Upload folder using huggingface_hub

.gitattributes

@@ -57,3 +57,5 @@ saved_model/**/* filter=lfs diff=lfs merge=lfs -text
 # Video files - compressed
 *.mp4 filter=lfs diff=lfs merge=lfs -text
 *.webm filter=lfs diff=lfs merge=lfs -text
+transcript/crowd/train.tsv filter=lfs diff=lfs merge=lfs -text
+transcript/farfield/train.tsv filter=lfs diff=lfs merge=lfs -text

README.md

@@ -1,223 +1,71 @@
----
-license: other
-license_link: https://github.com/salute-developers/golos/blob/master/license/ru.pdf
-dataset_info:
-- config_name: test_crowd_chunk_0
-  features:
-  - name: id
-    dtype: string
-  - name: audio
-    dtype:
-      audio:
-        sampling_rate: 16000
-  - name: text
-    dtype: string
-  - name: duration
-    dtype: float32
-  - name: source
-    dtype: string
-  splits:
-  - name: train
-    num_bytes: 174005420.684
-    num_examples: 9994
-  download_size: 171247463
-  dataset_size: 174005420.684
-- config_name: test_farfield_chunk_0
-  features:
-  - name: id
-    dtype: string
-  - name: audio
-    dtype:
-      audio:
-        sampling_rate: 16000
-  - name: text
-    dtype: string
-  - name: duration
-    dtype: float32
-  - name: source
-    dtype: string
-  splits:
-  - name: train
-    num_bytes: 22773121.4
-    num_examples: 1916
-  download_size: 20748082
-  dataset_size: 22773121.4
-- config_name: train_crowd_chunk_0
-  features:
-  - name: id
-    dtype: string
-  - name: audio
-    dtype:
-      audio:
-        sampling_rate: 16000
-  - name: text
-    dtype: string
-  - name: duration
-    dtype: float32
-  - name: source
-    dtype: string
-  splits:
-  - name: train
-    num_bytes: 3543023006.0
-    num_examples: 200000
-  download_size: 3348370117
-  dataset_size: 3543023006.0
-- config_name: train_crowd_chunk_1
-  features:
-  - name: id
-    dtype: string
-  - name: audio
-    dtype:
-      audio:
-        sampling_rate: 16000
-  - name: text
-    dtype: string
-  - name: duration
-    dtype: float32
-  - name: source
-    dtype: string
-  splits:
-  - name: train
-    num_bytes: 3955337318.0
-    num_examples: 200000
-  download_size: 3286435754
-  dataset_size: 3955337318.0
-- config_name: train_crowd_chunk_2
-  features:
-  - name: id
-    dtype: string
-  - name: audio
-    dtype:
-      audio:
-        sampling_rate: 16000
-  - name: text
-    dtype: string
-  - name: duration
-    dtype: float32
-  - name: source
-    dtype: string
-  splits:
-  - name: train
-    num_bytes: 4217539962.0
-    num_examples: 200000
-  download_size: 3833360601
-  dataset_size: 4217539962.0
-- config_name: train_crowd_chunk_3
-  features:
-  - name: id
-    dtype: string
-  - name: audio
-    dtype:
-      audio:
-        sampling_rate: 16000
-  - name: text
-    dtype: string
-  - name: duration
-    dtype: float32
-  - name: source
-    dtype: string
-  splits:
-  - name: train
-    num_bytes: 5302673779.0
-    num_examples: 200000
-  download_size: 3744720070
-  dataset_size: 5302673779.0
-- config_name: train_crowd_chunk_4
-  features:
-  - name: id
-    dtype: string
-  - name: audio
-    dtype:
-      audio:
-        sampling_rate: 16000
-  - name: text
-    dtype: string
-  - name: duration
-    dtype: float32
-  - name: source
-    dtype: string
-  splits:
-  - name: train
-    num_bytes: 3259869309.996
-    num_examples: 179796
-  download_size: 2598124991
-  dataset_size: 3259869309.996
-- config_name: train_farfield_chunk_0
-  features:
-  - name: id
-    dtype: string
-  - name: audio
-    dtype:
-      audio:
-        sampling_rate: 16000
-  - name: text
-    dtype: string
-  - name: duration
-    dtype: float32
-  - name: source
-    dtype: string
-  splits:
-  - name: train
-    num_bytes: 2858014558.777
-    num_examples: 124003
-  download_size: 1940591595
-  dataset_size: 2858014558.777
-configs:
-- config_name: test_crowd_chunk_0
-  data_files:
-  - split: train
-    path: test_crowd_chunk_0/train-*
-- config_name: test_farfield_chunk_0
-  data_files:
-  - split: train
-    path: test_farfield_chunk_0/train-*
-- config_name: train_crowd_chunk_0
-  data_files:
-  - split: train
-    path: train_crowd_chunk_0/train-*
-- config_name: train_crowd_chunk_1
-  data_files:
-  - split: train
-    path: train_crowd_chunk_1/train-*
-- config_name: train_crowd_chunk_2
-  data_files:
-  - split: train
-    path: train_crowd_chunk_2/train-*
-- config_name: train_crowd_chunk_3
-  data_files:
-  - split: train
-    path: train_crowd_chunk_3/train-*
-- config_name: train_crowd_chunk_4
-  data_files:
-  - split: train
-    path: train_crowd_chunk_4/train-*
-- config_name: train_farfield_chunk_0
-  data_files:
-  - split: train
-    path: train_farfield_chunk_0/train-*
----
-
-Licensing Information
-
-
-English Version
-https://github.com/salute-developers/golos/blob/master/license/en_us.pdf
-
-Russian Version
-https://github.com/salute-developers/golos/blob/master/license/ru.pdf
+# GOLOS OPUS Dataset
 
+## Dataset Description
+
+GOLOS is a Russian dataset for speech research. This is the OPUS version of the GOLOS dataset.
+
+## Subsets
+
+The dataset contains two subsets:
+- **crowd**: Crowd-sourced recordings
+- **farfield**: Far-field recordings
+
+## Dataset Structure
+
+This dataset is organized in the Common Voice format:
+
+- `/audio/{subset}/{split}/` - Contains TAR files with audio files
+- `/transcript/{subset}/` - Contains TSV files with transcriptions
+- `/n_shards.json` - Contains information about the number of shards for each subset and split
+
+## Usage
+
+```python
+from datasets import load_dataset
+from torch.utils.data import DataLoader
+
+# Загрузка crowd подмножества, тренировочного сплита
+dataset = load_dataset("Sh1man/golos_opus", "crowd", split="train")
+
+# Или использование всего датасета
+dataset = load_dataset("Sh1man/golos_opus", "crowd")
+train_dataset = dataset["train"]
+test_dataset = dataset["test"]
+
+# Создание DataLoader
+dataloader = DataLoader(dataset, batch_size=32)
+
+# Доступ к данным
+for batch in dataloader:
+    print(batch["audio"])
+    print(batch["text"])
 ```
-@inproceedings{karpov21_interspeech,
-  author={Nikolay Karpov and Alexander Denisenko and Fedor Minkin},
-  title={{Golos: Russian Dataset for Speech Research}},
-  year=2021,
-  booktitle={Proc. Interspeech 2021},
-  pages={1419--1423},
-  doi={10.21437/Interspeech.2021-462}
-}/github.com/salute-developers/golos/blob/master/license/en_us.pdf
 
-Russian Version
-https://github.com/salute-developers/golos/blob/master/license/ru.pdf
+## Dataset Statistics
+
+
+### crowd subset
+
+| Split | Samples | Hours |
+|-------|---------|-------|
+| train | 979796 | 1094.99 |
+| test | 9994 | 11.20 |
+| **Total** | **989790** | **1106.19** |
+
+### farfield subset
+
+| Split | Samples | Hours |
+|-------|---------|-------|
+| train | 124003 | 132.46 |
+| test | 1916 | 1.41 |
+| **Total** | **125919** | **133.87** |
+
+## License
+
+English Version: https://github.com/salute-developers/golos/blob/master/license/en_us.pdf
+Russian Version: https://github.com/salute-developers/golos/blob/master/license/ru.pdf
+
+## Citation
 
 ```
 @inproceedings{karpov21_interspeech,
@@ -227,4 +75,5 @@ https://github.com/salute-developers/golos/blob/master/license/ru.pdf
   booktitle={Proc. Interspeech 2021},
   pages={1419--1423},
   doi={10.21437/Interspeech.2021-462}
-}
+}
+```

dataset_infos/crowd.json

@@ -0,0 +1,72 @@
+{
+  "crowd": {
+    "description": "GOLOS OPUS Russian speech dataset - crowd subset",
+    "citation": "@inproceedings{karpov21_interspeech,\n  author={Nikolay Karpov and Alexander Denisenko and Fedor Minkin},\n  title={{Golos: Russian Dataset for Speech Research}},\n  year=2021,\n  booktitle={Proc. Interspeech 2021},\n  pages={1419--1423},\n  doi={10.21437/Interspeech.2021-462}\n}",
+    "homepage": "https://github.com/salute-developers/golos",
+    "license": "CC-BY 4.0",
+    "features": {
+      "id": {
+        "dtype": "string",
+        "id": null,
+        "_type": "Value"
+      },
+      "path": {
+        "dtype": "string",
+        "id": null,
+        "_type": "Value"
+      },
+      "text": {
+        "dtype": "string",
+        "id": null,
+        "_type": "Value"
+      },
+      "duration": {
+        "dtype": "float32",
+        "id": null,
+        "_type": "Value"
+      },
+      "audio": {
+        "sampling_rate": 16000,
+        "id": null,
+        "_type": "Audio"
+      }
+    },
+    "post_processed": null,
+    "supervised_keys": null,
+    "task_templates": [
+      {
+        "task": "automatic-speech-recognition",
+        "audio_column": "audio",
+        "transcription_column": "text"
+      }
+    ],
+    "builder_name": "golos_opus",
+    "config_name": "crowd",
+    "version": {
+      "version_str": "1.0.0",
+      "description": null,
+      "major": 1,
+      "minor": 0,
+      "patch": 0
+    },
+    "splits": {
+      "train": {
+        "name": "train",
+        "num_bytes": null,
+        "num_examples": 979796,
+        "dataset_name": "golos_opus.crowd"
+      },
+      "test": {
+        "name": "test",
+        "num_bytes": null,
+        "num_examples": 9994,
+        "dataset_name": "golos_opus.crowd"
+      }
+    },
+    "download_checksums": {},
+    "download_size": null,
+    "post_processing_size": null,
+    "dataset_size": null,
+    "size_in_bytes": null
+  }
+}

dataset_infos/farfield.json

@@ -0,0 +1,72 @@
+{
+  "farfield": {
+    "description": "GOLOS OPUS Russian speech dataset - farfield subset",
+    "citation": "@inproceedings{karpov21_interspeech,\n  author={Nikolay Karpov and Alexander Denisenko and Fedor Minkin},\n  title={{Golos: Russian Dataset for Speech Research}},\n  year=2021,\n  booktitle={Proc. Interspeech 2021},\n  pages={1419--1423},\n  doi={10.21437/Interspeech.2021-462}\n}",
+    "homepage": "https://github.com/salute-developers/golos",
+    "license": "CC-BY 4.0",
+    "features": {
+      "id": {
+        "dtype": "string",
+        "id": null,
+        "_type": "Value"
+      },
+      "path": {
+        "dtype": "string",
+        "id": null,
+        "_type": "Value"
+      },
+      "text": {
+        "dtype": "string",
+        "id": null,
+        "_type": "Value"
+      },
+      "duration": {
+        "dtype": "float32",
+        "id": null,
+        "_type": "Value"
+      },
+      "audio": {
+        "sampling_rate": 16000,
+        "id": null,
+        "_type": "Audio"
+      }
+    },
+    "post_processed": null,
+    "supervised_keys": null,
+    "task_templates": [
+      {
+        "task": "automatic-speech-recognition",
+        "audio_column": "audio",
+        "transcription_column": "text"
+      }
+    ],
+    "builder_name": "golos_opus",
+    "config_name": "farfield",
+    "version": {
+      "version_str": "1.0.0",
+      "description": null,
+      "major": 1,
+      "minor": 0,
+      "patch": 0
+    },
+    "splits": {
+      "train": {
+        "name": "train",
+        "num_bytes": null,
+        "num_examples": 124003,
+        "dataset_name": "golos_opus.farfield"
+      },
+      "test": {
+        "name": "test",
+        "num_bytes": null,
+        "num_examples": 1916,
+        "dataset_name": "golos_opus.farfield"
+      }
+    },
+    "download_checksums": {},
+    "download_size": null,
+    "post_processing_size": null,
+    "dataset_size": null,
+    "size_in_bytes": null
+  }
+}

golos_opus.py

@@ -0,0 +1,169 @@
+# coding=utf-8
+# Copyright 2023 The HuggingFace Datasets Authors and the GOLOS OPUS dataset creators.
+#
+# Licensed under the Apache License, Version 2.0 (the "License");
+# you may not use this file except in compliance with the License.
+# You may obtain a copy of the License at
+#
+#     http://www.apache.org/licenses/LICENSE-2.0
+#
+# Unless required by applicable law or agreed to in writing, software
+# distributed under the License is distributed on an "AS IS" BASIS,
+# WITHOUT WARRANTIES OR CONDITIONS OF ANY KIND, either express or implied.
+# See the License for the specific language governing permissions and
+# limitations under the License.
+
+import csv
+import json
+import os
+from pathlib import Path
+from typing import Dict, List, Tuple
+
+import datasets
+
+logger = datasets.logging.get_logger(__name__)
+
+_DESCRIPTION = '''
+GOLOS is a Russian dataset for speech research.
+This is the OPUS version of the GOLOS dataset.
+'''
+
+_CITATION = '''
+@inproceedings{karpov21_interspeech,
+  author={Nikolay Karpov and Alexander Denisenko and Fedor Minkin},
+  title={{Golos: Russian Dataset for Speech Research}},
+  year=2021,
+  booktitle={Proc. Interspeech 2021},
+  pages={1419--1423},
+  doi={10.21437/Interspeech.2021-462}
+}
+'''
+
+_HOMEPAGE = "https://github.com/salute-developers/golos"
+_LICENSE = "CC-BY 4.0"
+
+_SUBSETS = ["crowd", "farfield"]
+
+
+class GolosOpusConfig(datasets.BuilderConfig):
+    """BuilderConfig for GolosOpus."""
+
+    def __init__(self, name, **kwargs):
+        """BuilderConfig for GolosOpus.
+        Args:
+          name: Subset of the dataset: crowd or farfield
+          **kwargs: keyword arguments forwarded to super.
+        """
+        super(GolosOpusConfig, self).__init__(name=name, **kwargs)
+
+
+class GolosOpus(datasets.GeneratorBasedBuilder):
+    """GOLOS: Russian Dataset for Speech Research."""
+
+    VERSION = datasets.Version("1.0.0")
+    BUILDER_CONFIGS = [
+        GolosOpusConfig(name=subset, description=f"GOLOS OPUS Russian speech dataset - {subset} subset")
+        for subset in _SUBSETS
+    ]
+
+    def _info(self):
+        features = datasets.Features(
+            {
+                "id": datasets.Value("string"),
+                "path": datasets.Value("string"),
+                "text": datasets.Value("string"),
+                "duration": datasets.Value("float32"),
+                "audio": datasets.Audio(sampling_rate=16000),
+            }
+        )
+
+        return datasets.DatasetInfo(
+            description=_DESCRIPTION,
+            features=features,
+            supervised_keys=None,
+            homepage=_HOMEPAGE,
+            license=_LICENSE,
+            citation=_CITATION,
+            task_templates=[
+                datasets.AutomaticSpeechRecognition(audio_column="audio", transcription_column="text")
+            ],
+        )
+
+    def _split_generators(self, dl_manager):
+        """Returns SplitGenerators."""
+        subset = self.config.name
+        return [
+            datasets.SplitGenerator(
+                name=datasets.Split.TRAIN,
+                gen_kwargs={
+                    "filepath": f"transcript/{subset}/train.tsv",
+                    "split": "train",
+                },
+            ),
+            datasets.SplitGenerator(
+                name=datasets.Split.TEST,
+                gen_kwargs={
+                    "filepath": f"transcript/{subset}/test.tsv",
+                    "split": "test",
+                },
+            ),
+        ]
+
+    def _generate_examples(self, filepath, split):
+        """Yields examples."""
+        data_dir = Path(self.config.data_dir)
+        subset = self.config.name
+
+        # Загружаем информацию о количестве шардов
+        with open(data_dir / "n_shards.json", encoding="utf-8") as f:
+            n_shards = json.load(f)
+
+        num_shards = n_shards[subset][split]
+
+        # Создаем список имен шардов
+        audio_shards = [
+            data_dir / "audio" / subset / split / f"{split}_{i:05d}.tar"
+            for i in range(num_shards)
+        ]
+
+        with open(data_dir / filepath, encoding="utf-8") as f:
+            reader = csv.DictReader(f, delimiter="\t")
+            for row_id, row in enumerate(reader):
+                # Ищем файл в шардах
+                sample_id = row["id"]
+                file_path = row["path"]
+
+                # Поиск шарда, в котором находится файл
+                import tarfile
+                from io import BytesIO
+
+                # В Common Voice формате нужно искать файл в каждом шарде
+                audio_data = None
+
+                for shard_path in audio_shards:
+                    try:
+                        with tarfile.open(shard_path, "r") as tar:
+                            try:
+                                member = tar.getmember(file_path)
+                                f = tar.extractfile(member)
+                                if f:
+                                    audio_data = {"path": str(shard_path), "bytes": file_path}
+                                    break
+                            except KeyError:
+                                continue
+                    except Exception as e:
+                        logger.warning(f"Error opening {shard_path}: {e}")
+
+                if audio_data is None:
+                    logger.warning(f"Could not find {file_path} in any shard, skipping")
+                    continue
+
+                example = {
+                    "id": row["id"],
+                    "path": row["path"],
+                    "text": row["text"],
+                    "duration": float(row["duration"]) if row["duration"] else 0.0,
+                    "audio": audio_data,
+                }
+
+                yield row_id, example

n_shards.json

@@ -0,0 +1,10 @@
+{
+  "crowd": {
+    "train": 16,
+    "test": 1
+  },
+  "farfield": {
+    "train": 2,
+    "test": 1
+  }
+}