renumics/speech_commands_enrichment_only

SpeechCommands数据集包含一秒长的.wav音频文件，每个文件包含一个单独的英语单词或背景噪音。这些单词来自一组小的命令，由不同的说话者录制。数据集旨在帮助训练简单的机器学习模型，特别是用于关键词检测任务。数据集有两个版本，v0.01和v0.02，分别包含64,727和105,829个音频文件。数据集还提供了两种配置：仅包含增强信息的enrichment_only和包含原始数据与增强信息结合的raw_and_enrichment_combined。

The SpeechCommands dataset consists of 1-second-long .wav audio files, each containing a single English word or background noise. These words are sourced from a small set of commands, recorded by various speakers. The dataset is designed to facilitate the training of simple machine learning models, particularly for keyword detection tasks. There are two versions of the dataset, v0.01 and v0.02, which contain 64,727 and 105,829 audio files respectively. The dataset also provides two configurations: enrichment_only, which only includes augmented information, and raw_and_enrichment_combined, which combines raw data with augmented information.

原始信息汇总

数据集概述

基本信息

• 数据集名称: SpeechCommands
• 语言: 英语 (en)
• 许可证: CC-BY-4.0
• 多语言性: 单语种
• 大小类别: 10K<n<100K, 100K<n<1M
• 源数据集: 扩展自 speech_commands
• 任务类别: 音频分类
• 任务ID: 关键词识别
• 配置名称: v0.01, v0.02
• 标签: spotlight, enriched, renumics, enhanced, audio, classification, extended

数据集结构

配置详情

配置: enrichment_only

• 特征:
  • label_string: 字符串
  • probability: 浮点数 (float64)
  • probability_vector: 浮点数序列 (float32)
  • prediction: 整数 (int64)
  • prediction_string: 字符串
  • embedding_reduced: 浮点数序列 (float32)
• 分割:
  • train: 8763867 字节, 51093 样本
  • validation: 1165942 字节, 6799 样本
  • test: 528408 字节, 3081 样本
• 下载大小: 0 字节
• 数据集大小: 10458217 字节

配置: raw_and_enrichment_combined

• 特征:
  • file: 字符串
  • audio: 音频 (采样率: 16000)
  • label: 类别标签 (名称: 0-30)
  • is_unknown: 布尔值
  • speaker_id: 字符串
  • utterance_id: 整数 (int8)
  • logits: 浮点数序列 (float64)
  • embedding: 浮点数序列 (float32)
  • label_string: 字符串
  • probability: 浮点数 (float64)
  • probability_vector: 浮点数序列 (float32)
  • prediction: 整数 (int64)
  • prediction_string: 字符串
  • embedding_reduced: 浮点数序列 (float32)
• 分割:
  • train: 1803565876.375 字节, 51093 样本
  • validation: 240795605.125 字节, 6799 样本
  • test: 109673146.875 字节, 3081 样本
• 下载大小: 0 字节
• 数据集大小: 2154034628.375 字节

数据文件

• 配置: enrichment_only
  • train: enrichment_only/train-*
  • validation: enrichment_only/validation-*
  • test: enrichment_only/test-*
• 配置: raw_and_enrichment_combined
  • train: raw_and_enrichment_combined/train-*
  • validation: raw_and_enrichment_combined/validation-*
  • test: raw_and_enrichment_combined/test-*

数据实例

核心词示例

python { "file": "no/7846fd85_nohash_0.wav", "audio": { "path": "no/7846fd85_nohash_0.wav", "array": array([ -0.00021362, -0.00027466, -0.00036621, ..., 0.00079346, 0.00091553, 0.00079346]), "sampling_rate": 16000 }, "label": 1, # "no" "is_unknown": False, "speaker_id": "7846fd85", "utterance_id": 0 }

辅助词示例

python { "file": "tree/8b775397_nohash_0.wav", "audio": { "path": "tree/8b775397_nohash_0.wav", "array": array([ -0.00854492, -0.01339722, -0.02026367, ..., 0.00274658, 0.00335693, 0.0005188]), "sampling_rate": 16000 }, "label": 28, # "tree" "is_unknown": True, "speaker_id": "1b88bf70", "utterance_id": 0 }

背景噪声示例

python { "file": "silence/doing_the_dishes.wav", "audio": { "path": "silence/doing_the_dishes.wav", "array": array([ 0. , 0. , 0. , ..., -0.00592041, -0.00405884, -0.00253296]), "sampling_rate": 16000 }, "label": 30, # "silence" "is_unknown": False, "speaker_id": "None", "utterance_id": 0 # doesnt make sense here }

数据字段

• file: 音频文件的相对路径
• audio: 包含音频文件路径、解码后的音频数组和采样率
• label: 音频样本中的单词或背景噪声类别
• is_unknown: 单词是否为辅助词
• speaker_id: 说话者的唯一ID
• utterance_id: 同一说话者内的单词发音增量ID

数据分割

• v0.01:
  • train: 51093 样本
  • validation: 6799 样本
  • test: 3081 样本
• v0.02:
  • train: 84848 样本
  • validation: 9982 样本
  • test: 4890 样本