v1.1.5: spectrogram fix and folder settings

[1.1.5] — 2026-09-09

Added

• Folder names can be built from parts, like file names. Subfolder →
  From parts … offers calendar week, week span (07.09-13.09), date, year,
  month, day and free text, tickable and reorderable, joined with spaces —
  folders may hold those, file names here may not. KW37 07.09-13.09 Voice Training is what the default selection plus a word of your own produces,
  which is how such folders usually get made by hand. The fixed periods stay
  where they were; picking one changes nothing about how it behaves.
• A Day name part. With Month now contributing only the month, year,
  month and day can be ticked separately and add up to what Date gives in
  one piece.

Fixed

• The detail spectrogram stays sharp when you zoom in. It was computed
  once at the width of the chart, so zooming stretched a handful of columns
  into blocks. Both charts are now redrawn for whatever stretch of time is on
  screen. The spectrogram gets the same number of columns over less time, and
  its analysis window shrinks with the view — 1024 samples is 64 ms at 16 kHz,
  so at a 25 ms zoom every column used to see almost the same piece of audio
  and the picture turned into horizontal streaks. Below roughly a tenth of a
  second you now see the individual glottal pulses. The trade is frequency
  resolution, which is exactly the trade you want when zooming in. Brightness
  is still derived from the whole recording, so a quiet passage does not
  brighten up just because you looked at it closely. The image also sits where
  it belongs now: the first column starts half a window in, and that offset
  used to be ignored.
• The waveform above it is redrawn from the samples themselves. Its
  envelope had 2400 points for the whole recording — one every 2.5 ms in a
  six-second take, which at high zoom left a staircase with no relation to the
  signal. A narrow view now plots one point per sample.
• The spectrogram comes back at the height it left. Both charts in the
  advanced area now have fixed stretch factors; without them Qt redistributed
  the space by size hints and the chart returned smaller or larger than
  before.

v1.1.4 picture fix

[1.1.4] — 2026-09-09

Fixed

• The tutorial looks for its screenshots in more places, and says so when it
  finds none. It searched next to paths.py and under /usr/share only.
  That misses two real cases: the launcher being a symlink, since .resolve()
  follows it out of the folder the files actually live in, and the Arch
  package, which installs them under /usr/lib/dream-voicetraining. Both are
  covered now, along with the directory of the launched script and the working
  directory. A miss is written to the debug log with every path it tried, and
  the diagnostics show where the screenshots were found — a page without
  pictures is no longer a guessing game.

v1.1.3 UX

[1.1.3] — 2026-09-09

Added

• A spectrogram in the detail window. Advanced now shows one under the
  waveform, on the same time axis, so every moment sits at the same place in
  both. The selection can be dragged in either of them — a spot you recognise
  by its formants no longer has to be found in the waveform first. A checkbox
  next to the range switches it off. The contrast comes from the recording
  itself rather than from fixed limits, so a quiet take is not simply black.
• Import existing recordings. Import … on the Sessions tab takes WAV
  files from anywhere on disk, analyses them and files them like any other
  session. Files are copied, never moved — the originals stay where they are.
  Each one keeps its own file name, so it shows up in the list under the name
  it had on disk; two files of the same name get a number, and a name that
  cannot be used at all falls back to the normal scheme. A later move puts
  them in the right subfolder without renaming them. You pick one type for the
  whole selection and can change it per recording afterwards. Each file keeps
  its own sample rate rather than being assumed to match the microphone, which
  would put pitch and formants off by exactly that ratio, and its modification
  date becomes the recording time, so it lands in the right month folder and
  the right place in the list. A file that cannot be read is reported by name
  at the end instead of ending the run.
• 8, 24 and 32-bit WAV files can be read, not just 16-bit. Files with
  floating-point samples still cannot — those have to be converted first, and
  the import now says so instead of failing obscurely.
• Recording types of your own. Settings → Options now lists every type
  and lets you add your own — a warm-up, a passage, whatever you keep coming
  back to. They show up in the type picker on the Live tab, in the
  right-click menu of the session list and in the detail window, and they can
  go into the file name like the built-in ones do. Each gets its short form
  once, when it is created, and keeps it: recordings made weeks ago carry that
  short form in their file name, so renaming would lose them. Deleting a type
  says how many recordings use it and what they will read as afterwards —
  their files and values are untouched, but the name behind the type is gone.
• Practice texts you can choose between. The field on the Live tab now
  sits under a dropdown. The built-in text stays where it was, and anything
  you type can be saved under a name of its own — a warm-up, a paragraph from
  a newspaper, a passage you keep coming back to. Save as … offers the name
  of the entry you are on, so revising one is two clicks; on the built-in text
  the field starts empty, and it keeps its own wording. The choice is
  remembered, so the same text is up again next time. Switching away from
  unsaved changes asks first.
• The pitch history reads out under the mouse. Moving the pointer across
  Last 30 seconds draws a dashed line at that height with the frequency
  next to it, so you can measure off the curve instead of estimating between
  the grid lines. The readout follows the line, so it stays put while the
  time window scrolls underneath.
• Both live charts can be cleared by hand. Right-clicking the pitch
  history offers Clear history, the spectrogram Clear spectrogram. Since
  neither is wiped on start any more, this is the way to start fresh — after
  the third exercise the beginning of the first was still in the picture.
  The charts' built-in pyqtgraph menu is gone with it: zoom and axis ranges
  are switched off here, so every entry in it was without effect.

Fixed

• A saved design survives a restart again. Colours, background image and
  card opacity were written to the configuration correctly, but never read
  back: the call that applies them ran before the configuration was loaded and
  therefore always saw the empty default, which reset everything to the
  preset. The startup sequence now loads first and applies afterwards, and
  restoring an empty design is a no-op rather than a reset.
• "Whole recording" in the detail window now does something visible. It
  only reset the numbers back to the stored ones, so with no analysed
  selection in play the button appeared dead. It now opens the selection up
  to the entire recording as well, which is what the numbers in the table
  refer to. While the waveform has not loaded — or the file is gone — the
  button stays disabled instead of pretending.

Changed

• The Month name part is now just the month. It used to contribute the
  year as well, so switching on Year and Month together produced
  2026_2026-03. It now contributes 03, and the year comes from the part
  that is named after it. Recordings named under the old scheme are still
  recognised as self-generated, so a later move sorts them into folders
  instead of leaving them behind. The month subfolder keeps the year —
  it also drives the counter reset, and without the year January 2027 would
  land in January 2026's folder.
• Starting the stream no longer wipes the spectrogram and the pitch
  history. Both carry on where the previous run left off, so stopping to
  clear your throat no longer costs you the last thirty seconds. The clock
  behind the history only advances while the stream is running, so the pause
  leaves no gap in the curve.

v1.1.2 changelog and option tab rename folder and data name

[1.1.2] — 2026-09-07

Added

• A tab for how recordings are named. Settings → Options holds the
  recording folder and the whole naming scheme in one place, with a live
  preview of the next file name and its full path at the very top. It updates
  on every change and never consumes a counter number while you try things out.
• Subfolders per day, calendar week, month or year — or none at all.
  Recordings are filed by their own date, not by the day they were moved.
• A file name assembled from parts. Custom text front and back, year,
  month, calendar week, date, time of day, recording type and a running
  counter. Each part can be switched on and off and dragged into any order by
  its grip on the left, with arrow buttons for those who would rather not
  drag. Every row shows what it currently contributes, even while switched
  off. Parts that come out empty are simply left out.
• The preview shows the name in coloured badges, one per part, so it is
  obvious which switch produced which piece of the name. It sits in a card of
  its own above everything else: the folder small and dim on top, the file
  name large below it.
• Each part carries its own settings. Time with seconds sits under the
  time row, the number of digits and the reset interval under the counter row,
  and the two custom text fields are the rows themselves. The settings appear
  with their part and disappear with it — choosing seconds while the time is
  not in the name at all was a question without an effect.
• Fixed: the two buttons that move a part up and down showed a stub of a
  dash instead of their arrows. They carried ▲ and ▼ all along, but the
  side padding every button inherits left two pixels for the glyph and Qt cut
  it away. They now use their own narrow style.
• Longer explanations live behind a small ⓘ as tooltips instead of standing as
  grey paragraphs under every field, and the page is laid out in two columns:
  what makes up the name on the left, what happens to files that already exist
  on the right. Both together take the scrolling out of the tab.
• Time of day with or without seconds (14-30-15 or 14-30).
• A running counter with a chosen number of digits and a reset interval:
  never, daily, weekly, monthly or yearly. A button sets it back to 1 by hand.
• The calendar week uses the ISO year, so the 1st of January still lands in the
  last week of the year before instead of opening a second folder for the same
  week. KW stays fixed regardless of interface language, for the same reason
  the type short forms do: a language switch must not rename existing files.

Changed

• A Changelog button under Settings → Info. It opens the version history
  in a window of its own, rendered as formatted text rather than raw Markdown.
  The text is fetched from the project page on GitHub, because the copy that
  ships with an installation ends at the version you already have — and what
  changed since your version is the thing you came to read. Without a
  connection, or when the fetch fails, the shipped copy is shown instead and
  the window says which of the two you are looking at.
• The only network connection this program makes. It happens when you
  press that button and never in the background; there is still no update
  check and nothing is sent but the request itself. The fetch runs in a thread
  of its own, so the window opens immediately instead of standing still until
  a timeout runs out. The README statement about network use was corrected
  accordingly.
• The introduction knows about the new tab. Its settings page lists
  Options first and says what it holds, and the sessions page points at
  Settings → Options for the file names now that Choose folder only picks a
  folder.
• Settings tabs reordered to Options, Analysis, Target profiles,
  Design, Info — roughly how often one changes something there. The dialog
  still opens on Info, which is what you want in front of you when you open
  the settings without a particular parameter in mind.
• Choose folder no longer opens a dialog first. It opens the folder dialog
  of the system and nothing else. Asking about the naming scheme when someone
  is looking for a folder was one question too many; the scheme moved to the
  Options tab.
• Existing recordings are moved to the assembled scheme, not just to month
  folders and the type. Whether a name was given by the program or by hand is
  now decided by taking the name apart into its parts, so hand-given names stay
  untouched under any scheme. During a move the counter is handed out per
  period in recording order, not from the stored value — a move brings along
  months that are long over.
• Empty subfolders left behind by a scheme change are cleaned up, not only
  month folders.
• Settings now roll back the recording folder and the naming scheme on
  Cancel, along with everything else. A move is excluded: moved files are not
  coming back, so the move button applies its selection right away.

Migration

• A configuration from 1.1.1 keeps its behaviour: month_folders becomes one
  folder per month, type_in_name switches on the type part. Nothing is
  rewritten until the Options tab is used.

v1.1.1 save folder

[1.1.1] — 2026-09-05

Added

• A place to say where recordings go. Choose folder sits next to Open
  folder under Sessions and opens a small dialog: the folder itself at the
  top, then a checkbox for one folder per month and one for the recording type
  in the file name, with a live example of what a new file will be called. An
  external drive or a synced folder was possible before only by moving the
  whole data directory with DREAM_VOICETRAINING_HOME, which took the settings
  along with it.
• A button that moves what is already there. At the bottom of the same
  dialog: existing recordings are moved into the chosen folder and renamed to
  the chosen scheme. Names given by hand are kept, a file of the same name at
  the destination is never overwritten, and files that cannot be found are
  reported instead of dropped from the list. The setting is only stored once
  the files have actually arrived.
• Recordings can now live in subfolders. sessions.json stores the name
  relative to the recording folder with forward slashes, so the list stays
  usable across systems, and it stays in the program data itself — a list that
  travels with the WAV files onto an external drive is gone the moment the
  drive is not plugged in. Recordings left behind at the old location are still
  found and played.
• The type in the file name uses a fixed English short form — reading, hum,
  vowel-a, vowel-i, vowel-u, free — so switching the interface language
  does not rename anything.

v1.1.0 mikrofon change fix

[1.1.0] — 2026-09-04

Added

• A proper input picker beside the microphone box. The button next to it
  opens a wide list with the full name of every input, the technical name of
  the selected one underneath, and a refresh button. Windows users could not
  read past the middle of names like USB Advanced Audio Device Analoges
  Stereo in a drop-down that had to share the toolbar with everything else.
• The microphone box now carries the full name as a tooltip as well.

Fixed

• The same microphone appearing twice is now explained instead of puzzling.
  It was never a duplicate: a sound card with two capture inputs reports both
  under one description, and every output has its own monitor source. Entries
  that share a name now get the piece of their technical name that tells them
  apart appended — two USB Audio Device become (00) and (01) — and the
  picker says so in a sentence at the bottom. Where the technical names match
  too, the entries are numbered, so the list is never ambiguous.

Changed

• The recording type and target voice boxes no longer take the width of
  their longest entry. A stylesheet rule set a 180 px minimum on every
  drop-down, so Reading text was as wide as Pitch test (hum) and the space
  was missing where it was actually needed. The minimum is now 90 px, the
  microphone box asks for 220 px and grows with the window, and the two short
  boxes are capped.

v1.0.9 windows desktop icon

[1.0.9] — 2026-09-03

Added

• A downloaded EXE offers to set itself up on the first start. Anyone who
  takes the plain executable instead of the setup runs it out of the download
  folder, ends up without an icon anywhere, and loses the program the next
  time that folder gets tidied. It now asks once — the answer, yes or no, is
  remembered — and on yes copies itself to
  %LOCALAPPDATA%\Programs\Dream-VoiceTraining, puts an icon on the desktop
  and an entry in the start menu, registers under Apps & features, and
  restarts from the new location. The file left behind in the download folder
  is deleted by the restarted copy, because Windows will not let a running
  program delete itself.
• The question is deliberately not asked for the portable build (it is meant
  to stay where it is), for a copy already under Program Files (the setup did
  all of this), or when running from source.

Changed

• The setup now installs to C:\Program Files\Dream-VoiceTraining and
  asks for administrator rights once, instead of installing per user. The
  wizard has a page for the desktop icon and the start menu entry, both ticked
  by default, and puts an Uninstall entry in the start menu beside the
  program. Recordings and settings stay where they are on uninstall — they are
  in %APPDATA% and %LOCALAPPDATA%, never in the install folder.

v1.0.8 windows fix and tutorial

[1.0.8] — 2026-09-03

Added

• The last page of the introduction carries the project links — source
  code and issues, Discord, Ko-fi — each with the address written out
  underneath, so it is clear where a click leads before making it. They were
  only in Settings → Info before, which is not where somebody looks who has
  just finished the introduction and hit a problem.

Fixed

• Those same links on the info page rendered without their colour: the style
  attribute was built inside an f-string as style=f"...", and the stray f
  made the attribute invalid. Both places now build the link from one helper,
  so the next change to it cannot drift apart again.

[1.0.7] — unreleased

Fixed

• Step 9 of the introduction pointed at a button nobody could see. The
  page about the settings marks the ⚙ button, which sits in the Live tab —
  but the page before it switches to Sessions to show the view button, and
  nothing switched back, so the marker hid itself and the step explained a
  button while pointing at nothing. Which tab to open is no longer decided
  from a list of keys: the window looks up which tab actually contains the
  target and brings that one forward. A control moved to another tab now
  keeps working without anyone remembering to update a special case.

v1.0.6 tutorial + info for f1 f2 etc

[1.0.6] — 2026-09-03

Removed

• The guided run is gone. The toggle in the toolbar and the panel in the
  Live tab have both been removed, along with their strings and styles. On some
  setups the countdown could leave the panel hanging: the timer kept the stream
  and the recording state to itself, and once that went wrong there was no way
  out other than restarting the program. Everything it did can be done by hand
  — pick a type, record, trim in Advanced — so a feature that freezes is
  worse than no feature.

Added

• A recommended first round in the first-run introduction, taking over what
  the guided run used to explain: pitch test (a held hum, about four seconds),
  then /a/, /i/ and /u/ at about three seconds each, every one as its own take
  with its own type, plus the hint to select the calm middle in Advanced
  afterwards. Reachable again at any time from Settings → Info → Show
  introduction again.
• A pulsing golden ⓘ points at the control each page is about — the
  microphone list, the type box, the ⓘ in the toolbar, the Sessions tab, the
  view button. It sits on the real main window rather than on a picture of it,
  ignores mouse clicks so the control underneath stays usable, and survives a
  language switch. Pages about the detail view, which is not open at that
  moment, carry a screenshot with the same marker on the spot that matters.
• Four new introduction pages: what F0, F1 and F2 mean and where the two
  ⓘ buttons explain them; advanced mode, with the waveform, the draggable
  region and why cutting out the calm middle is the whole point; making the
  column layout yours, covering the header menu, dragging columns and the view
  dialog; and what sits behind the settings button, tab by tab.
• packaging/make-intro-shots.py renders the introduction screenshots from the
  real interface with invented data, once per language. The marker positions
  fall out of it: they are computed from where the tab, the Details button, the
  row ⓘ, the Advanced toggle and the analyse button actually sit, and are
  written to assets/intro/shots.json. Hand-taken screenshots go stale and are
  always in one language.
• Screenshots for those pages ship in assets/intro and are installed by all
  four packaging paths. If they are missing, the pages stay readable without
  pictures instead of failing.

Changed

• Picking a language on the first page moves straight on. That page has one
  job; making people confirm it with Next afterwards was a click for nothing.

• The microphone page now says what to do in one line: pick the microphone from
  the list, press Start, speak normally, watch the level card.

• The introduction window sizes itself to the page. One fixed size for all
  of them meant a page with three paragraphs sat in a window built for a
  screenshot, three quarters empty. Text pages now open at roughly 760×380,
  screenshot pages as wide as the picture needs. Heading and text stand in
  their own 700 px column, so line length stays readable under a wider image.

• The introduction shows screenshots in the interface language, and swaps
  them when the language changes rather than keeping the pictures it was built
  with.

• The settings dialog opens on Info. That tab holds the version, the links,
  the debug window and the button that shows the introduction again — what
  people are looking for when they open settings without a specific parameter
  in mind.

• Built-in analysis templates are no longer named in German in the English
  interface. They carried their German names in the code, so Leises
  Mikrofon and Formantmessung (Vokal halten) showed up untranslated. They
  now have keys and translated labels; an old name in an existing config.json
  is migrated on load, so a saved selection survives the update.

• The About dialog is now an Info tab in the settings, next to Design.
  Version, links, licence, the note on network use and the file locations all
  live there, and it is reading material rather than a window that wants
  something, so a tab fits it better than a dialog.

• The buttons that used to crowd the settings button row moved to the top of
  that tab: show introduction again, debug window, copy system info and
  open folder. The button row is back to OK, Cancel and Apply.

• The debug button keeps its error counter and turns red there just as before.

[1.0.5] — unreleased

Added

• First-run introduction. On the very first start the program asks for a
  language and then walks through five short pages: set your level, pick a
  recording type, where sessions and details live, where to look things up,
  and one page on not training through pain. It is not modal, so the
  microphone can be set up while reading. Reachable again at any time from
  Settings → Show introduction again.

Changed

• Windows builds are now two single files. Dream-VoiceTraining.exe for
  the installer and Dream-VoiceTraining-Portable.exe for carrying around.
  Neither needs an _internal folder or anything else beside it — one file is
  the whole program.
• The portable build keeps settings and recordings in a
  Dream-VoiceTraining-Data folder next to the executable instead of under
  %APPDATA%. It recognises portable mode by the file name, or by a
  portable.txt placed beside it if the executable was renamed. If that
  folder cannot be written — a read-only drive, or Program Files — it falls
  back to the normal user folders rather than failing to start.

v1.0.4

[1.0.4] — 02.09.26

Added

• Reference window behind the ⓘ in the toolbar. Nineteen topics in six
  sections explaining what every number in the program actually is: pitch and
  its percentiles, intonation, the formants F1 to F3 and what physically moves
  them, H1/H2 and weight, harmonicity, jitter and shimmer, how to read the
  spectrogram, what the recording types are for, how target profiles and the
  analysis parameters work, and how to get a first measurement that is worth
  anything.
• The window is not modal: it stays open beside the main window, so you can
  look something up while recording rather than instead of recording. Pressing
  ⓘ again brings the existing window forward instead of opening a second one.
• Full-text search across titles, bodies and keywords in the active language.
• Written in English and German, switching with the interface.

[1.0.3] — 02.09.26

Added

• Windows support. Settings go to %APPDATA%\Dream-VoiceTraining,
  recordings to %LOCALAPPDATA% — deliberately the local branch, or a roaming
  profile would drag hundreds of WAV files across the network at every login.
• Device list works without pactl. Windows reports the same microphone once
  per host API; the WASAPI entry is kept, the rest dropped, and the API is
  shown in the label so the choice stays traceable. "Stereo Mix" and
  "What U Hear" are classified as monitors, VB-Cable and Voicemeeter as
  virtual sources.
• Build scripts under packaging/windows: a PyInstaller spec that bundles the
  PortAudio DLL shipped with sounddevice and drops the Qt modules the program
  never uses, a PowerShell script producing a portable ZIP, and an Inno Setup
  installer that installs without administrator rights and leaves recordings
  and settings alone when uninstalling.
• Segoe UI added to the font stack.

Fixed

• Spin box arrows sat on top of the number rather than beside it, so
  clicking the up arrow put the cursor in the field instead of raising the
  value. Styling a QSpinBox at all makes Qt stop drawing its sub-controls
  in the right place, and the border-triangle trick from CSS does nothing
  there — Qt wants an image. The arrows are now generated as SVG in the
  theme's text colour and regenerated when the theme changes. On Linux the
  native style happened to hide the problem; it only showed on Windows.

Note

The Windows build is untested by the author of this changelog entry — it was
written on Linux and could not be executed there. Treat 1.0.3 on Windows as a
first attempt and report what breaks.