Create to_generate.py

to_generate.py

@@ -0,0 +1,131 @@
+#!/usr/bin/env python3
+# -*- coding: utf-8 -*-
+import glob
+import json
+import os
+from functools import partial
+from pathlib import Path
+
+import datasets
+
+VERSION = datasets.Version("0.0.1")
+
+SAMPLE_RATE = 16000
+
+URLS = {'introduction_psychology': 'data/introduction_psychology.zip',
+        'a_descriptive_statistics': 'data/a_descriptive_statistics.zip',
+        'inherited_how_does_the_internet_work': 'data/inherited_how_does_the_internet_work.zip',
+        'data_structures': 'data/data_structures.zip',
+        'computational_thinking': 'data/computational_thinking.zip',
+        'physics_intro': 'data/physics_intro.zip',
+        'data_intro': 'data/data_intro.zip',
+        'introduction_to_the_philosophy_of_education': 'data/introduction_to_the_philosophy_of_education.zip',
+        'yad_vashem': 'data/yad_vashem.zip',
+        'algorithms': 'data/algorithms.zip',
+        'blue-and-white_tv': 'data/blue-and-white_tv.zip',
+        'covid_mindfulness': 'data/covid_mindfulness.zip',
+        'preparation_for_a_job_interview-_ways_to_success': 'data/preparation_for_a_job_interview-_ways_to_success.zip',
+        'what_is_the_world_introduction_to_general_chemistry': 'data/what_is_the_world_introduction_to_general_chemistry.zip',
+        'networking_create_a_network_of_professional_progress': 'data/networking_create_a_network_of_professional_progress.zip',
+        'how_to_learn': 'data/how_to_learn.zip',
+        'post_modern_education': 'data/post_modern_education.zip',
+        'introduction_to_renewable_energy': 'data/introduction_to_renewable_energy.zip',
+        'science_communication': 'data/science_communication.zip',
+        'introduction_to_physics_-_mechanics': 'data/introduction_to_physics_-_mechanics.zip',
+        'negotiations_different_cultures_and_what_between_them': 'data/negotiations_different_cultures_and_what_between_them.zip',
+        'computation_models': 'data/computation_models.zip',
+        'object_oriented_programming': 'data/object_oriented_programming.zip',
+        'from_another_angle_math': 'data/from_another_angle_math.zip'}
+
+
+class CampusHebrewSpeech(datasets.GeneratorBasedBuilder):
+    BUILDER_CONFIGS = [
+        datasets.BuilderConfig(name="campus_hebrew_speech",
+                               version=VERSION,
+                               description=f"Campus Hebrew Speech Recognition dataset")
+    ]
+
+    def _info(self):
+        return datasets.DatasetInfo(
+            description="Hebrew speech datasets",
+            features=datasets.Features(
+                {
+                    "uid": datasets.Value("string"),
+                    "file_id": datasets.Value("string"),
+                    "audio": datasets.Audio(sampling_rate=SAMPLE_RATE),
+                    "sentence": datasets.Value("string"),
+                    "n_segment": datasets.Value("int32"),
+                    "duration_ms": datasets.Value("float32"),
+                    "language": datasets.Value("string"),
+                    "sample_rate": datasets.Value("int32"),
+                    "course": datasets.Value("string"),
+                    "sentence_length": datasets.Value("int32"),
+                    "n_tokens": datasets.Value("int32"),
+                }
+            ),
+            supervised_keys=("audio", "sentence"),
+            homepage="https://huggingface.co/datasets/imvladikon/hebrew_speech_campus",
+            citation="TODO",
+        )
+
+    def _split_generators(self, dl_manager):
+        # course_links = {
+        #     course: dl_manager.download_and_extract(link) + "/" + course
+        #     for course, link in URLS.items()
+        # }
+        course_links = {
+            'from_another_angle-_mathematics_teaching_practices': 'data/from_another_angle-_mathematics_teaching_practices',
+            'introduction_to_renewable_energy': 'data/introduction_to_renewable_energy',
+            'negotiations_different_cultures_and_what_between_them': 'data/negotiations_different_cultures_and_what_between_them',
+            'introduction_to_physics_-_mechanics': 'data/introduction_to_physics_-_mechanics',
+            'networking-_create_a_network_of_professional_progress': 'data/networking-_create_a_network_of_professional_progress',
+            'post_modern_education': 'data/post_modern_education',
+            'physics_intro': 'data/physics_intro',
+            'algorithms': 'data/algorithms',
+            'blue-and-white_tv': 'data/blue-and-white_tv',
+            'what_is_the_world-_introduction_to_general_chemistry': 'data/what_is_the_world-_introduction_to_general_chemistry',
+            'computational_thinking': 'data/computational_thinking',
+            'preparation_for_a_job_interview-_ways_to_success': 'data/preparation_for_a_job_interview-_ways_to_success',
+            'a_descriptive_statistics': 'data/a_descriptive_statistics',
+            'yad_vashem': 'data/yad_vashem',
+            'introduction_to_the_philosophy_of_education': 'data/introduction_to_the_philosophy_of_education',
+            'covid_mindfulness': 'data/covid_mindfulness',
+            'data_structures': 'data/data_structures',
+            'computation_models': 'data/computation_models',
+            'introduction_psychology': 'data/introduction_psychology',
+            'how_to_learn': 'data/how_to_learn',
+            'inherited_-_how_does_the_internet_work': 'data/inherited_-_how_does_the_internet_work',
+            'object_oriented_programming': 'data/object_oriented_programming',
+            'science_communication': 'data/science_communication',
+            'data_intro': 'data/data_intro'}
+
+        return [
+            datasets.SplitGenerator(
+                name=datasets.Split.TRAIN,
+                gen_kwargs={
+                    "course_links": course_links,
+                    "split": "train"},
+            )
+        ]
+
+    def _generate_examples(self, course_links, split):
+        idx = 0
+        for course, root_path in course_links.items():
+            root_path = "/content/hebrew_speech_campus/" + root_path
+            for metadata_file in Path(root_path).glob("*.json"):
+                audio_file = Path(metadata_file).stem + ".wav"
+                metadata = json.load(open(metadata_file, encoding="utf-8"))
+                yield idx, {
+                    "uid": metadata["file"].split("_")[0],
+                    "file_id": Path(metadata["file"]).stem,
+                    "audio": os.path.join(root_path, audio_file),
+                    "sentence": metadata["text"],
+                    "n_segment": metadata["n_segment"],
+                    "duration_ms": 1000 * metadata["duration"],
+                    "language": metadata["language"],
+                    "sample_rate": SAMPLE_RATE,
+                    "course": course,
+                    "sentence_length": len(metadata["text"]),
+                    "n_tokens": metadata["text"].count(" ") + 1,
+                }
+                idx += 1