Coqui stt model for large Arabic vocabulary

LICENCE.txt

@@ -0,0 +1,21 @@
+MIT License
+
+Copyright (c) 2021 Anas Hasni
+
+Permission is hereby granted, free of charge, to any person obtaining a copy
+of this software and associated documentation files (the "Software"), to deal
+in the Software without restriction, including without limitation the rights
+to use, copy, modify, merge, publish, distribute, sublicense, and/or sell
+copies of the Software, and to permit persons to whom the Software is
+furnished to do so, subject to the following conditions:
+
+The above copyright notice and this permission notice shall be included in all
+copies or substantial portions of the Software.
+
+THE SOFTWARE IS PROVIDED "AS IS", WITHOUT WARRANTY OF ANY KIND, EXPRESS OR
+IMPLIED, INCLUDING BUT NOT LIMITED TO THE WARRANTIES OF MERCHANTABILITY,
+FITNESS FOR A PARTICULAR PURPOSE AND NONINFRINGEMENT. IN NO EVENT SHALL THE
+AUTHORS OR COPYRIGHT HOLDERS BE LIABLE FOR ANY CLAIM, DAMAGES OR OTHER
+LIABILITY, WHETHER IN AN ACTION OF CONTRACT, TORT OR OTHERWISE, ARISING FROM,
+OUT OF OR IN CONNECTION WITH THE SOFTWARE OR THE USE OR OTHER DEALINGS IN THE
+SOFTWARE.

alphabet.txt

@@ -0,0 +1,36 @@
+ 
+ء
+ؤ
+ا
+ب
+ة
+ت
+ث
+ج
+ح
+خ
+د
+ذ
+ر
+ز
+س
+ش
+ص
+ض
+ط
+ظ
+ع
+غ
+ف
+ق
+ك
+ل
+م
+ن
+ه
+و
+ى
+ي
+ٓ
+ٔ
+ی

best_dev_checkpoint

@@ -0,0 +1,2 @@
+model_checkpoint_path: "best_dev-25350"
+all_model_checkpoint_paths: "best_dev-25350"

flags.txt

@@ -0,0 +1,98 @@
+{
+  "train_files": [
+    "ar-train-dev/train.csv"
+  ],
+  "dev_files": [
+    "ar-train-dev/dev.csv"
+  ],
+  "test_files": [],
+  "metrics_files": [],
+  "read_buffer": 1048576,
+  "feature_cache": "",
+  "cache_for_epochs": 0,
+  "feature_win_len": 32,
+  "feature_win_step": 20,
+  "audio_sample_rate": 16000,
+  "normalize_sample_rate": true,
+  "augment": null,
+  "epochs": 20,
+  "dropout_rate": 0.05,
+  "dropout_rate2": 0.05,
+  "dropout_rate3": 0.05,
+  "dropout_rate4": 0.0,
+  "dropout_rate5": 0.0,
+  "dropout_rate6": 0.05,
+  "relu_clip": 20.0,
+  "beta1": 0.9,
+  "beta2": 0.999,
+  "epsilon": 1e-08,
+  "learning_rate": 0.001,
+  "train_batch_size": 16,
+  "dev_batch_size": 16,
+  "test_batch_size": 1,
+  "export_batch_size": 1,
+  "inter_op_parallelism_threads": 0,
+  "intra_op_parallelism_threads": 0,
+  "use_allow_growth": false,
+  "load_cudnn": false,
+  "train_cudnn": false,
+  "automatic_mixed_precision": false,
+  "limit_train": 0,
+  "limit_dev": 0,
+  "limit_test": 0,
+  "reverse_train": false,
+  "reverse_dev": false,
+  "reverse_test": false,
+  "checkpoint_dir": "",
+  "load_checkpoint_dir": "/workspace/.local/share/stt/checkpoints",
+  "save_checkpoint_dir": "checkpoints_arabic",
+  "checkpoint_secs": 600,
+  "max_to_keep": 5,
+  "load_train": "init",
+  "load_evaluate": "auto",
+  "drop_source_layers": 0,
+  "export_dir": "",
+  "remove_export": false,
+  "export_tflite": false,
+  "n_steps": 16,
+  "export_zip": false,
+  "export_file_name": "output_graph",
+  "export_beam_width": 500,
+  "export_author_id": "author",
+  "export_model_name": "model",
+  "export_model_version": "0.0.1",
+  "export_contact_info": "<public contact information of the author. Can be an email address, or a link to a contact form, issue tracker, or discussion forum. Must provide a way to reach the model authors>",
+  "export_license": "<SPDX identifier of the license of the exported model. See https://spdx.org/licenses/. If the license does not have an SPDX identifier, use the license name.>",
+  "export_language": "<language the model was trained on - IETF BCP 47 language tag including at least language, script and region subtags. E.g. \"en-Latn-UK\" or \"de-Latn-DE\" or \"cmn-Hans-CN\". Include as much info as you can without loss of precision. For example, if a model is trained on Scottish English, include the variant subtag: \"en-Latn-GB-Scotland\".>",
+  "export_min_stt_version": "<minimum Coqui STT version (inclusive) the exported model is compatible with>",
+  "export_max_stt_version": "<maximum Coqui STT version (inclusive) the exported model is compatible with>",
+  "export_description": "<Freeform description of the model being exported. Markdown accepted. You can also leave this flag unchanged and edit the generated .md file directly. Useful things to describe are demographic and acoustic characteristics of the data used to train the model, any architectural changes, names of public datasets that were used when applicable, hyperparameters used for training, evaluation results on standard benchmark datasets, etc.>",
+  "log_level": 1,
+  "show_progressbar": true,
+  "log_placement": false,
+  "report_count": 5,
+  "summary_dir": "/workspace/.local/share/stt/summaries",
+  "test_output_file": "",
+  "n_hidden": 256,
+  "layer_norm": false,
+  "random_seed": 4568,
+  "early_stop": false,
+  "es_epochs": 25,
+  "es_min_delta": 0.05,
+  "reduce_lr_on_plateau": false,
+  "plateau_epochs": 10,
+  "plateau_reduction": 0.1,
+  "force_initialize_learning_rate": false,
+  "bytes_output_mode": false,
+  "alphabet_config_path": "ar-train-dev/alphabet.txt",
+  "scorer_path": "",
+  "beam_width": 1024,
+  "lm_alpha": 0.931289039105002,
+  "lm_beta": 1.1834137581510284,
+  "cutoff_prob": 1.0,
+  "cutoff_top_n": 300,
+  "one_shot_infer": null,
+  "lm_alpha_max": 5,
+  "lm_beta_max": 5,
+  "n_trials": 2400
+}