v1.2.0

Added

• Input method and output method choice at the start of every session,
  asked right after native language and saved to config.<hostname>.ini
  alongside your other picks. Input can be voice (microphone + speech
  recognition, the previous default) or text (typed at a prompt); output
  can be voice (spoken aloud) or text (printed only). Choosing text for
  either side skips its device-selection prompt and, at startup, its model
  preload (Whisper for text input, the TTS voice model for text output),
  so a text-only session starts faster and needs no microphone or speaker.

Changed

• In role-play mode, the tutor now introduces herself by name and role
  together once the scenario starts, so her fixed tutor persona and the
  in-scene character (e.g. a waiter, a shopkeeper) stay linked from the
  first line instead of only being connected implicitly.

v1.1.0

Added

• Practice-mode selection at the start of every session: free conversation,
  role-play, vocabulary quiz, translation practice, or a custom goal. In
  role-play mode the tutor asks the student for a scenario and which of the
  two roles they want to play, then stays in character for the rest of the
  session. Built-in modes live in the new practice_modes.py; a custom goal
  is read from a .md file the student drops in custom_goals/ (see
  custom_goals/put_custom_goal_files_here.txt), listing the 5 most
  recently modified files to choose from. The choice is asked fresh on
  every startup and is never saved to config.<hostname>.ini. Custom goal
  files can reference {NATIVE_LANGUAGE}/{TARGET_LANGUAGE}, substituted
  with the session's actual languages, and can be as short as one line or a
  full multi-paragraph drill with its own instructions -- the goal's text is
  given to the tutor on its own line rather than wrapped inline in quotes, so
  longer, structured goals stay readable in the assembled prompt.
• --practice-mode <mode> and --custom-goal <filename> flags for both
  main.py and llm_engine.py's standalone CLI, for scripted/non-interactive
  runs. Passing --practice-mode skips main.py's interactive picker
  entirely; an invalid mode, or --practice-mode custom without a
  --custom-goal naming an existing file in custom_goals/, prints a clear
  error and exits immediately instead of falling back silently. Passing
  --custom-goal alone, with no --practice-mode, implies custom mode.
• --native-lang <language> flag for llm_engine.py's standalone CLI,
  mirroring --lang-target/--level.
• --debug flag for main.py, enabling debug-level logging across all
  engines (including the full assembled tutor system prompt on every turn),
  and for llm_engine.py's, tts_engine.py's, and rag_engine.py's own
  standalone CLIs.

Full Changelog: v1.0.0...v1.1.0

v1.0.0

First stable release. LangTeacher is a voice-based language tutor that runs entirely on your own hardware, verified on macOS, Windows, Linux, and Raspberry Pi 5+.

Changes since v0.1.0

• Split requirements for a lighter Raspberry Pi install
• Resample audio to the output device's native rate before playback
• Route STT test playback through a chosen device, add --debug flag
• Suffix log/transcript filenames with the machine's hostname
• Fix log encoding by specifying UTF-8 for log files and settings
• Fix missing vocabulary translations for genderless target languages
• Fix nonsensical vocabulary quiz prompts asking to translate a word already given in the target language
• Suppress noisy but non-fatal console output on the Pi
• Add CHANGELOG.md, version.py, --version flag, and bump_version.py
• Documentation updates (Raspberry Pi push-to-talk/Wayland limitation, stt_engine.py flags)
• Minor bug fixes and cleanup

v0.1.0

Full Changelog: https://github.com/janosdios/langteacher/commits/v0.1.0