issai/SpokenMQA_Kazakh

该数据集是SpokenMQA基准测试的哈萨克语机器翻译版本，用于评估基于语音和音频语言模型的数学推理能力。数据集包含四个难度递增的子集：`short_digit`（短数字序列识别）、`long_digit`（长数字序列识别）、`single_step_reasoning`（需要单步算术操作的数学问题）和`multi_step_reasoning`（需要多步推理的数学问题）。原始基准测试的文本字段被机器翻译成哈萨克语，音频则使用MangiSozTTS从翻译后的文本重新合成。

This dataset is a machine-translated Kazakh adaptation of the SpokenMQA benchmark for evaluating spoken mathematical reasoning in speech-based and audio-language models. The benchmark covers four tracks of increasing difficulty: `short_digit` (recognition of short numerical sequences spoken aloud), `long_digit` (recognition of longer numerical sequences spoken aloud), `single_step_reasoning` (mathematical problems requiring a single arithmetic operation), and `multi_step_reasoning` (mathematical problems requiring multi-step reasoning over spoken input). The textual fields of the original benchmark were machine-translated into Kazakh, and the audio was re-synthesized in Kazakh from the translated transcripts using MangiSozTTS.