VoiceAssistant-430K-vocalnet

# VoiceAssistant-430K-vocalnet

This dataset supports the reproduction of [VocalNet](https://github.com/SJTU-OmniAgent/VocalNet).

## Data Construction

1. **Data Source**: We used the [VoiceAssistant-400K](https://huggingface.co/datasets/gpt-omni/VoiceAssistant-400K) from Mini-Omni, which contains about 470K instances.
2. **Data Filtering**: We removed samples with excessively long data. The resulted corpus contains 430K instances.
3. **Response Speech**: We perform speech synthesis using [CosyVoice](https://github.com/FunAudioLLM/CosyVoice) to generate the response speech.
4. **Response Token**: We generate the speech token using [CosyVoice2](https://github.com/FunAudioLLM/CosyVoice).

## Acknowledgment

1. The original data is from [Mini-Omni](https://github.com/gpt-omni/mini-omni).
2. The generation of speech wave and token is from [CosyVoice/CosyVoice2](https://github.com/gpt-omni/mini-omni).

# VoiceAssistant-430K-vocalnet

本数据集可用于复现[VocalNet](https://github.com/SJTU-OmniAgent/VocalNet)。

## 数据集构建

1. **数据来源**：我们使用了Mini-Omni发布的[VoiceAssistant-400K](https://huggingface.co/datasets/gpt-omni/VoiceAssistant-400K)数据集，该数据集共包含约47万个样本实例。
2. **数据筛选**：我们移除了时长过长的样本，最终得到包含43万个样本实例的语料库。
3. **应答语音**：我们采用[CosyVoice](https://github.com/FunAudioLLM/CosyVoice)进行语音合成，以生成应答语音。
4. **应答Token(Token)**：我们通过[CosyVoice2](https://github.com/FunAudioLLM/CosyVoice)生成语音Token。

## 致谢

1. 原始数据集源自[Mini-Omni](https://github.com/gpt-omni/mini-omni)。
2. 语音波形与Token的生成工作基于[CosyVoice/CosyVoice2](https://github.com/gpt-omni/mini-omni)实现。

搜集汇总

背景与挑战

背景概述

该数据集基于VoiceAssistant-400K构建，通过过滤过长样本得到43万条实例，并利用CosyVoice合成响应语音，同时使用CosyVoice2生成语音标记。