dsh-voice-input: Voice Dictation and Transcription Cleaning Plugin for DeepSeek Harness

# Introduction In DeepSeek Harness, the content entered into the composer is often colloquial: supplementing context, describing requirements, and stating constraints. Speaking is faster than typing, but the resulting transcription is raw—it contains filler words, repetitions, and mid-sentence corrections. Manual cleanup is required before pasting it into the input box. The approach of **dsh-voice-input** is to separate these two steps: use browser or system dictation for recording, and delegate text cleaning to the model currently selected for the current session, without the need to configure an additional API key. The following sections introduce the plugin's positioning, installation methods, and daily usage. # What is this? **dsh-voice-input** is an open-source plugin maintained by lhenlihai-hub.