Every spoken word, instantly searchable
Generate highly accurate transcripts for any file in your library — podcast, audiobook, voice memo — processed locally on your device's Neural Engine. No internet. No data sent anywhere.
Free on supported devices. Private by architecture.
On-device vs cloud transcription
Cloud-based transcription services — including those offered by Snipd and similar apps — send your audio to a remote server for processing. That server receives the content of what you were listening to, when, and generates a permanent record of the result. For public podcast episodes, this may be acceptable. For private recordings, confidential interviews, proprietary lectures, or any audio you would not share publicly, it is a meaningful privacy exposure.
Nuotit's transcription runs entirely on the Neural Engine inside your iPhone or iPad. The audio file never leaves your device. The transcript it produces is stored locally. No server receives either the audio or the text. This is not a policy — it is an architectural fact about how the feature is built.
- Processes any file type — RSS downloads or imported local files
- Audio file never transmitted to any server
- Transcript stored on-device in the app's local container
- Works fully offline once the episode is downloaded
- Runs in the background — processing continues seamlessly
- Free on supported devices (iPhone 15 Pro+ / M1 iPad+)
Device requirements
Why cloud transcripts go out of sync — and why Nuotit's don't
If you have ever noticed a podcast transcript that seems to lag behind the audio by a few seconds, or has timestamps that jump unexpectedly, you have encountered the dynamic ad insertion desynchronization problem.
Major podcast hosting platforms — Spotify, Acast, Megaphone, and others — use server-side dynamic ad insertion. When you download an episode, the hosting server inserts localized, real-time advertisements into the audio file at that moment. The inserted ads alter the total duration of the file. A transcript generated by a cloud service from the original pre-insertion audio now has timestamps mapped to a different version than the one on your device — by exactly the combined duration of the injected ads.
Nuotit generates its transcript locally from the exact audio file stored on your device — the version with all dynamic ads already inserted. The timestamp mapping is always accurate because the transcript is generated from the same file that plays back.
How the desync happens
Cloud service transcript → mapped to original audio (pre-ad-insertion, 42:00)
Your downloaded file → 44:30 (with two 45-second ads inserted dynamically)
Result → every timestamp after the first ad break is offset by 45 seconds or more.
Transcript features in detail
Synchronized read-along
The transcript scrolls automatically as audio plays, highlighting the current sentence. Tap any word or sentence and playback jumps to that exact moment. Works for both generated and feed-provided transcripts.
Keyword search
Search for any word or phrase within the transcript. Results highlight throughout the text. Jump to any match to hear the context. "Copy All" respects search results and copies only matching sections if a search is active.
Copy and export
Select any portion of a transcript to copy as plain text. Use "Copy All" for the full transcript. Paste into Notes, Obsidian, Notion, or any text editor — the text is plain and portable.
Feed-provided transcripts
When a podcast creator includes a transcript in the RSS feed (WebVTT or SRT format), Nuotit displays it immediately without any generation step. Tap-to-seek and search work identically.
Private transcription for a personal knowledge workflow
Apps like Snipd have built a strong following among knowledge workers who export podcast highlights to tools like Obsidian, Notion, and Readwise. The attraction is clear: audio becomes searchable, quotable text integrated into a personal note-taking system.
The limitation is equally clear: Snipd's AI processing runs on its cloud servers. Every highlight you send for AI summarization passes through Snipd's infrastructure. For knowledge workers using Obsidian in local-only mode precisely to avoid cloud exposure, this introduces an inconsistency.
Nuotit generates the transcript locally. The text is then available to copy directly into any local notes application. A workflow from audio to Obsidian vault that never touches a cloud server is entirely achievable: generate transcript on-device → copy selected passage → paste into local Obsidian note. No intermediary API. No cloud account required.
- Transcripts export as plain text — paste into any PKM tool
- No API intermediary required — direct local-to-local workflow
- Works for imported files — transcribe a private interview for your research notes
- Clip export generates captioned video highlights on-device for social sharing
Frequently asked questions
What devices support on-device transcript generation?
iPhone 15 Pro or later, or M1 iPad or later. All other Nuotit features work on any device running iOS 26 or iPadOS 26. Feed-provided transcripts (from the RSS feed) display on all devices.
Can Nuotit transcribe imported audiobooks and voice memos?
Yes — any file in the library. This distinguishes Nuotit from Apple Podcasts and Overcast, which only transcribe public RSS feed episodes.
How accurate are the transcripts?
Nuotit uses Apple's native on-device speech recognition. Accuracy is high for clear English speech. Heavy accents, poor recording quality, or multiple overlapping speakers reduce accuracy. Feed-provided transcripts are used when available and are typically better for non-English content.
Why are cloud podcast transcripts sometimes out of sync?
Dynamic server-side ad insertion alters the total audio duration after a transcript is generated, shifting all timestamps. Nuotit generates transcripts from the exact file on your device, so mapping is always accurate.
How long does generation take?
A one-hour episode typically takes a few minutes on iPhone 15 Pro or M1 iPad. The process runs in the background.
Your audio becomes knowledge
Every lecture, every interview, every voice memo — searchable, navigable, private.