Vikhrmodels/Ficbook-Audio-Instruct-10K

Ficbook Audio Instruct 10K是一个用于训练俄语音频语言模型的合成音频指令数据集。它包含约10K个样本，每个样本包含以下内容：
- **音频**：使用OpenAI的`gpt-4o-mini-tts`模型生成的Ficbook故事文本的音频
- **文本**：来自Ficbook故事的原始文本
- **问题**：为模型设计的指令/任务
- **答案**：由Gemini 2.5 Flash生成的预期响应
- **任务类型**：12个任务类别之一

数据集旨在训练和评估俄语虚构内容的音频语言模型。音频以MP3格式生成，采样率为16kHz，使用了11种不同的声音。任务类型包括分类、摘要、命名实体识别、JSON提取、JSON结构、JSON分析、翻译、问答、情感分析、关键词提取、改写和续写。

Ficbook Audio Instruct 10K is a synthetic audio instruction dataset for training Russian audio-language models. It contains ~10K samples of fiction text voiced with OpenAI TTS and paired with diverse instruction tasks. Each sample includes:
- **Audio**: Fiction text voiced using OpenAIs `gpt-4o-mini-tts` model
- **Text**: Original text from ficbook stories
- **Question**: Instruction/task for the model
- **Answer**: Expected response generated by Gemini 2.5 Flash
- **Task Type**: One of 12 task categories

The dataset was created for training and evaluating audio-language models on Russian fiction content. Audio is in MP3 format, mono, generated at variable rates using 11 different voices. Task types include classification, summarization, NER, JSON extraction, JSON structure, JSON analysis, translation, question answering, sentiment analysis, keywords extraction, paraphrase, and continuation.