Vendor page
Tidytalk: Private, On-Device Voice Input for Mac
byteyu2436, Tidytalk (tidytalk.zayen.dev), 2026
Read the originalRead on October 23, 2026
- What it is
- The official site of Tidytalk, a free, open-source Mac dictation app that recognizes speech on the Mac and can tidy the text with a local or cloud AI model. It describes offline use, file transcription, voice editing of selected text and the privacy setup.
- How much weight it carries
- The maker's own site. It links the byteyu2436/tidytalk GitHub repository, which is the source of the byteyu2436 attribution. The claims are the maker's and were not tested.
Facts taken from this source
- Tidytalk's site says the Qwen3-ASR model recognizes names and terms from your vocabulary.
“Qwen3-ASR recognizes names and terms from your vocabulary” (Speech recognition models)
- Tidytalk's site says Tidytalk is free and open source under the MIT License.
“Yes. Tidytalk is free and open source under the MIT License.” (FAQ, Is Tidytalk free)
- Tidytalk's site says recordings are never uploaded.
“The SenseVoice model runs on your Mac. Recordings are never uploaded.” (How it works, Speech recognition)
- Tidytalk's site says that when you transcribe a YouTube or Bilibili link, only the audio is downloaded and then deleted. The site gives no time for the deletion.
“For links, only the audio is downloaded, then deleted” (Transcribe audio, video files, and links too)
- Tidytalk's site says its Qwen3-ASR speech model covers 30 languages, and gives 22 Chinese dialects separately.
“For other languages, switch to Qwen3-ASR (30 languages and 22 Chinese dialects)” (FAQ, Which languages does it support?)
- Tidytalk's site says your voice is never uploaded and that recording and recognition happen on your Mac.
“Your voice never is. Recording and speech recognition happen on your Mac.” (FAQ, Is my voice or text uploaded)
- Tidytalk's site says its recent results live in memory and are cleared when you quit the app. The site gives no number of days.
“Recent results live in memory and are cleared when you quit.” (Everything else: Nothing collected)
- Tidytalk's site says its AI tidying removes fillers and slips without summarizing or adding.
“Removes fillers and slips without summarizing or adding.” (Tidies, never invents)
- Tidytalk's site says that with a cloud AI service, only the recognized text is sent to that service.
“With a cloud service, only the recognized text is sent.” (FAQ, Is my voice or text uploaded)
- Tidytalk's site says its default SenseVoice model supports five languages: Mandarin Chinese (including mixed-in English), English, Japanese, Korean and Cantonese.
“With the default SenseVoice model it detects the spoken language automatically and supports Mandarin Chinese (including mixed-in English), English, Japanese, Korean, and Cantonese.” (FAQ, Which languages does it support?)
- Tidytalk's site says the SenseVoice speech model runs on the Mac and recordings are never uploaded.
“The SenseVoice model runs on your Mac. Recordings are never uploaded.” (How it works, Speech recognition)
- Tidytalk's site says you hold a shortcut in any text field and clean text appears at your cursor.
“Option + Command, in any text field. ... Clean text appears right at your cursor.” (How it works, Three steps)
- Tidytalk's site says you can select text, hold Right Option and say what to change.
“Select text, hold Right Option, and say what to change.” (Everything else, Rewrite with your voice)
- Tidytalk's site says its Whisper Turbo speech model covers about 99 languages.
“switch to Qwen3-ASR (30 languages and 22 Chinese dialects) or Whisper Turbo (about 99)” (FAQ, Which languages does it support?)
- Tidytalk's site says that with a local model the app works without an internet connection.
“With a local model, Tidytalk works without an internet connection.” (FAQ, Is there a free, open-source alternative)