From 8b4a37d8cb2630710c219df833551fc4fa69ba03 Mon Sep 17 00:00:00 2001
From: =?UTF-8?q?Eren=20G=C3=B6lge?= <egolge@coqui.ai>
Date: Thu, 30 Sep 2021 14:43:59 +0000
Subject: [PATCH] Add DeepSpeech LibriSpeech recipe

---
 .../stt/deep_speech/train_deep_speech.py      | 96 +++++++++++++++++++
 1 file changed, 96 insertions(+)
 create mode 100644 recipes/librispeech/stt/deep_speech/train_deep_speech.py

diff --git a/recipes/librispeech/stt/deep_speech/train_deep_speech.py b/recipes/librispeech/stt/deep_speech/train_deep_speech.py
new file mode 100644
index 00000000..b691fc57
--- /dev/null
+++ b/recipes/librispeech/stt/deep_speech/train_deep_speech.py
@@ -0,0 +1,96 @@
+import os
+
+from TTS.config.shared_configs import BaseAudioConfig, BaseDatasetConfig
+from TTS.stt.configs import DeepSpeechConfig
+from TTS.stt.datasets import load_stt_samples
+from TTS.stt.datasets.downloaders import download_librispeech
+from TTS.stt.datasets.tokenizer import Tokenizer
+from TTS.stt.models.deep_speech import DeepSpeech
+from TTS.trainer import Trainer, TrainingArgs
+from TTS.utils.audio import AudioProcessor
+
+output_path = os.path.dirname(os.path.abspath(__file__))
+
+if not os.path.exists("/home/ubuntu/librispeech/LibriSpeech/train-clean-100"):
+    download_librispeech("/home/ubuntu/librispeech/", "train-clean-100")
+if not os.path.exists("/home/ubuntu/librispeech/LibriSpeech/dev-clean"):
+    download_librispeech("/home/ubuntu/librispeech/", "dev-clean")
+
+# train_dataset_config = BaseDatasetConfig(
+# name="librispeech", meta_file_train=None, path="/home/ubuntu/librispeech/LibriSpeech/train-clean-100"
+# )
+
+# eval_dataset_config = BaseDatasetConfig(
+# name="librispeech", meta_file_train=None, path="/home/ubuntu/librispeech/LibriSpeech/dev-clean"
+# )
+
+train_dataset_config = BaseDatasetConfig(
+    name="ljspeech",
+    meta_file_train="metadata.csv",
+    path="/home/ubuntu/ljspeech/LJSpeech-1.1/",
+)
+
+
+audio_config = BaseAudioConfig(
+    sample_rate=16000,
+    resample=False,
+    frame_shift_ms=20,
+    frame_length_ms=40,
+    num_mels=26,
+    num_mfcc=26,
+    do_trim_silence=False,
+    mel_fmin=0,
+    mel_fmax=8000,
+)
+
+config = DeepSpeechConfig(
+    audio=audio_config,
+    run_name="deepspeech_librispeech",
+    batch_size=128,
+    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,
+    print_step=25,
+    print_eval=False,
+    mixed_precision=True,
+    max_seq_len=500000,
+    output_path=output_path,
+    train_datasets=[train_dataset_config],
+    # eval_datasets=[eval_dataset_config],
+)
+
+# init audio processor
+ap = AudioProcessor(**config.audio.to_dict())
+
+# load training samples
+train_samples, eval_samples = load_stt_samples(train_dataset_config, eval_split=True)
+# eval_samples, _ = load_stt_samples(eval_dataset_config, eval_split=False)
+transcripts = [s["text"] for s in train_samples]
+
+# init tokenizer
+tokenizer = Tokenizer(transcripts=transcripts)
+n_tokens = tokenizer.vocab_size
+config.model_args.n_tokens = n_tokens
+config.vocabulary = tokenizer.vocab_dict
+
+# init model
+model = DeepSpeech(config)
+
+# init training and kick it 🚀
+# args, config, output_path, _, c_logger, tb_logger = init_training(TrainingArgs(), config)
+trainer = Trainer(
+    TrainingArgs(),
+    config,
+    output_path,
+    model=model,
+    tokenizer=tokenizer,
+    train_samples=train_samples,
+    eval_samples=eval_samples,
+    cudnn_benchmark=False,
+    training_assets={"tokenizer": tokenizer, "audio_processor": ap},
+)
+trainer.fit()