Bamlis
Speech to text. On your device.
Bamlis turns recordings, voice memos and videos into text without sending a single byte to a server. Finnish by default, more than twenty other languages, automatic speaker names and subtitle export. Free, no account, no tracking.
- iPhone
- iPad
- Mac
- Apple Watch
- English · Suomi · Svenska · Deutsch · Français · Español
What it does
Transcribes any recording
Audio files, voice memos and the audio track of videos: MP4, MOV, M4A, MP3, WAV, MKV, WebM, FLV and more. Share a file to Bamlis from any app, or pick it from Files.
Finnish and 20+ languages
WhisperKit (OpenAI Whisper on Core ML) gives accurate text with punctuation and sentence breaks. Apple's built-in speech recognition is available as a second engine.
Knows who is speaking
Speakers are identified automatically. Rename “Speaker 1” to a real name in one step, or fix a single sentence.
Listen, check, edit
Play the recording with the current sentence highlighted, tap a sentence to jump to it, and edit the text in place.
Record anywhere
Record on iPhone, iPad, Mac or Apple Watch. Watch recordings go to your iPhone for transcription, and the Apple Watch Ultra Action button can start and stop them.
Export for every use
Text with paragraphs, text without speaker names, text with timestamps, subtitles (.srt) for video editing and a spreadsheet (.csv) for analysis.
Private by design
Your audio and your text never leave the device. Transcripts sync only through your own iCloud. No account, no ads, no analytics.
Free and open
Built on open source models and software: WhisperKit and SpeakerKit by Argmax, OpenAI Whisper and pyannote speaker diarization.
Support
Frequently asked questions
Why does the first transcription take so long?
The first time you use a speech model, Bamlis downloads it (about 630 MB for the default Whisper model) and then prepares it for your device. On an iPhone this preparation can take a few minutes and happens once per installation. Keep Bamlis open; the screen stays on by itself. After that, models load in seconds.
Does Bamlis need an internet connection?
Only to download a speech model the first time it is needed, and for iCloud sync if you use it. Transcription itself runs offline on your device.
Do I need an account?
No. Bamlis has no account or sign-in. iCloud is optional: with it, your transcripts appear on your iPhone, iPad and Mac; without it, the app works on one device.
Which engine should I use?
WhisperKit is the default and gives the best results for Finnish and most languages, including punctuation. Apple Speech is faster to start and needs no large download, but it is less accurate for Finnish. Switch under the globe menu → Engine.
How do I send a file to Bamlis?
Tap + in Bamlis and pick a file, or use Share → Bamlis from Voice Memos, Files or any other app. On the Mac, use Open With in Finder or drop the file on the app icon. Several files can be queued at once.
The speakers are wrong or mixed up.
Automatic identification works best with clear, separate voices. Tap Edit, open a sentence’s speaker menu and choose the right name, or use “Rename … Everywhere” to give a speaker a real name throughout. You can also turn Identify Speakers off in the globe menu.
How do I use the Apple Watch Ultra Action button?
Open Bamlis on the watch once, then on the watch go to Settings → Action Button → Action: Control → Control: Bamlis. One press starts a recording, the next press stops it and sends it to your iPhone. On other watches, say “Record with Bamlis” to Siri or add the shortcut from the Shortcuts app.
How do I free the space the models use?
Globe menu → About Bamlis → Downloaded Models, then delete a model. It is downloaded again the next time it is needed.
My transcripts are not syncing between devices.
Check that every device is signed in to the same Apple Account and that iCloud is allowed for Bamlis in Settings → [your name] → iCloud. Audio files themselves stay on the device where they were imported; only the transcripts sync.
Privacy policy
Bamlis does not collect any personal data.
- Your audio files, recordings and transcripts are stored on your device and, if you use iCloud, in your own private iCloud database (Apple CloudKit). The developer cannot access them.
- Transcription and speaker identification run entirely on your device. No audio or text is ever sent to the developer or to any server.
- The microphone is used only when you start a recording. Recordings stay on your device.
- Bamlis connects to the internet only to download speech models from Hugging Face (huggingface.co) the first time a model is needed. The request contains no personal data or identifiers beyond what any file download includes, such as your IP address, which Hugging Face handles under its own privacy policy. Downloaded models can be deleted in the app.
- Recordings made on Apple Watch are transferred to your iPhone over Apple’s WatchConnectivity.
- There is no account, no advertising, no analytics and no tracking.
Contact: stindesofta@gmail.com
Tietosuojakäytäntö (suomeksi)
Bamlis ei kerää mitään henkilötietoja.
- Äänitiedostot, äänitykset ja litteroinnit tallennetaan laitteellesi ja, jos käytät iCloudia, omaan yksityiseen iCloud-tietokantaasi (Apple CloudKit). Kehittäjä ei pääse niihin käsiksi.
- Litterointi ja puhujantunnistus tapahtuvat kokonaan laitteellasi. Ääntä tai tekstiä ei koskaan lähetetä kehittäjälle eikä millekään palvelimelle.
- Mikrofonia käytetään vain, kun aloitat äänityksen. Äänitykset pysyvät laitteellasi.
- Bamlis käyttää verkkoa vain puhemallien lataamiseen Hugging Facesta (huggingface.co), kun mallia tarvitaan ensimmäisen kerran. Pyyntö ei sisällä henkilötietoja tai tunnisteita muuten kuin minkä tahansa tiedostolatauksen tavoin, esimerkiksi IP-osoitteen, jota Hugging Face käsittelee oman tietosuojakäytäntönsä mukaisesti. Ladatut mallit voi poistaa sovelluksessa.
- Apple Watchilla tehdyt äänitykset siirretään iPhoneen Applen WatchConnectivity-yhteydellä.
- Ei tunnuksia, ei mainontaa, ei analytiikkaa, ei seurantaa.
Open source
Bamlis uses WhisperKit and SpeakerKit (Argmax, Inc., MIT License), OpenAI Whisper models (MIT License), pyannote speaker-diarization-community-1 (MIT License) and Swift Argument Parser (Apache License 2.0). The full license texts are in the app under About Bamlis.