Menlo/instruction-speech-whispervq-v1

--- dataset_info: features: - name: index dtype: int64 - name: tokens sequence: int64 - name: text_prompt dtype: string - name: answer dtype: string - name: length dtype: int64 - name: prompt dtype: string - name: conversations list: - name: content dtype: string - name: role dtype: string splits: - name: train num_bytes: 6872437718 num_examples: 679149 download_size: 2032992881 dataset_size: 6872437718 configs: - config_name: default data_files: - split: train path: data/train-* --- ## Dataset Overview This dataset contains nearly over 679,000 English `speech instruction to text answer` samples, using: - The combination of [homebrewltd/instruction-speech-encodec-v1](https://huggingface.co/datasets/homebrewltd/instruction-speech-encodec-v1) and [homebrewltd/instruction-speech-encodec-v1.5](https://huggingface.co/datasets/homebrewltd/instruction-speech-encodec-v1.5) - Tokenized using [WhisperVQ](https://huggingface.co/WhisperSpeech/WhisperSpeech/blob/main/whisper-vq-stoks-medium-en%2Bpl.model). ## Usage ```python from datasets import load_dataset, Audio # Load Instruction Speech dataset dataset = load_dataset("homebrewltd/raw-speech-whispervq-v1",split='train') ``` ## Dataset Fields Field | Type | Description | |------------------|------------|--------------------------------------------------| | `tokens` | sequence | Tokenized using WhisperVQ | | `text_prompt` | string | User's query | | `answer` | string | Assistant's answer | | `length` | int | Length of user's query | | `prompt` | string | User's query tokenized using WhisperVQ | | `conversations` | sequence | Conversation between User and Assistant | ## Bias, Risks, and Limitations - Dataset may reflect biases inherent in its source. - Current version lacks quality control for prompts and responses. - The usage of Encodec may compromise sound tokens quality. - Users should consider these limitations when applying the dataset. ## Licensing Information The dataset is released under the [MIT license](https://opensource.org/license/MIT). ## Citation Information ``` @article{Instruction Speech 2024, title={Instruction Speech}, author={JanAI}, year=2024, month=June}, url={https://huggingface.co/datasets/jan-hq/instruction-speech} ```数据集信息： 特征： - 名称：index，数据类型：int64 - 名称：tokens，数据类型：int64序列 - 名称：text_prompt，数据类型：字符串 - 名称：answer，数据类型：字符串 - 名称：length，数据类型：int64 - 名称：prompt，数据类型：字符串 - 名称：conversations，数据类型：列表，其子项为： - 名称：content，数据类型：字符串 - 名称：role，数据类型：字符串 数据划分： - 划分名称：train，字节数：6872437718，样本数量：679149 下载大小：2032992881 数据集总大小：6872437718 配置项： - 配置名称：default，数据文件： - 划分：train，路径：data/train-* ## 数据集概览 本数据集包含近67.9万条英文「语音指令转文本回复」样本，构建时结合了[homebrewltd/instruction-speech-encodec-v1](https://huggingface.co/datasets/homebrewltd/instruction-speech-encodec-v1)与[homebrewltd/instruction-speech-encodec-v1.5](https://huggingface.co/datasets/homebrewltd/instruction-speech-encodec-v1.5)两个数据集，并使用[WhisperVQ](https://huggingface.co/WhisperSpeech/WhisperSpeech/blob/main/whisper-vq-stoks-medium-en%2Bpl.model)完成了Token分词处理。 ## 使用方法 python from datasets import load_dataset, Audio # 加载语音指令数据集 dataset = load_dataset("homebrewltd/raw-speech-whispervq-v1", split='train') ## 数据集字段 | 字段名 | 数据类型 | 描述 | |------------------|------------|--------------------------------------------------| | `tokens` | 序列 | 使用WhisperVQ完成Token分词后的结果 | | `text_prompt` | 字符串 | 用户的查询请求 | | `answer` | 字符串 | 助手的回复内容 | | `length` | 整数 | 用户查询请求的长度 | | `prompt` | 字符串 | 使用WhisperVQ完成Token分词后的用户查询请求 | | `conversations` | 序列 | 用户与助手之间的对话记录 | ## 偏见、风险与局限性 - 本数据集可能反映其源数据中固有的偏见。 - 当前版本未对查询请求与回复内容进行质量管控。 - Encodec的使用可能会损害语音Token的质量。 - 用户在使用该数据集时需充分考虑上述局限性。 ## 许可证信息 本数据集采用[MIT许可证](https://opensource.org/license/MIT)进行发布。 ## 引用信息 @article{Instruction Speech 2024, title={Instruction Speech}, author={JanAI}, year=2024, month=June}, url={https://huggingface.co/datasets/jan-hq/instruction-speech} }