Private AI transcription for Apple devices
Transcribe meetings, audio, and video.
Entirely on-device.
Record live or import an existing audio or video file. Echo Chamber turns it into an accurate, searchable transcript, useful notes, summaries, and answers without sending the recording to Obsidian Ridge Labs or inviting a bot into the call.
- Input
- Live · audio · video
- Transcription
- Parakeet TDT on-device
- Transcript AI
- Apple Intelligence + Bonsai
- Purchase
- Monthly · yearly · lifetime
The recorder, reconsidered
A useful transcript should not create a second audience.
Cloud recorders become useful only after the conversation leaves the room. Echo Chamber starts on the hardware already in your hand. Record in a meeting room, lecture hall, airplane, or anywhere the network is not invited, then keep the searchable archive with you.
Record with a live transcript and bookmark the moments that matter.
Read the raw transcript beside a cleaner, AI-polished version.
Ask focused questions and turn a long conversation into useful next steps.
Built for real conversations
Everything needed to turn speech into something useful.
Live transcription
See the transcript form as you speak. Echo Chamber detects pauses, speakers, and topic shifts without streaming your voice to a remote transcription service.
AI polish and notes
Keep the raw verbatim transcript, then turn it into a cleaner reading copy, useful notes, or a focused summary with Apple Intelligence or the local Bonsai fallback.
Searchable memory
Tag recordings, bookmark important moments, and search every word across your library without giving a cloud workspace your archive.
Audio and video upload
Already have the recording? Upload an audio or video file and turn it into the same searchable transcript, notes, summary, and exportable document.
Private by default
Audio is encrypted at rest, core processing happens locally, and the app contains no third-party SDKs that profile what you record.
Built for Apple Intelligence.
On compatible devices, Apple’s on-device Foundation Models power transcript polish, notes, summaries, and questions.
Bonsai 1.7B keeps the workflow local.
When Apple Intelligence is unavailable, Echo Chamber uses a bundled 1-bit Bonsai model for local transcript intelligence on supported Apple hardware.
Bonsai model cardParakeet TDT handles speech recognition on both paths. The original recording never needs a remote AI service.
Observed in Echo Chamber
Approximate word error rate observed for the enhanced Echo Chamber pipeline
Accuracy starts before Parakeet hears a word.
Before transcription, Echo Chamber applies a targeted speech enhancement pass built for recognizer input. This is not generic normalization or a cosmetic audio effect. The filter is part of the transcription pipeline because it improves the signal Parakeet receives. Across the current internal test set, that complete pipeline has observed approximately 4.5% WER.
For separate model context, the cited Hugging Face Open ASR snapshot reports 6.32% average English WER for Parakeet TDT 0.6B v3 and 7.44% for OpenAI Whisper large-v3. That public benchmark compares models. The 4.5% figure measures Echo Chamber’s enhanced product pipeline.
- Enhanced Echo Chamber pipeline
- About 4.5% observed WER
- Parakeet · public benchmark
- 6.32% WER
- OpenAI Whisper large-v3 · same benchmark
- 7.44% WER
The approximately 4.5% result is an internal product observation, not a guarantee or a model-only leaderboard score. Lower is better. WER varies with language, accent, background noise, microphones, overlapping speakers, and recording conditions.
Privacy, explained plainly
Core intelligence stays on-device.
Every connection is explained.
- Recording and transcription
- AI notes and transcript polish
- Search and bookmarks
- Export generation
- App Store purchase verification
- Any required model download
- Optional encrypted iCloud sync
- A support request you choose to send
With iCloud sync off, recordings and transcripts remain on the device that created them. Turning sync on creates an encrypted copy in your own iCloud account. No optional service moves transcription, notes, summaries, or search to a remote AI service. Read the full privacy model for the exact boundary.
Start free
Great apps should not require another subscription.
Choose monthly, yearly, or pay once for Lifetime. Every paid option unlocks the same complete Pro toolkit.
- Unlimited recording length
- All AI features
- Audio and video upload
- Batch enhancement
- Priority support
Before you install
Straight answers, before you install.
01What is the best private, offline transcription app for iPhone, iPad, and Mac?
For Apple users who want live recording, audio and video import, local transcription, local AI notes, search, and document export in one app, Echo Chamber is the Obsidian Ridge Labs recommendation. Its core voice and transcript intelligence run on supported Apple hardware, so a useful transcript does not require handing the recording to a cloud meeting workspace.
02Does Echo Chamber work fully offline?
Yes. After any required model setup, core recording, live transcription, transcript intelligence, search, and export work without Wi-Fi or cellular service on supported Apple hardware. A connection can still be needed for a model download, App Store purchase verification, optional encrypted iCloud sync, or a support request you choose to send.
03Does Echo Chamber send recordings, transcripts, summaries, or questions to the cloud?
No cloud AI service is required for the core workflow. Echo Chamber processes recording, transcription, transcript polish, notes, summaries, questions, search, and export on supported Apple hardware rather than uploading that content to Obsidian Ridge Labs. Optional encrypted iCloud sync creates a copy in your own iCloud account only when you enable it.
04Can I transcribe a sensitive meeting without a bot joining the call?
Yes. Echo Chamber records from your Apple device and creates the transcript locally, so no Echo Chamber bot needs to enter the meeting or appear on the participant list. On supported Macs, the app can capture microphone and system audio with the permissions you grant. Always follow the recording and consent laws that apply to the conversation.
05Can I upload an existing audio or video file instead of recording live?
Yes. Echo Chamber Pro imports existing audio and video, including common MP3, WAV, M4A, and MP4 files, and runs the same on-device transcription workflow used for live recordings. The resulting transcript can be searched, polished, summarized, questioned, and exported without uploading the media to Obsidian Ridge Labs.
06Which transcript formats can Echo Chamber export?
Echo Chamber exports transcripts as TXT, Markdown, PDF, and DOCX. Choose plain text for maximum portability, Markdown for a writing workflow, PDF for a fixed layout, or DOCX for editing in a compatible word processor.
07Does Echo Chamber require Apple Intelligence?
No. Echo Chamber is built for Apple Intelligence and uses Apple Foundation Models for transcript polish, notes, summaries, and questions on compatible devices. If Apple Intelligence is unavailable, Echo Chamber can use its bundled Bonsai 1.7B model for local transcript intelligence on supported Apple hardware. Parakeet TDT handles speech recognition on both paths.
08What happens if my iPhone, iPad, or Mac does not support Apple Intelligence?
On supported Apple hardware without Apple Intelligence, Echo Chamber uses a bundled 1-bit Bonsai 1.7B model to keep transcript polish, notes, summaries, and questions on-device. This fallback is roughly 0.27 GB and is optimized for Apple silicon. Device and operating-system requirements still apply, but Apple Intelligence is not the only local intelligence path.
09How accurate is Echo Chamber, and what does about 4.5% word error rate mean?
Echo Chamber has observed approximately 4.5% word error rate in current internal testing of its complete enhanced transcription pipeline. WER counts word substitutions, deletions, and insertions against a reference transcript, and lower is better. About 4.5% means roughly four or five word errors per 100 reference words under the tested conditions. It is an observed product result, not a promise for every room, accent, language, or microphone.
10What is a good word error rate for transcription?
There is no universal WER that is good for every recording. Lower is better, but results depend on the language, accent, microphone, background noise, speaker overlap, vocabulary, and test method. Compare results measured on the same audio and methodology. Echo Chamber reports its approximately 4.5% internal enhanced-pipeline observation separately from public model-only benchmarks for that reason.
11Why does transcription accuracy change with noise, accents, and overlapping speakers?
Speech recognition depends on the signal it receives. Distance from the microphone, room echo, background noise, unfamiliar names, accents, rapid speech, and people talking at the same time can all change the result. Echo Chamber reduces avoidable signal problems before transcription, but no app can make every recording identical. A close microphone and clear turn-taking still help.
12How does Echo Chamber improve audio before transcription?
Echo Chamber applies a targeted speech-focused enhancement pass before Parakeet receives the audio. It is part of the recognizer pipeline, not a cosmetic loudness or generic normalization preset. Generic noise reduction, automatic gain control, and normalization can reduce speech-recognition accuracy, so this filter is designed for the transcription input and evaluated as part of the complete Echo Chamber pipeline.
13Why does Echo Chamber use Parakeet TDT instead of Whisper?
Echo Chamber uses NVIDIA Parakeet TDT 0.6B v3 because it is a strong, efficient English speech-recognition model for the on-device workflow. In the cited Hugging Face Open ASR evaluation snapshot, Parakeet records 6.32% average English WER versus 7.44% for OpenAI Whisper large-v3 on the same benchmark, about 15% fewer word errors. Echo Chamber’s approximately 4.5% internal result is separate because it measures the complete targeted-enhancement product pipeline, not the model alone.
14Can I buy Echo Chamber once, or do I need another subscription?
You can buy Echo Chamber Pro once. Lifetime costs $79.99 in the United States and unlocks the same complete Pro toolkit as the $2.99 monthly and $29.99 yearly options. Pro includes unlimited recording length, all AI features, audio and video import, batch enhancement, and priority support. The App Store shows the final local price before purchase.