摘要：当前最先进的皮层内脑信号转文本系统，通常将神经序列音素解码器与外部语言模型配对使用。两个设计维度仍未得到充分探索：选择性状态空间模型（Mamba）是否优于循环解码器，以及输出目标（语音/音素与字符）如何与这一模型选择相互作用。在公开的 Brain-to-Text '25 基准上，我们在一套可复现实验协议下，采用 CTC 目标函数训练并研究了一个受控的 2×2 组合：GRU 与混合 Mamba 解码器，以及音素与字符目标。循环基线仍然最强：最佳音素 GRU 达到 12.62% 的音素错误率（PER）和 21.19% 的词错误率（WER）；经语言模型重评分后，最佳文本 GRU 达到 13.39% 的字符错误率（CER）和 26.28% 的词错误率（WER）。混合 Mamba 模型具备竞争力，但未能超越该结果。消融实验分离了不同架构组件的贡献，错误分析则显示了依赖表征方式的失败模式：类似发音动作的音素混淆，以及词汇和词边界错误。

Abstract:State-of-the-art intracortical brain-to-text systems pair a neural-sequence phone decoder with an external language model. Two design axes remain underexplored: whether selective state-space models (Mamba) improve on recurrent decoders, and how the output target (phonetic vs.\ character) interacts with that choice. On the public Brain-to-Text '25 benchmark, we study a controlled 2x2 grid (GRU vs.\ hybrid Mamba decoder; phonetic vs.\ character targets) trained with a CTC objective under one reproducible protocol. The recurrent baseline remains strongest: the best phonetic GRU reaches 12.62% PER and 21.19% WER, while the best textual GRU after LM rescoring reaches 13.39% CER and 26.28% WER. The Mamba hybrid is competitive but does not surpass it. Ablations isolate architectural contributions, and error analysis shows representation-dependent failures: articulatory-like phoneme confusions vs.\ lexical and word-boundary errors.