Hosted transcription uses our free Hugging Face Space with Shook for Amharic and Whisper Small English for English. Audio is sent there only when you choose Transcribe. Free queues and daily limits apply. Automatic translation is not connected; you can type a translation yourself.
Advanced: enter your own Python backend below to use local transcription and Ollama translation. Leave the address empty to use the hosted speech service.
Recordings and stories are saved in this browser. When using a custom backend, Transcribe or Translate sends the relevant audio or text to that backend. Hosted transcription sends audio to Hugging Face for in-memory processing. File upload/download routes are disabled and transcript results are encrypted for the requesting browser. Your browser may request permission to connect to a local service.
The project includes a Python backend, requirements, a Dockerfile, and a setup guide. Run ./start.sh from the source checkout, then connect to http://localhost:8000. For the fully local experience, open that address directly. A hosted preview needs a reachable backend for AI features.
./start.sh
http://localhost:8000
Record a parent or grandparent’s memories in Amharic or English, review the transcript, and turn their words into a picture storybook.
Add your own family photos or illustrations and captions to the cover and pages. Automatic AI illustrations are not connected yet. You can add a translation yourself; automatic translation is not available in the hosted app.
Download digital book: an HTML storybook with your pictures, text, and original voice recording embedded. Open it in a browser to read and listen offline.
Print / Save PDF: a printable storybook with your pictures, captions, and text. PDF files do not include playable audio.
Download keepsake: a simpler HTML copy with the original recording, transcript, and any translation. There is no separate audio-only download button yet.
Saved memories can be protected with a password in this browser. Protection encrypts words, names, pictures, and recordings; use Lock stories when you finish. The key stays in memory only, so reloading locks protected storage. There is no password recovery. Existing unprotected memories are encrypted when you enable protection. Downloads are not encrypted: anyone with the downloaded file can read it and play any included recording. Your saved memories and storybooks stay on this device. They do not sync between devices. Download a digital book or keepsake to back up the original voice. Pictures are resized in your browser.
When you choose Transcribe, audio is sent to our Hugging Face Space running Shook Tiny Amharic or Whisper Small English. Free queues and usage limits apply. Listen and review names, dates, and meaning before sharing.
The written example is fictional and its Amharic translation is illustrative. Your recordings preserve the original voice.
Choose Protect stories to set a password. This encrypts saved story text, names, questions, photos, and original recordings in this browser using AES-256-GCM. The password is used on your device and is not sent to us or Hugging Face. Older memories remain unprotected until you enable this.
Choose Lock stories when you finish. Protected storage also starts locked after a reload. Closing other app tabs before setup helps protect existing memories consistently. Protection does not lock your whole computer or defend against malicious software or browser extensions.
Use a long, unique password and keep it somewhere safe. There is no password recovery. Clearing browser data removes saved memories, and the app does not sync them or provide a cloud backup.
Recording, writing, and arranging a storybook happen in your browser. Before transcription, we ask for permission to send that recording over HTTPS to our Hugging Face Space. The service needs readable audio to transcribe it: this step is not end-to-end encrypted.
Our hosted speech code processes audio in memory without creating downloadable audio files. File upload and file download routes are disabled. Transcript results are encrypted for a one-request key held only in your browser, so a copied result link alone cannot reveal the words. The model and hosting infrastructure still handle readable audio during processing; we cannot guarantee provider-wide deletion or secure erasure from memory. Read Hugging Face’s privacy policy. An optional custom backend receives the audio or text instead when you configure one.
If you allow analytics, Google Analytics uses cookies to count visits and report general device and referring-site information. We send a fixed page title and website address, and only the referring site’s origin, without query strings. We do not send story text, names, recordings, pictures, passwords, searches, or form entries as analytics events. Automatic form, search, link, and download tracking are disabled. Google still receives connection information, such as your IP address, when contacted. Analytics stays off until you opt in; you can decline or withdraw using Analytics choices in the footer. Browser privacy signals are respected. See Google’s privacy policy.
HTML books and keepsakes include any attached original voice. PDFs include pictures and text, but no playable audio. These downloaded files are not password-protected; anyone with a copy can read them and play any included recording.
These protections reduce risk; this prototype has not had an independent security audit. Never record or share someone’s story without their permission.