Marek324/speech-music-classification

--- pretty_name: Speech / Music classification license: unknown language: en tags: - audio - parquet configs: - config_name: full data_files: - split: train path: full/train/**/*.parquet - split: validation path: full/validation/**/*.parquet - split: test path: full/test/**/*.parquet - config_name: mid data_files: - split: train path: mid/train/**/*.parquet - split: validation path: mid/validation/**/*.parquet - split: test path: mid/test/**/*.parquet - config_name: mini data_files: - split: train path: mini/train/**/*.parquet - split: validation path: mini/validation/**/*.parquet - split: test path: mini/test/**/*.parquet --- ## Structure After `uv run python build.py {mini|mid|full}` for each tier (same `--out-dir` staging parent): ```text <staging>/ full/ # or mid/, mini/ train/ speech/ part_*.parquet music/ inactive/ validation/ speech/ ... test/ ... ``` Parquet columns: `audio` (HF Audio struct), `sampling_rate`, `class`, `subclass`, `source`, `row_idx`, `labels` (list of `{label, start, end}` in ms). ## Upload From the **staging parent** (folder that contains `mini/`, `mid/`, `full/`): ```bash hf upload-large-folder --repo-type dataset YOUR_ORG/YOUR_REPO . ``` ## Load ```python from datasets import load_dataset ds = load_dataset("YOUR_ORG/YOUR_REPO", "full", split="train") # Rows include columns from parquet; decode WAV from bytes if needed. ``` Adjust `YOUR_ORG/YOUR_REPO` and verify YAML against current [Hub dataset data files](https://huggingface.co/docs/hub/datasets-data-files-configuration) if `load_dataset` errors.pretty_name: 语音/音乐分类（Speech / Music Classification） license: 未知 language: 英语 tags: - 音频（audio） - Parquet（parquet） configs: - config_name: 完整集（full） data_files: - split: 训练集（train） path: full/train/**/*.parquet - split: 验证集（validation） path: full/validation/**/*.parquet - split: 测试集（test） path: full/test/**/*.parquet - config_name: 中等集（mid） data_files: - split: 训练集（train） path: mid/train/**/*.parquet - split: 验证集（validation） path: mid/validation/**/*.parquet - split: 测试集（test） path: mid/test/**/*.parquet - config_name: 小型集（mini） data_files: - split: 训练集（train） path: mini/train/**/*.parquet - split: 验证集（validation） path: mini/validation/**/*.parquet - split: 测试集（test） path: mini/test/**/*.parquet ## 目录结构 针对每个数据集层级（需使用相同的`--out-dir`参数指定暂存父目录）执行`uv run python build.py {mini|mid|full}`命令后，将生成如下目录结构： text <staging>/ full/ # 或 mid/, mini/ train/ speech/ part_*.parquet music/ inactive/ validation/ speech/ ... test/ ... Parquet文件的列字段包括：`audio`（HF Audio 结构体（HF Audio struct））、`sampling_rate`（采样率）、`class`（类别）、`subclass`（子类别）、`source`（数据源）、`row_idx`（行索引）、`labels`（由`{label, start, end}`组成的列表，单位为毫秒）。 ## 上传方式 在包含`mini/`、`mid/`、`full/`的暂存父目录下执行如下命令： bash hf upload-large-folder --repo-type dataset YOUR_ORG/YOUR_REPO . ## 加载示例 python from datasets import load_dataset ds = load_dataset("YOUR_ORG/YOUR_REPO", "full", split="train") # 数据行包含Parquet文件的全部列字段；若需从字节数据中解码WAV音频，请按需自行处理。 若`load_dataset`函数调用报错，请调整`YOUR_ORG/YOUR_REPO`参数，并根据当前[Hub数据集数据文件配置文档](https://huggingface.co/docs/hub/datasets-data-files-configuration)验证YAML配置文件。