This build replaces builds 32 through 35. It contains the same two new screens as build 32, plus four fixes. All of them were verified on a physical Android device before submitting. Sign-in, subscriptions, in-app purchases, permissions, and data handling are unchanged. No new permissions are requested. 1) Feedback (free-text, one line) Where: Settings > "Send feedback". It also appears once in a while after a dictation result screen, with a Skip button of equal weight. What it does: sends one short text note to the developer's own database. The text is NEVER shown to any other user, is not published anywhere in the app, and there is no feed, profile, or messaging. It is write-only: the app cannot read back any note, including the user's own. We do not ask for a name, an email address, or any contact detail. Notes are deleted with the account. 2) Listen again to a finished clip Where: Home > pick a topic > tap a clip you have already worked on. Under that row there is a "Listen straight through" button. What it does: plays the clip audio end to end. It does not open the dictation exercise and does not consume a trial episode. Fixed in this build • Clip list: the highlight colour and the tap ripple of rows you have already worked on were painted by a container placed above the row, which covered them. The colour is now painted on the row itself. • Conversation audio: when the storage gateway timed out, the audio attached to a saved phrase card was dropped silently, so the card appeared in review without sound. That upload is idempotent, so it is now retried once. • Dropped connections were not recognised as network failures, so they were reported as application errors instead of connectivity warnings. They are classified correctly now. No network call was added or repeated. • Sign-up sheet: with the on-screen keyboard open, the sheet was taller than the space left for it, so the "Send code" button was clipped on shorter screens and at larger text sizes. The sheet now scrolls. This is the only change in this build compared with build 35. How to try it • Settings > "Send feedback" > type anything > Send. A thank-you toast appears. Nothing is displayed back. • Home > any topic > finish one clip > return to the topic list. The row for that clip is tinted, taps show a ripple, and a "Listen straight through" button sits under it. Also in this build: spelling variants such as "alright" / "all right" are no longer graded as wrong, and internal analytics fields were added. Neither is visible as a new screen.

This build replaces build 32 and 33. It contains the same two new screens as build 32, plus two fixes. Sign-in, subscriptions, in-app purchases, permissions, and data handling are unchanged. No new permissions are requested. 1) Feedback (free-text, one line) Where: Settings > "Send feedback". It also appears once in a while after a dictation result screen, with a Skip button of equal weight. What it does: sends one short text note to the developer's own database. The text is NEVER shown to any other user, is not published anywhere in the app, and there is no feed, profile, or messaging. It is write-only: the app cannot read back any note, including the user's own. We do not ask for a name, an email address, or any contact detail. Notes are deleted with the account. 2) Listen again to a finished clip Where: Home > pick a topic > tap a clip you have already worked on. Under that row there is a "Listen straight through" button. What it does: plays the clip audio end to end. It does not open the dictation exercise and does not consume a trial episode. Fixed in this build • Clip list: the highlight colour and the tap ripple of rows you have already worked on were painted by a container placed above the row, which covered them. The colour is now painted on the row itself. • Conversation audio: when the storage gateway timed out, the audio attached to a saved phrase card was dropped silently, so the card appeared in review without sound. That upload is idempotent, so it is now retried once. How to try it • Settings > "Send feedback" > type anything > Send. A thank-you toast appears. Nothing is displayed back. • Home > any topic > finish one clip > return to the topic list. The row for that clip is tinted, taps show a ripple, and a "Listen straight through" button sits under it. Also in this build: spelling variants such as "alright" / "all right" are no longer graded as wrong, and internal analytics fields were added. Neither is visible as a new screen.

Two new screens in this build. Sign-in, subscriptions, in-app purchases, permissions, and data handling are unchanged from the previous build. No new permissions are requested. 1) Feedback (free-text, one line) Where: Settings > "Send feedback". It also appears once in a while after a dictation result screen, with a Skip button of equal weight. What it does: sends one short text note to the developer's own database. The text is NEVER shown to any other user, is not published anywhere in the app, and there is no feed, profile, or messaging. It is write-only: the app cannot read back any note, including the user's own. We do not ask for a name, an email address, or any contact detail. Notes are deleted with the account. 2) Listen again to a finished clip Where: Home > pick a topic > tap a clip you have already worked on. Under that row there is a "Listen straight through" button. What it does: plays the clip audio end to end. It does not open the dictation exercise and does not consume a trial episode. How to try it - Settings > "Send feedback" > type anything > Send. A thank-you toast appears. Nothing is displayed back. - Home > any topic > finish one clip > return to the topic list > tap "Listen straight through" on that clip. Also in this build: spelling variants such as "alright" / "all right" are no longer graded as wrong, and internal analytics fields were added. Neither is visible as a new screen.

This is a bug-fix update. No screens, flows, or features were added or changed. Sign-in, subscriptions, in-app purchases, permissions, and data handling are unchanged from the previous build. What was fixed • Review screen: moving quickly from one card to the next cancelled the audio that was still loading, and the app showed a "couldn't play the audio" message even though the new card was playing correctly. The cancelled load is no longer treated as an error. • Network hiccups: a few read requests now retry once before showing an error, so a brief interruption no longer surfaces to the learner. Writes and paid operations are deliberately not retried. • Saved conversations: uploading the conversation audio now retries once, so fewer turns are missing when replaying a past conversation. Nothing to test differently from the previous build.

This update changes how imported audio is cut into sentences and chunks, and adds a few controls. Sign-in, subscriptions, in-app purchases, permissions, and data handling are unchanged. Imported audio ("Dictation with my file") • Transcripts used to be cut at the boundaries the speech-to-text service returned, which often fell mid-phrase. They are now cut at sentence punctuation, and a long sentence is split where the speaker actually pauses, or at a clause boundary if there is no pause. • A blank now covers a whole chunk of a few words instead of a single word. • The full-text view shows chunk boundaries as wider spacing between words. No characters are added to the text. • The editing screen was reorganised around chunks and renamed. Tapping between two words merges or splits a chunk. Tap targets were enlarged. • A short how-to page for this screen was added under Settings. Settings • Pause length between clauses is now a user choice (four steps). It multiplies the automatic value; it does not replace it. • A how-to page for conversation practice was added. • Retaking the level test now draws different questions each time. It previously always showed the same first questions. Reviewing the imported-audio feature • It requires a signed-in account, and on the free tier one import is allowed in total. This limit is unchanged from the shipped version; it is mentioned only because this update is mostly about that screen. • Existing content (the built-in clips) needs no account to try. The cutting described above runs on our backend and is already live, so a transcript imported today is already cut the new way. The chunk display, the editing screen, the pause-length control, and the level-test change are in the app and ship with this build.

This update changes audio timing and how a missed answer is replayed. Audio timing • Pauses between speakers were rebalanced so a change of speaker is easier to hear. • Speech rate between difficulty levels was corrected — level 2 audio was slower than level 1 by mistake and is now faster, as intended. • Long sentences now pause once mid-sentence. The speech itself is not slowed down; the learner is simply given time before the next clause. Replaying a missed answer (dictation) • When an answer is wrong, the app now replays the whole sentence once. On the next wrong attempt it replays only the blanked part. • A new text button, "Just that part", replays only the blanked part at any time. It is a playback control. It does not reveal the answer. • The replayed range is now taken from stored word timings instead of being estimated from character positions, so it no longer cuts off the target word or runs far past it. • After a wrong answer, words in the learner's own typed answer that did not match are struck through. Only what the learner typed is shown; the correct answer stays behind the existing "Show answer" button. The audio files are served from our backend, so the timing changes above are already live for the current App Store version. The pause and the replay behaviour are in the app and ship with this build. No changes to sign-in, subscriptions, in-app purchases, permissions, or data handling.