts.py

@@ -1,7 +1,7 @@
 import os
 
 from TTS.trainer import Trainer, TrainingArgs
-from TTS.tts.configs import AlignTTSConfig, BaseDatasetConfig
+from TTS.tts.configs.align_tts_config import AlignTTSConfig, BaseDatasetConfig
 from TTS.tts.datasets import load_tts_samples
 from TTS.tts.models.align_tts import AlignTTS
 from TTS.utils.audio import AudioProcessor

recipes/ljspeech/fast_pitch/train_fast_pitch.py

@@ -1,8 +1,8 @@
 import os
 
-from TTS.config import BaseAudioConfig, BaseDatasetConfig
+from TTS.config.shared_configs import BaseAudioConfig, BaseDatasetConfig
 from TTS.trainer import Trainer, TrainingArgs
-from TTS.tts.configs import FastPitchConfig
+from TTS.tts.configs.fast_pitch_config import FastPitchConfig
 from TTS.tts.datasets import load_tts_samples
 from TTS.tts.models.forward_tts import ForwardTTS
 from TTS.utils.audio import AudioProcessor

recipes/ljspeech/fast_speech/train_fast_speech.py

@@ -2,7 +2,7 @@ import os
 
 from TTS.config import BaseAudioConfig, BaseDatasetConfig
 from TTS.trainer import Trainer, TrainingArgs
-from TTS.tts.configs import FastSpeechConfig
+from TTS.tts.configs.fast_speech_config import FastSpeechConfig
 from TTS.tts.datasets import load_tts_samples
 from TTS.tts.models.forward_tts import ForwardTTS
 from TTS.utils.audio import AudioProcessor

recipes/ljspeech/glow_tts/train_glowtts.py

@@ -1,7 +1,8 @@
 import os
 
 from TTS.trainer import Trainer, TrainingArgs
-from TTS.tts.configs import BaseDatasetConfig, GlowTTSConfig
+from TTS.tts.configs.glow_tts_config import GlowTTSConfig
+from TTS.tts.configs.shared_configs import BaseDatasetConfig
 from TTS.tts.datasets import load_tts_samples
 from TTS.tts.models.glow_tts import GlowTTS
 from TTS.utils.audio import AudioProcessor

recipes/ljspeech/speedy_speech/train_speedy_speech.py

@@ -2,7 +2,7 @@ import os
 
 from TTS.config import BaseAudioConfig, BaseDatasetConfig
 from TTS.trainer import Trainer, TrainingArgs
-from TTS.tts.configs import SpeedySpeechConfig
+from TTS.tts.configs.speedy_speech_config import SpeedySpeechConfig
 from TTS.tts.datasets import load_tts_samples
 from TTS.tts.models.forward_tts import ForwardTTS
 from TTS.utils.audio import AudioProcessor

recipes/vctk/download_vctk.sh

@@ -0,0 +1,12 @@
+#!/usr/bin/env bash
+# take the scripts's parent's directory to prefix all the output paths.
+RUN_DIR="$( cd "$( dirname "${BASH_SOURCE[0]}" )" &> /dev/null && pwd )"
+echo $RUN_DIR
+# download LJSpeech dataset
+wget https://datashare.ed.ac.uk/bitstream/handle/10283/3443/VCTK-Corpus-0.92.zip -O VCTK-Corpus-0.92.zip
+# extract
+mkdir VCTK
+unzip VCTK-Corpus-0.92 -d VCTK
+# create train-val splits
+mv VCTK $RUN_DIR/recipes/vctk/
+rm VCTK-Corpus-0.92.zip

recipes/vctk/fast_pitch/train_fast_pitch.py

@@ -0,0 +1,80 @@
+import os
+
+from TTS.config import BaseAudioConfig, BaseDatasetConfig
+from TTS.trainer import Trainer, TrainingArgs
+from TTS.tts.configs.fast_pitch_config import FastPitchConfig
+from TTS.tts.datasets import load_tts_samples
+from TTS.tts.models.forward_tts import ForwardTTS
+from TTS.tts.utils.speakers import SpeakerManager
+from TTS.utils.audio import AudioProcessor
+
+output_path = os.path.dirname(os.path.abspath(__file__))
+dataset_config = BaseDatasetConfig(name="vctk", meta_file_train="", path=os.path.join(output_path, "../VCTK/"))
+
+audio_config = BaseAudioConfig(
+    sample_rate=22050,
+    do_trim_silence=True,
+    trim_db=23.0,
+    signal_norm=False,
+    mel_fmin=0.0,
+    mel_fmax=8000,
+    spec_gain=1.0,
+    log_func="np.log",
+    ref_level_db=20,
+    preemphasis=0.0,
+)
+
+config = FastPitchConfig(
+    run_name="fast_pitch_ljspeech",
+    audio=audio_config,
+    batch_size=32,
+    eval_batch_size=16,
+    num_loader_workers=8,
+    num_eval_loader_workers=4,
+    compute_input_seq_cache=True,
+    compute_f0=True,
+    f0_cache_path=os.path.join(output_path, "f0_cache"),
+    run_eval=True,
+    test_delay_epochs=-1,
+    epochs=1000,
+    text_cleaner="english_cleaners",
+    use_phonemes=True,
+    use_espeak_phonemes=False,
+    phoneme_language="en-us",
+    phoneme_cache_path=os.path.join(output_path, "phoneme_cache"),
+    print_step=50,
+    print_eval=False,
+    mixed_precision=False,
+    sort_by_audio_len=True,
+    max_seq_len=500000,
+    output_path=output_path,
+    datasets=[dataset_config],
+    use_speaker_embedding=True,
+)
+
+# init audio processor
+ap = AudioProcessor(**config.audio)
+
+# load training samples
+train_samples, eval_samples = load_tts_samples(dataset_config, eval_split=True)
+
+# init speaker manager for multi-speaker training
+# it maps speaker-id to speaker-name in the model and data-loader
+speaker_manager = SpeakerManager()
+speaker_manager.set_speaker_ids_from_data(train_samples + eval_samples)
+config.model_args.num_speakers = speaker_manager.num_speakers
+
+# init model
+model = ForwardTTS(config, speaker_manager)
+
+# init the trainer and 🚀
+trainer = Trainer(
+    TrainingArgs(),
+    config,
+    output_path,
+    model=model,
+    train_samples=train_samples,
+    eval_samples=eval_samples,
+    training_assets={"audio_processor": ap},
+)
+trainer.fit()

recipes/vctk/fast_speech/train_fast_speech.py

@@ -0,0 +1,80 @@
+import os
+
+from TTS.config import BaseAudioConfig, BaseDatasetConfig
+from TTS.trainer import Trainer, TrainingArgs
+from TTS.tts.configs.fast_speech_config import FastSpeechConfig
+from TTS.tts.datasets import load_tts_samples
+from TTS.tts.models.forward_tts import ForwardTTS
+from TTS.tts.utils.speakers import SpeakerManager
+from TTS.utils.audio import AudioProcessor
+
+output_path = os.path.dirname(os.path.abspath(__file__))
+dataset_config = BaseDatasetConfig(name="vctk", meta_file_train="", path=os.path.join(output_path, "../VCTK/"))
+
+audio_config = BaseAudioConfig(
+    sample_rate=22050,
+    do_trim_silence=True,
+    trim_db=23.0,
+    signal_norm=False,
+    mel_fmin=0.0,
+    mel_fmax=8000,
+    spec_gain=1.0,
+    log_func="np.log",
+    ref_level_db=20,
+    preemphasis=0.0,
+)
+
+config = FastSpeechConfig(
+    run_name="fast_pitch_ljspeech",
+    audio=audio_config,
+    batch_size=32,
+    eval_batch_size=16,
+    num_loader_workers=8,
+    num_eval_loader_workers=4,
+    compute_input_seq_cache=True,
+    compute_f0=True,
+    f0_cache_path=os.path.join(output_path, "f0_cache"),
+    run_eval=True,
+    test_delay_epochs=-1,
+    epochs=1000,
+    text_cleaner="english_cleaners",
+    use_phonemes=True,
+    use_espeak_phonemes=False,
+    phoneme_language="en-us",
+    phoneme_cache_path=os.path.join(output_path, "phoneme_cache"),
+    print_step=50,
+    print_eval=False,
+    mixed_precision=False,
+    sort_by_audio_len=True,
+    max_seq_len=500000,
+    output_path=output_path,
+    datasets=[dataset_config],
+    use_speaker_embedding=True,
+)
+
+# init audio processor
+ap = AudioProcessor(**config.audio)
+
+# load training samples
+train_samples, eval_samples = load_tts_samples(dataset_config, eval_split=True)
+
+# init speaker manager for multi-speaker training
+# it maps speaker-id to speaker-name in the model and data-loader
+speaker_manager = SpeakerManager()
+speaker_manager.set_speaker_ids_from_data(train_samples + eval_samples)
+config.model_args.num_speakers = speaker_manager.num_speakers
+
+# init model
+model = ForwardTTS(config, speaker_manager)
+
+# init the trainer and 🚀
+trainer = Trainer(
+    TrainingArgs(),
+    config,
+    output_path,
+    model=model,
+    train_samples=train_samples,
+    eval_samples=eval_samples,
+    training_assets={"audio_processor": ap},
+)
+trainer.fit()

recipes/vctk/glow_tts/train_glow_tts.py

@@ -0,0 +1,62 @@
+import os
+
+from TTS.config.shared_configs import BaseAudioConfig
+from TTS.trainer import Trainer, TrainingArgs
+from TTS.tts.configs.glow_tts_config import GlowTTSConfig
+from TTS.tts.configs.shared_configs import BaseDatasetConfig
+from TTS.tts.datasets import load_tts_samples
+from TTS.tts.models.glow_tts import GlowTTS
+from TTS.tts.utils.speakers import SpeakerManager
+from TTS.utils.audio import AudioProcessor
+
+output_path = os.path.dirname(os.path.abspath(__file__))
+dataset_config = BaseDatasetConfig(name="vctk", meta_file_train="", path=os.path.join(output_path, "../VCTK/"))
+
+audio_config = BaseAudioConfig(sample_rate=22050, do_trim_silence=True, trim_db=23.0)
+
+config = GlowTTSConfig(
+    batch_size=64,
+    eval_batch_size=16,
+    num_loader_workers=4,
+    num_eval_loader_workers=4,
+    run_eval=True,
+    test_delay_epochs=-1,
+    epochs=1000,
+    text_cleaner="phoneme_cleaners",
+    use_phonemes=True,
+    phoneme_language="en-us",
+    phoneme_cache_path=os.path.join(output_path, "phoneme_cache"),
+    print_step=25,
+    print_eval=False,
+    mixed_precision=True,
+    output_path=output_path,
+    datasets=[dataset_config],
+    use_speaker_embedding=True,
+)
+
+# init audio processor
+ap = AudioProcessor(**config.audio.to_dict())
+
+# load training samples
+train_samples, eval_samples = load_tts_samples(dataset_config, eval_split=True)
+
+# init speaker manager for multi-speaker training
+# it maps speaker-id to speaker-name in the model and data-loader
+speaker_manager = SpeakerManager()
+speaker_manager.set_speaker_ids_from_data(train_samples + eval_samples)
+config.num_speakers = speaker_manager.num_speakers
+
+# init model
+model = GlowTTS(config, speaker_manager)
+
+# init the trainer and 🚀
+trainer = Trainer(
+    TrainingArgs(),
+    config,
+    output_path,
+    model=model,
+    train_samples=train_samples,
+    eval_samples=eval_samples,
+    training_assets={"audio_processor": ap},
+)
+trainer.fit()

recipes/vctk/speedy_speech/train_speedy_speech.py

@@ -0,0 +1,80 @@
+import os
+
+from TTS.config import BaseAudioConfig, BaseDatasetConfig
+from TTS.trainer import Trainer, TrainingArgs
+from TTS.tts.configs.speedy_speech_config import SpeedySpeechConfig
+from TTS.tts.datasets import load_tts_samples
+from TTS.tts.models.forward_tts import ForwardTTS
+from TTS.tts.utils.speakers import SpeakerManager
+from TTS.utils.audio import AudioProcessor
+
+output_path = os.path.dirname(os.path.abspath(__file__))
+dataset_config = BaseDatasetConfig(name="vctk", meta_file_train="", path=os.path.join(output_path, "../VCTK/"))
+
+audio_config = BaseAudioConfig(
+    sample_rate=22050,
+    do_trim_silence=True,
+    trim_db=23.0,
+    signal_norm=False,
+    mel_fmin=0.0,
+    mel_fmax=8000,
+    spec_gain=1.0,
+    log_func="np.log",
+    ref_level_db=20,
+    preemphasis=0.0,
+)
+
+config = SpeedySpeechConfig(
+    run_name="fast_pitch_ljspeech",
+    audio=audio_config,
+    batch_size=32,
+    eval_batch_size=16,
+    num_loader_workers=8,
+    num_eval_loader_workers=4,
+    compute_input_seq_cache=True,
+    compute_f0=True,
+    f0_cache_path=os.path.join(output_path, "f0_cache"),
+    run_eval=True,
+    test_delay_epochs=-1,
+    epochs=1000,
+    text_cleaner="english_cleaners",
+    use_phonemes=True,
+    use_espeak_phonemes=False,
+    phoneme_language="en-us",
+    phoneme_cache_path=os.path.join(output_path, "phoneme_cache"),
+    print_step=50,
+    print_eval=False,
+    mixed_precision=False,
+    sort_by_audio_len=True,
+    max_seq_len=500000,
+    output_path=output_path,
+    datasets=[dataset_config],
+    use_speaker_embedding=True,
+)
+
+# init audio processor
+ap = AudioProcessor(**config.audio)
+
+# load training samples
+train_samples, eval_samples = load_tts_samples(dataset_config, eval_split=True)
+
+# init speaker manager for multi-speaker training
+# it maps speaker-id to speaker-name in the model and data-loader
+speaker_manager = SpeakerManager()
+speaker_manager.set_speaker_ids_from_data(train_samples + eval_samples)
+config.model_args.num_speakers = speaker_manager.num_speakers
+
+# init model
+model = ForwardTTS(config, speaker_manager)
+
+# init the trainer and 🚀
+trainer = Trainer(
+    TrainingArgs(),
+    config,
+    output_path,
+    model=model,
+    train_samples=train_samples,
+    eval_samples=eval_samples,
+    training_assets={"audio_processor": ap},
+)
+trainer.fit()

recipes/vctk/tacotron-DDC/train_tacotron-DDC.py

@@ -0,0 +1,80 @@
+import os
+
+from TTS.config.shared_configs import BaseAudioConfig
+from TTS.trainer import Trainer, TrainingArgs
+from TTS.tts.configs.shared_configs import BaseDatasetConfig
+from TTS.tts.configs.tacotron_config import TacotronConfig
+from TTS.tts.datasets import load_tts_samples
+from TTS.tts.models.tacotron import Tacotron
+from TTS.tts.utils.speakers import SpeakerManager
+from TTS.utils.audio import AudioProcessor
+
+output_path = os.path.dirname(os.path.abspath(__file__))
+dataset_config = BaseDatasetConfig(name="vctk", meta_file_train="", path=os.path.join(output_path, "../VCTK/"))
+
+audio_config = BaseAudioConfig(
+    sample_rate=22050,
+    resample=True,  # Resample to 22050 Hz. It slows down training. Use `TTS/bin/resample.py` to pre-resample and set this False for faster training.
+    do_trim_silence=True,
+    trim_db=23.0,
+    signal_norm=False,
+    mel_fmin=0.0,
+    mel_fmax=8000,
+    spec_gain=1.0,
+    log_func="np.log",
+    ref_level_db=20,
+    preemphasis=0.0,
+)
+
+config = TacotronConfig(  # This is the config that is saved for the future use
+    audio=audio_config,
+    batch_size=48,
+    eval_batch_size=16,
+    num_loader_workers=4,
+    num_eval_loader_workers=4,
+    run_eval=True,
+    test_delay_epochs=-1,
+    r=6,
+    gradual_training=[[0, 6, 48], [10000, 4, 32], [50000, 3, 32], [100000, 2, 32]],
+    double_decoder_consistency=True,
+    epochs=1000,
+    text_cleaner="phoneme_cleaners",
+    use_phonemes=True,
+    phoneme_language="en-us",
+    phoneme_cache_path=os.path.join(output_path, "phoneme_cache"),
+    print_step=25,
+    print_eval=False,
+    mixed_precision=True,
+    sort_by_audio_len=True,
+    min_seq_len=0,
+    max_seq_len=44000 * 10,  # 44k is the original sampling rate before resampling, corresponds to 10 seconds of audio
+    output_path=output_path,
+    datasets=[dataset_config],
+    use_speaker_embedding=True,  # set this to enable multi-sepeaker training
+)
+
+# init audio processor
+ap = AudioProcessor(**config.audio.to_dict())
+
+# load training samples
+train_samples, eval_samples = load_tts_samples(dataset_config, eval_split=True)
+
+# init speaker manager for multi-speaker training
+# it mainly handles speaker-id to speaker-name for the model and the data-loader
+speaker_manager = SpeakerManager()
+speaker_manager.set_speaker_ids_from_data(train_samples + eval_samples)
+
+# init model
+model = Tacotron(config, speaker_manager)
+
+# init the trainer and 🚀
+trainer = Trainer(
+    TrainingArgs(),
+    config,
+    output_path,
+    model=model,
+    train_samples=train_samples,
+    eval_samples=eval_samples,
+    training_assets={"audio_processor": ap},
+)
+trainer.fit()

recipes/vctk/tacotron2-DDC/train_tacotron2-ddc.py

@@ -0,0 +1,87 @@
+import os
+
+from TTS.config.shared_configs import BaseAudioConfig
+from TTS.trainer import Trainer, TrainingArgs
+from TTS.tts.configs.shared_configs import BaseDatasetConfig
+from TTS.tts.configs.tacotron2_config import Tacotron2Config
+from TTS.tts.datasets import load_tts_samples
+from TTS.tts.models.tacotron2 import Tacotron2
+from TTS.tts.utils.speakers import SpeakerManager
+from TTS.utils.audio import AudioProcessor
+
+output_path = os.path.dirname(os.path.abspath(__file__))
+dataset_config = BaseDatasetConfig(name="vctk", meta_file_train="", path=os.path.join(output_path, "../VCTK/"))
+
+audio_config = BaseAudioConfig(
+    sample_rate=22050,
+    resample=False,  # Resample to 22050 Hz. It slows down training. Use `TTS/bin/resample.py` to pre-resample and set this False for faster training.
+    do_trim_silence=True,
+    trim_db=23.0,
+    signal_norm=False,
+    mel_fmin=0.0,
+    mel_fmax=8000,
+    spec_gain=1.0,
+    log_func="np.log",
+    preemphasis=0.0,
+)
+
+config = Tacotron2Config(  # This is the config that is saved for the future use
+    audio=audio_config,
+    batch_size=32,
+    eval_batch_size=16,
+    num_loader_workers=4,
+    num_eval_loader_workers=4,
+    run_eval=True,
+    test_delay_epochs=-1,
+    r=2,
+    # gradual_training=[[0, 6, 48], [10000, 4, 32], [50000, 3, 32], [100000, 2, 32]],
+    double_decoder_consistency=False,
+    epochs=1000,
+    text_cleaner="phoneme_cleaners",
+    use_phonemes=True,
+    phoneme_language="en-us",
+    phoneme_cache_path=os.path.join(output_path, "phoneme_cache"),
+    print_step=150,
+    print_eval=False,
+    mixed_precision=True,
+    sort_by_audio_len=True,
+    min_seq_len=14800,
+    max_seq_len=22050 * 10,  # 44k is the original sampling rate before resampling, corresponds to 10 seconds of audio
+    output_path=output_path,
+    datasets=[dataset_config],
+    use_speaker_embedding=True,  # set this to enable multi-sepeaker training
+    decoder_ssim_alpha=0.0,  # disable ssim losses that causes NaN for some runs.
+    postnet_ssim_alpha=0.0,
+    postnet_diff_spec_alpha=0.0,
+    decoder_diff_spec_alpha=0.0,
+    attention_norm="softmax",
+    optimizer="Adam",
+    lr_scheduler=None,
+    lr=3e-5,
+)
+
+# init audio processor
+ap = AudioProcessor(**config.audio.to_dict())
+
+# load training samples
+train_samples, eval_samples = load_tts_samples(dataset_config, eval_split=True)
+
+# init speaker manager for multi-speaker training
+# it mainly handles speaker-id to speaker-name for the model and the data-loader
+speaker_manager = SpeakerManager()
+speaker_manager.set_speaker_ids_from_data(train_samples + eval_samples)
+
+# init model
+model = Tacotron2(config, speaker_manager)
+
+# init the trainer and 🚀
+trainer = Trainer(
+    TrainingArgs(),
+    config,
+    output_path,
+    model=model,
+    train_samples=train_samples,
+    eval_samples=eval_samples,
+    training_assets={"audio_processor": ap},
+)
+trainer.fit()

recipes/vctk/vits/train_vits.py

@@ -0,0 +1,86 @@
+import os
+
+from TTS.config.shared_configs import BaseAudioConfig
+from TTS.trainer import Trainer, TrainingArgs
+from TTS.tts.configs.shared_configs import BaseDatasetConfig
+from TTS.tts.configs.vits_config import VitsConfig
+from TTS.tts.datasets import load_tts_samples
+from TTS.tts.models.vits import Vits
+from TTS.tts.utils.speakers import SpeakerManager
+from TTS.utils.audio import AudioProcessor
+
+output_path = os.path.dirname(os.path.abspath(__file__))
+dataset_config = BaseDatasetConfig(name="vctk", meta_file_train="", path=os.path.join(output_path, "../VCTK/"))
+
+
+audio_config = BaseAudioConfig(
+    sample_rate=22050,
+    win_length=1024,
+    hop_length=256,
+    num_mels=80,
+    preemphasis=0.0,
+    ref_level_db=20,
+    log_func="np.log",
+    do_trim_silence=True,
+    trim_db=23.0,
+    mel_fmin=0,
+    mel_fmax=None,
+    spec_gain=1.0,
+    signal_norm=False,
+    do_amp_to_db_linear=False,
+    resample=True,
+)
+
+config = VitsConfig(
+    audio=audio_config,
+    run_name="vits_vctk",
+    use_speaker_embedding=True,
+    batch_size=32,
+    eval_batch_size=16,
+    batch_group_size=5,
+    num_loader_workers=4,
+    num_eval_loader_workers=4,
+    run_eval=True,
+    test_delay_epochs=-1,
+    epochs=1000,
+    text_cleaner="english_cleaners",
+    use_phonemes=True,
+    phoneme_language="en-us",
+    phoneme_cache_path=os.path.join(output_path, "phoneme_cache"),
+    compute_input_seq_cache=True,
+    print_step=25,
+    print_eval=False,
+    mixed_precision=True,
+    sort_by_audio_len=True,
+    min_seq_len=32 * 256 * 4,
+    max_seq_len=1500000,
+    output_path=output_path,
+    datasets=[dataset_config],
+)
+
+# init audio processor
+ap = AudioProcessor(**config.audio.to_dict())
+
+# load training samples
+train_samples, eval_samples = load_tts_samples(dataset_config, eval_split=True)
+
+# init speaker manager for multi-speaker training
+# it maps speaker-id to speaker-name in the model and data-loader
+speaker_manager = SpeakerManager()
+speaker_manager.set_speaker_ids_from_data(train_samples + eval_samples)
+config.model_args.num_speakers = speaker_manager.num_speakers
+
+# init model
+model = Vits(config, speaker_manager)
+
+# init the trainer and 🚀
+trainer = Trainer(
+    TrainingArgs(),
+    config,
+    output_path,
+    model=model,
+    train_samples=train_samples,
+    eval_samples=eval_samples,
+    training_assets={"audio_processor": ap},
+)
+trainer.fit()