Repository navigation
Releases: qwertymuzaffar/parrot
Release list
Parrot v0.5.1
Thirty-six fixes from a review of Speak freely and two stress tests of the app. Nothing new to learn: everything works as 0.5.0 meant it to.
Improved
- Long texts are checked fast. Comparing a long text with what was heard was slow in the length of the text: a 30-minute talk sat at 100% for half a minute before its check screen showed, and every take of a long text read whole took seconds to score. Both now take moments.
- Home, and the part count under Short parts, no longer stall on a long text without punctuation, such as pasted captions or dictation.
- The library is read off the window's thread, and the lists read only a try's score and times, so a library with long talks in it no longer freezes the window on every load.
- A talk's sound is let go once it has been heard, saving some 350 MB of memory by the end of a 30-minute talk; and the splits of a text's old words are forgotten when its words change or it is deleted.
- Home shows its cards at once on every return, with no "Nothing to practise yet" frame first, and Save and practise keeps the check screen up until the new text is open, instead of flashing Home.
- A URL or other long word stays inside its box instead of scrolling the sheet sideways, and a long title is held to two lines on the Continue card.
- The window is granted only what it uses: its version, the engine's events, dragging and zooming by its top strip, and recordings from the texts and talks folders only.
Fixed
Talks
- Cancel while the microphone gets ready, then Start talking again, no longer leaves a talk that nothing can stop.
- A talk the engine refuses is dropped at once, with what went wrong said on the card, instead of leaving a new talk running with no Stop.
- After Stop, the card says Stopping… with nothing to press, instead of offering a Cancel that set the talk aside under the stop.
- A talk the engine stops taking no longer leaves the card at "Stopping…" for the session: after ten seconds it is kept as the engine has it, to write down or let go.
- Writing a talk down stops if that talk was discarded meanwhile, so it can no longer finish over a new one.
- Keeping a talk files the try and the sentences' takes first and moves the recording in last, so a failure on the way, such as a full disk, leaves the talk to keep again instead of losing its recording.
- A keep cut short by a quit, crash or power cut is finished at the next start, instead of being offered again and filed twice.
- A talk kept into a text that already has its words leaves that text its title, and a text whose words were fixed no longer fails to keep a talk corrected to its original words.
Takes and scores
- A take shows only on its own part: pressing an arrow, a step or a text while a take is scored no longer dresses the part moved to with its colours, chart and score.
- No take starts while a text or mode is still loading, so Record pressed right after a click no longer files the take under the text left behind.
- A take still being checked when another text was opened keeps its score with its own text, not the text opened.
- A word's take whose start was left while the microphone got ready no longer stops the take of the word clicked after it.
- A control no longer stays disabled for the rest of the session when the engine does not answer: each wait has a limit, after which the error is shown and the control comes back.
- When the dictionary, the engine or the trash cannot be read, the word card and Settings say so instead of freezing.
Texts and the library
- A text opening slowly gives way to Home, a talk, or another text or mode chosen meanwhile, and says it is on its way while it waits.
- A rename, delete, mode change or new text made while the library reloads is kept, instead of undone when the old list lands.
- Switching mode or text shows the new parts, their scores and the part together, instead of the new words in the old try's colours.
- A text can be edited whatever its tries' recordings, and an edit moves the tries as a whole: a failure halfway leaves the text as it was.
- A changed part's tries go to the part of the new text most of its words went to, and a text's scores count only the parts it has.
- Edit the words offers Cancel once the engine has gone a while without a word, keeping the edit in the box to save again.
- The same words are one text whichever folder holds them; a new text whose words name a folder already taken gets -2, -3… on the end.
What counts as heard right
- "it's" and "its", "let's" and "lets", "who's" and "whose" match whichever one Whisper writes.
- "U.S.", "U.K.", "p.m." and other dotted abbreviations match when Whisper writes them plain, and "Node.js" matches either way.
- "2nd" matches "second" and "22nd" matches "twenty-second", while "twenty seconds" stays what it was.
- A letter with two marks, as in "Nguyễn" or "Hà Nội", is one letter: such a name is no longer cut into pieces and marked misheard.
- "2.5" is read out as 2.5 and "Node.js" is kept whole: the window trims only a word's ends, as the engine does.
Your files
- Two changes to one file at once each get their own temporary file, so neither is lost and meta.json cannot be left in pieces.
- A try or details file that cannot be read is set aside as
<name>.damaged-<time>and noted in the log, instead of being written over. - The waveform of a damaged recording is refused at once, instead of keeping the engine reading for minutes on every open.
Installing
The installers are still unsigned. On macOS, right-click Parrot and choose Open (or go to System Settings → Privacy & Security → Open Anyway). On Windows, click More info → Run anyway.
If you have an earlier version installed, your texts and recordings are kept.
Parrot v0.5.0
New
- Talk for up to 30 minutes. Just talk now runs for up to 30 minutes, up from 10. It still stops by itself after 2 quiet minutes.
- Written down as you talk. Parrot writes down what it hears while you are talking, with the last words heard on the card, so even a long talk is ready moments after you stop.
- Prompts. For something to talk about, pick a set of prompts on the Just talk card (interview, small talk or presentations). Parrot shows one at a time, with another one a click away, and the prompt becomes the title of the text the talk is saved as.
- A take per sentence. A saved talk can be practised sentence by sentence, each sentence with its first try cut from the talk.
- Recover a talk. A talk not saved when Parrot quits or crashes is offered again the next time it starts, to write down or discard. Going Home from the check screen leaves the talk, with the words you fixed, on the Just talk card to come back to.
- Edit a text. Fix a text's words at any time, from the pencil beside its title. The tries of each part you change are scored again from their recordings; the other parts keep theirs as they are.
Improved
- A word with an accent, such as "café" or "Zürich", is heard as one word, and no longer marked misheard when said right.
- A long take is scored a stretch at a time, in far less memory: reading a long text whole, or a long talk again, no longer risks the app running out of memory.
- A line sets the window buttons apart from the app, and the headers of Settings and Edit the words from what is under them.
- Renaming a text on Home shows a check mark to save the title.
- A screen reader is told when a talk starts listening, when there is a minute left and when it is being written down, instead of the clock every second and the words heard so far.
Fixed
- A talk can no longer be discarded or left while it is being saved.
- Opening a text while a talk was recorded left the talk recording. It now stops, and waits on the Just talk card.
- A talk's recording is written whole or not at all, so quitting while it is saved can't leave it empty.
- A long talk recorded from a microphone set to 96 kHz could not be written down.
- Adding a text you already have opens it as it is, instead of undoing the words you fixed in it.
- Esc that closes Settings or leaves a field no longer stops a talk.
- A talk recovered after a quit is no longer titled with a prompt it was never about.
- The words can't be edited while a take is being checked, which filed the take under the old words.
- The waveform fits in a narrow window, and a name stays in the line of a talk's words.
Installing
The installers are still unsigned. On macOS, right-click Parrot and choose Open (or go to System Settings → Privacy & Security → Open Anyway). On Windows, click More info → Run anyway.
If you have an earlier version installed, your texts and recordings are kept.
Parrot v0.4.0
New
- Just talk. Tell a story or answer an interview question with nothing to read, for up to 10 minutes. Parrot writes down what you said and shows how each word came out. Fix any word it misheard, give the talk a title, and save it: it becomes a text you can practise reading, with your talk as its first take. A talk stops by itself after 2 quiet minutes or at 10 minutes, and writing it down can be cancelled and run again.
- Pace and long pauses. Every take now shows your pace in words a minute, and marks each pause of over a second between words.
- Names. Mark a word as a name (in the word card), and a name Parrot hears differently no longer counts against your take. Names are one list for all texts, which you can review in Settings.
Fixed
- When the system withholds the microphone and Parrot gets only silence, it now says so, instead of "Nothing was heard".
- A score of 100% fits inside its ring.
Installing
The installers are still unsigned. On macOS, right-click Parrot and choose Open (or go to System Settings → Privacy & Security → Open Anyway). On Windows, click More info → Run anyway.
If you have an earlier version installed, your texts and recordings are kept.
Parrot v0.3.0
New
- Delete a try. Remove a single try from a part's tries chart or from a word card, without deleting the whole text.
- Trash that empties itself. A deleted text waits in the trash for 30 days and is then erased. Settings shows how much the trash holds and can empty it now, and Open data folder shows where your texts and recordings are kept.
- Check for updates. Settings has an opt-in Check for updates button. Parrot asks GitHub only when you click it.
- Keyboard and screen reader. The app can now be used from the keyboard and with a screen reader.
Fixed
- The speech models download more reliably: a broken download resumes where it stopped, each file is checked against its checksum, and a model that won't load is set aside and downloaded again.
- Long takes are scored more accurately: they are now aligned in windows of five seconds.
- A take is scored against the part it was started on, even if you move to another part before it finishes.
- When the engine fails, Parrot now says so, instead of quietly showing an empty or older result. A text the engine wouldn't add stays in the box, so you can try again.
- Word clips are no longer cut short, and stopping a comparison stops both clips.
- Parrot starts, and says so, when the system has no voice to read words aloud.
- The practice dock fits in narrow windows.
- On Windows, the speech models are kept in the local profile (
%LOCALAPPDATA%) rather than the roaming one. Models already downloaded are moved over on the first start. - The window now runs under a Content-Security-Policy.
- Third-party licence notices are included with the app.
Installing
The installers are still unsigned. On macOS, right-click Parrot and choose Open (or go to System Settings → Privacy & Security → Open Anyway). On Windows, click More info → Run anyway.
If you have 0.2 or 0.1 installed, your texts and recordings are kept. Note that anything already in the trash for more than 30 days is erased the first time 0.3.0 starts.
Parrot v0.2.0
New
- Practise one word on its own. Click a word and press Record the word. Say just that word, as often as you like: each try gets its own sound match and history, without changing the sentence's score or your progress.
- How to make the sound. When a sound comes out off, the word card now says how to make it: th, r, l, v, w, ng and the rest, silent letters, and a word's last sound that often gets dropped.
- Half-green sounds. A sound that was right but not sure shows half green, half amber, so a 92% word shows where the rest went. A clear word now explains its score instead of showing nothing.
- Pace. Words a minute now counts every word you read; a misheard last word no longer slows it.
Fixed
- The first recording after the microphone opens no longer starts quiet. On macOS the microphone comes up quiet for a few seconds; Parrot now opens it ahead of a take and waits for it to settle.
- A take that starts with a click, or a single short word stopped quickly, is no longer rejected as "Nothing was heard".
- Comparing a word plays yours, then the correct one; it used to stop after yours.
- Quitting no longer crashes.
- The recording timer stays still.
Installing
The installers are unsigned, as before: on macOS right-click Parrot and choose Open (or System Settings → Privacy & Security → Open Anyway); on Windows, More info → Run anyway. The speech models (about 800 MB) download on first launch. A 0.1 install keeps its texts and recordings.
Parrot 0.1.1
Download from the website: https://qwertymuzaffar.github.io/parrot/
What's new
- Intel Macs. Parrot now runs on Intel Macs with macOS 13.4 Ventura or newer.
- Takes keep their first words. Opening the microphone could take a second or more, and words said in that time were lost. Parrot now shows Recording only once sound is coming in, keeps the half second before your click, and starts the next take at once.
- Downloads start straight from the website instead of opening this page.
Which file
| System | File |
|---|---|
| macOS, Apple chip (M1 and newer), macOS 11+ | Parrot-macOS-Apple-chip.dmg |
| macOS, Intel, macOS 13.4+ | Parrot-macOS-Intel.dmg |
| Windows 10/11, 64-bit | Parrot-Windows-setup.exe or Parrot-Windows.msi |
| Linux (Ubuntu 24.04, Fedora 40, Debian 13 or newer) | Parrot-Linux.AppImage, Parrot-Linux.deb or the .rpm |
On first launch Parrot downloads its two speech models once (about 800 MB), then works offline.
Opening it the first time
The installers are not signed with paid developer certificates yet.
- macOS: if it says Apple cannot check Parrot, right-click the app and choose Open, or go to System Settings → Privacy & Security → Open Anyway. Once is enough.
- Windows: if SmartScreen says it protected your PC, click More info → Run anyway.
Parrot 0.1.0
The first release of Parrot: read a text aloud and see which words came out wrong.
Download from the website: https://qwertymuzaffar.github.io/parrot/
What's in it
- Paste any text and practise it sentence by sentence or in short parts; hear it read with the spoken word lit up.
- Record yourself: words that were heard as something else turn red (with what was heard), words that sound off turn amber.
- Click a word for how it sounds (a respelling like sik·STEEN and its IPA), hear it slowly, hear your own, or both.
- Every try is kept, so you can watch your score go up.
- Everything runs on your computer. On first launch Parrot downloads two speech models (about 800 MB) once, then works offline.
Which file
| System | File |
|---|---|
| macOS, Apple Silicon (M1 and newer), macOS 11+ | Parrot_0.1.0_aarch64.dmg |
| Windows 10/11, 64-bit | Parrot_0.1.0_x64-setup.exe or Parrot_0.1.0_x64_en-US.msi |
| Linux (Ubuntu 24.04, Fedora 40, Debian 13 or newer) | .AppImage, .deb or .rpm |
Intel Macs come in 0.1.1.
Opening it the first time
The installers are not signed with paid developer certificates yet.
- macOS: if it says Apple cannot check Parrot, right-click the app and choose Open, or go to System Settings → Privacy & Security → Open Anyway. Once is enough.
- Windows: if SmartScreen says it protected your PC, click More info → Run anyway.
Speech models v1
Models Parrot downloads on first launch. Not an app release: get Parrot from https://qwertymuzaffar.github.io/parrot/
- wav2vec2-xlsr-53-english-int8.onnx (317 MB): jonatasgrosman/wav2vec2-large-xlsr-53-english, fine-tuned on Mozilla Common Voice, exported to ONNX and quantised to 8 bits. Apache 2.0. Finds each word in a take and scores its sounds.
Whisper small (English) is downloaded from ggerganov/whisper.cpp (MIT).