Releases
What's new
Every echo99 release and what it brought, newest first.
Latest: version 1.4, released . Download echo99. Already installed? echo99 checks for a new version each time it launches and tells you when one is ready.
Version 1.4
- One person, one speaker: when someone's voice changes during a call and echo99 heard them as two speakers, it now merges them back into one, so naming them once names them everywhere in the transcript.
- Better speaker separation with the default speaker model: speakers are now grouped over the whole call, so a new voice is no longer filed under someone who spoke earlier.
- With the offline speaker model and the number of speakers on Auto, echo99 no longer forces a guessed speaker count, which often came out badly wrong on larger meetings and one-on-one calls.
- More reliable speaker recognition: long silences are kept out of a speaker's voiceprint, so their sample clip no longer plays silence and they are matched to the right person more often.
- Welcome window: shows download progress for the speech models, opens right under the menu-bar icon, stays on screen on smaller displays, and no longer crashes when the recordings folder can't be used.
Version 1.3
- New speaker model: Nemotron 3 (experimental), which tells apart up to 8 people on the other side of a call, is now offered in Config → Models, on Intel Macs too. It replaces Sortformer, which stays available for anyone already using it.
- Safer model downloads: an interrupted download picks up where it left off, and updating or reinstalling a speaker model keeps the working copy until the new one is complete.
- Recordings window: fixed a crash when selecting a recording, the window opens at a sensible size, the recordings list stays visible on smaller screens, and speaker rows stay readable in a narrow window. A recording transcribed with speaker recognition off now says so and offers to re-transcribe it.
- System audio speed correction moved to Config → Debug. It stays on by default.
Version 1.2
- New transcription model: Parakeet Ultra is now offered in Config → Models, alongside the existing speech models.
- Bigger summary model: Gemma 4 12B is now available in Config → LLM for more detailed call summaries on Macs with memory to spare. A model that's currently in use can no longer be reinstalled or deleted by mistake.
- Better speaker separation: two different people are less often merged into one speaker, and one person's sentence is no longer split across several lines of the transcript.
- Recording alignment: the other side of a call now lines up with your microphone from the very first second, and the recorded speed is right from the start with more audio devices.
- Small conveniences: the menu bar window shows how long the current recording has been running, the Recordings window can stay on top of other windows, the Config menu is grouped under headings, and echo99 checks for updates on every launch.
Version 1.1
- Call summaries, written on your Mac: echo99 can now summarize any transcribed recording with a small language model that runs entirely on your Mac, so the transcript never leaves it. Pick a model in Config → LLM, then open Summary in the Recordings window, or have every recording summarized automatically after transcription.
- Live transcript (experimental): watch what's being said while the call is still recording, in a small floating window. Turn it on in Config → Experiments and choose its speech model in Config → Models.
- System audio speed correction: with some headsets and audio devices, the other side of a call could be recorded sped up, and their part of the transcript came out garbled. echo99 now notices within a few seconds and records the rest at the right speed. It's on by default, in Config → Transcription.
- More reliable recording: fixed a rare case where the other side of a long call went silent partway through, and your microphone and the other side now stay in sync even when the audio device changes mid-call.
- Cleaner transcripts: a sentence is no longer split between two speakers, skipping silence no longer cuts off speech at the very end, and speaker echo removal now matches words by their timing, so it catches more duplicates.
Version 1.0
- Import a recording: hand echo99 an existing audio or video file (WAV, MP3, M4A, FLAC, MP4, MOV) and it transcribes it like a call you recorded: speakers separated, searchable, exportable.
- Refreshed menu bar: clicking echo99 now opens a small window instead of a plain menu, with Record and Import side by side, your latest recordings with their transcription status, and any warning right there.
- The rest of the app got the same visual pass: clearer buttons, calmer colors, and less repeated text in the lists.
Version 0.8
- Export a transcript as Markdown: the Recordings window can now save the whole conversation, with timestamps and speaker names, as a Markdown file, ready to paste into your notes.
- Voice samples for speakers: play a short identification clip of each speaker straight from the transcript, so naming who's who is easier.
- If the speaker-recognition model is missing, echo99 now warns you and offers to install it instead of silently skipping speaker matching.
- Fixed Bluetooth headsets: the other side of a call no longer comes out sped-up and distorted when the headset switches audio profiles mid-recording.
Version 0.7
- Full-call playback: play a recording straight from the Recordings window, scrub through it, change speed, and balance your mic against the other side.
- Speaker activity timeline shows who spoke when across the call.
- Recordings are now stored at a fixed 24 kHz mono: smaller files, same transcription quality.
Version 0.6
- New Recordings window: browse recordings, read transcripts, and play any line back.
- People: name a speaker once and echo99 recognizes their voice in future calls; your People list can be exported and imported.
- Re-transcribe any recording with a different model. If anything fails, the old transcript is kept.
- Global keyboard shortcut to start and stop recording (⌃⌥R by default, configurable in Config → Shortcuts).
Version 0.5
- Experimental Sortformer speaker-separation model for Apple silicon Macs (Config → Models).
- More accurate speaker counting with the standard speaker model, tuned on real multi-party calls.
- Recording now survives switching your microphone or headphones mid-call, including Bluetooth.
Version 0.4
- Skip silence during transcription (opt-in): long quiet stretches no longer slow transcription down. It uses the optional Silero VAD model from Config → Models.
- Mic and system tracks now record in mono: smaller files, same speech quality.
- Sturdier crash recovery: an interrupted recording is recovered even when only one track was written, and a finished recording can no longer be lost when quitting.
- Opt-in debug logging (Config → Debug) to help track down problems.
Version 0.3
- Czech localization completed: the Config window, model descriptions, and every error alert.
- Fixed alerts that mixed English and Czech in one message.
Version 0.2
- First public release as echo99, signed and notarized for macOS.
- Automatic update check: echo99 lets you know when a new version is available.
- The whole app speaks Czech as well as English.
- New app icon and a branded installer.