Transcription and recordings - Fixed an issue that could produce transcription in the wrong language or unreadable text on some devices: decoding now runs on the processor by default, guaranteeing transcription faithful to the selected language, still 100% on-device. - Removed periodic clicks (a "tick" about every half second) that could appear in new audio recordings. Reports - Previous version previews: all sections are now displayed — participants, topics discussed, decisions, action items, next steps — just like the current report. - Better preservation of numbers and dates during automatic text correction. - Fixed a crash when copying diagnostics.

- Fixed an issue where transcription could produce wrong-language or unreadable output on some iOS devices. - Improved transcription reliability across all languages. - Better preservation of numbers and dates during automatic text correction. - Improved overall stability.

Faster transcript correction (iOS 26+) - On Apple Intelligence-capable iPhones, transcript correction now relies on Apple's on-device model: significantly faster processing with lower memory use, with automatic fallback when unavailable. Everything remains 100 % on-device. Cleaner recordings and speakers - Noticeably better audio quality: removed a parasitic noise artifact and periodic clicks that degraded recordings, transcription and voice detection. - Better speaker distinction: up to 4 voices recognized on recent devices, up to 6 on the most powerful models; audio replayed through loudspeakers and genuine monologues are handled better. - Speaker editor: manually correct attribution per segment, merge speakers — regenerated reports use the new attribution. - Each device only downloads the components suited to its capacity, and superseded models can be deleted to free up space. More complete, faster reports - Complete reports on long transcripts: the entire text is now taken into account, whatever report template or canvas you pick — on every device. - Mid-range devices now produce complete, structured reports instead of partial ones. - Generation up to three times faster on modest devices, with hangs and abnormal waits fixed. - Displayed progress no longer moves backwards or stays stuck at 0%; properly capitalized titles and more accurate summaries for data reports. - Improved reliability across all 34 languages: apostrophes, contractions and copy adapted to each platform. Background processing and notifications - New "Background processing" setting: your choice persists and pending jobs resume when re-enabled. - Reliable resume of transcriptions and reports interrupted by backgrounding or app termination, within the windows granted by the system. - Completion notifications — including for a manually started generation — open the exact transcript or report. - The app stays responsive during processing: disk cleanup no longer competes with generation, and report navigation stays correct after a direct reopen.

Faster transcript correction (iOS 26+) - On Apple Intelligence-capable iPhones, transcript correction now relies on Apple's on-device model: significantly faster processing with lower memory use, with automatic fallback when unavailable. Everything remains 100 % on-device. Cleaner recordings and speakers - Noticeably better audio quality: removed a parasitic noise artifact and periodic clicks that degraded recordings, transcription and voice detection. - Better speaker distinction: up to 4 voices recognized on recent devices, up to 6 on the most powerful models; audio replayed through loudspeakers and genuine monologues are handled better. - Speaker editor: manually correct attribution per segment, merge speakers — regenerated reports use the new attribution. - Each device only downloads the components suited to its capacity, and superseded models can be deleted to free up space. More complete, faster reports - Complete reports on long transcripts: the entire text is now taken into account, whatever report template or canvas you pick — on every device. - Mid-range devices now produce complete, structured reports instead of partial ones. - Generation up to three times faster on modest devices, with hangs and abnormal waits fixed. - Displayed progress no longer moves backwards or stays stuck at 0%; properly capitalized titles and more accurate summaries for data reports. - Improved reliability across all 34 languages: apostrophes, contractions and copy adapted to each platform. Background processing and notifications - New "Background processing" setting: your choice persists and pending jobs resume when re-enabled. - Reliable resume of transcriptions and reports interrupted by backgrounding or app termination, within the windows granted by the system. - Completion notifications — including for a manually started generation — open the exact transcript or report. - The app stays responsive during processing: disk cleanup no longer competes with generation, and report navigation stays correct after a direct reopen.

Cleaner recordings - Fixed a gritty noise artifact added to every recording: the hiss filter ran per-sample and warped speech (which also degraded transcription and speaker detection). Filtering now works on smoothed 20 ms frames and only engages during sustained silences — speech passes through untouched. - Better speaker separation: 5.5–8 GB devices can now distinguish up to 4 voices (was 3), ≥8 GB devices up to 6. - Extra diagnostics when report generation fails ("insufficient content") to pinpoint the cause on-device. - Complete reports on long transcripts: the transcript is no longer shrunk to an excerpt before generation. The full text is processed in memory-sized chunks (facts, outline, entities, data) whatever report template or canvas the user picks — on every RAM tier.

Report quality - The "Bullet report" template produces real lists again: a bullet that merely repeats the summary or the title is now dropped. - Spelled-out numbers are recognized, including compound forms ("deux mille", "trois cents", "un million", "two thousand"…), across all supported Latin-script languages. - "Data report" again generates tables and charts when the transcript contains spelled-out quantities; when no numeric data is found, the report shows an explicit message. - Titles that are just a sentence fragment ("Dont …", "Duquel …") are rejected and rebuilt. - UI strings verified at full parity across all 34 locale catalogs.

SophiWhisper Premium is here: unlimited HQ transcription, advanced reports and pro exports. • High-quality transcription: better recognition in noisy environments and from a distance. • Advanced reports: more reliable extraction of names, numbers, decisions and action items. • Professional exports: PDF, Word, Markdown, email and print. • Smoother UI: fixed gestures, more responsive navigation, optimized animations. • One-time purchase, no subscription. Everything stays private and offline after resources are installed.

Install and data footprint - Significant reduction in download size at install: only essential resources are required, advanced voice modules are now optional. - Lighter default text-to-speech voices, with an option to upgrade to higher quality. - Automatic cleanup of space used by unused resources, without touching encrypted data or installed voices. Transcription quality - Better recognition for distant speech and noisy environments. - Automatic repair of words glued together or mis-segmented by transcription. - Filler-word removal ("um", "uh", etc.) across all 33 supported languages. - Spelling and agreement (gender, number) correction in the cleaned transcript. - First-person oral narration detection to prevent it from contaminating reports. Report quality - Removed stray labels and numeric ratings from topic bullets (e.g. "YES: 30%"). - No more raw transcript paste in summaries. - Cleaner summaries: better-formatted bullets, incomplete trailing sentences removed. - Clear in-app message when the device has limited capacity for generation. UI and navigation - Navigation locked during batch transcription to avoid memory conflicts. - Auto-resume of downloads when the app returns to the foreground. - Interface translated across all 33 supported languages.

- IOS component installation & UI fixes

- IOS component installation fixes

SophiWhisper — Your meeting, your privacy. SophiWhisper transcribes your meetings and turns them into actionable reports right on your device. Audio, transcripts, and reports never leave your phone — everything, including AI, runs locally. What’s new in v1.2: - More reliable reports: better identification of participants, numbers, tables, and decisions. - Improved readability: long text blocks are no longer cut off. - Cancel with care: a confirmation is now required before stopping transcription or report generation. - 33 supported languages for transcription and report generation. - Universal, multilingual fixes that work across all your content. - Fix the installation on Iphone 16 Perfect for professionals, defense and security forces, journalists, healthcare workers, and anyone who wants to keep their data under control.

SophiWhisper transcribes your meetings and turns them into actionable reports right on your device. Audio, transcripts, and reports never leave your phone — everything, including AI, runs locally. What’s new in v1.2: - More reliable reports: better identification of participants, numbers, tables, and decisions. - Improved readability: long text blocks are no longer cut off. - Cancel with care: a confirmation is now required before stopping transcription or report generation. - 33 supported languages for transcription and report generation. - Universal, multilingual fixes that work across all your content. Perfect for professionals, defense and security forces, journalists, healthcare workers, and anyone who wants to keep their data under control.

This version makes SophiWhisper even more useful day-to-day: - Audio groups : create groups of recordings and start transcription for all of them at once with the new Transcribe all button. - Delete transcript : when an audio file is removed, you can now delete the transcript — unless a report still depends on it. - Improved report history : the latest versions appear at the top, and your prompts are easier to read. - Prompts taken further : ask for a specific change and the report will better reflect your request. - As always, everything runs offline on your device, with no cloud, in 33 languages. Your meetings, interviews and notes stay private.