pauljunsukhan/throatmic_codered

Throat Microphone Dataset是一个高质量的喉部麦克风录音数据集，专门设计用于微调Whisper和其他语音识别模型。该数据集包含通过喉部麦克风录制的英语句子，这些录音通过喉咙的振动而非空气传导的声音来捕捉语音。数据集适用于自动语音识别（ASR）、语音转文本、语音活动检测和喉部麦克风适应等任务。数据集包含音频、文本转录和音频时长三个字段，且仅包含英语录音。数据集由Paul Han等人创建，遵循MIT许可证。

The Throat Microphone Dataset is a high-quality dataset of throat microphone (laryngophone) recordings specifically designed for fine-tuning Whisper and other speech recognition models. The dataset consists of carefully selected English sentences recorded using a throat microphone, which captures speech through vibrations in the throat rather than air-conducted sound. The dataset is suitable for Automatic Speech Recognition (ASR), Speech-to-Text, Voice Activity Detection, and Throat Microphone Adaptation. Each instance in the dataset contains audio, text, and duration fields. The dataset is provided as a single training split. The creation of this dataset addresses the lack of high-quality throat microphone data, making it useful for high-noise environments, military and emergency services communications, industrial settings, and assistive technology. The dataset may have biases related to language, pronunciation, accents, and the number of speakers.