Transcribe audio files to text using local offline speech recognition models.
npx skills add https://github.com/joeseesun/qiaomu-cut-skill --skill asr-joeseesun
Please help me install this Agent Skill. Skill: asr Source: https://github.com/joeseesun/qiaomu-cut-skill/tree/main/vendor/marswaveai-skills/asr Command: npx skills add https://github.com/joeseesun/qiaomu-cut-skill --skill asr-joeseesun
coli asr with the sensevoice model supporting Chinese, English, Japanese, Korean, and Cantonese, or whisper-tiny.en for English-only audio. - Rich Metadata Output: Returns detected language, emotion, audio events, and duration alongside the transcript text. - Optional AI Polish: Cleans up punctuation, removes filler words, and improves readability while preserving the original meaning, with an option to export the result as a Markdown file. - Use Case: You have a recorded meeting as meeting.m4a and need a clean, readable transcript. The Skill checks prerequisites, confirms settings, transcribes locally, polishes the text, and optionally saves it as a Markdown file. ## Quick Start Ask the assistant to transcribe your audio file, for example: transcribe the file meeting.m4a into text with polish enabled.High-intent search queries and answers about installing and using this skill.
Use the coli CLI with the command coli asr -j --model sensevoice followed by your file path. It runs local speech recognition models entirely offline, so no API key or internet upload is required after the initial model download.
The sensevoice model supports Chinese, English, Japanese, Korean, and Cantonese, and can also detect language, emotion, and audio events. The whisper-tiny.en model is available as an English-only alternative.
No API key is needed because transcription runs fully offline through local models. Internet access is only required once to download the roughly 60MB model files to ~/.coli/models on first use.
ffmpeg is required to decode compressed audio formats like m4a or mp3 before recognition. WAV files work without ffmpeg, but installing it via brew or apt ensures broader format compatibility.
sensevoice supports five languages and detects emotion and audio events, making it the recommended default. whisper-tiny.en only handles English and lacks the extra metadata detection capabilities.
silversteinSearches recent voice memos and local minutes for short-term recall.
SaberAlouiRepurpose podcast episodes into social posts, show notes, quote cards, and video clips.
J-StaR-Films-StudiosAutomate podcast production with text-to-speech, music generation, and audio editing.
framerslabIdentify and label individual speakers in multi-speaker audio streams.