Offline Transcriber v1.3.6

A cross platform offline audio transcription tool with both console and GUI interfaces, built with Python. It supports a wide range of common audio formats and can optionally translate transcripts into English. Transcribe audio locally without requiring an internet connection (open source models), with support for macOS and Windows.

This tool is especially useful for:

• Students and academic researchers transcribing interviews
• Journalists handling sensitive recordings
• Professionals working with confidential audio data
• Anyone who needs privacy-preserving offline speech-to-text transcription

---

Features

• Optional GUI (Graphical User Interface) for easy use
• Fully offline processing after setup
• Optional translation (-translate)
• Multi-model support (small, medium, large)
• Smart paragraphing
• Dynamic paragraph threshold by model
• Long-file chunking
• Resume support
• Subtitle export (.srt, .vtt)
• Cross-platform (Windows / macOS)
• Interactive installer with model selection
• Detailed logs
• MIT License

---

Privacy & Data Handling

All processing is performed locally on the user's machine. No audio, transcript, or metadata is sent, stored, or shared externally.

Internet access is only required:

• during installation
• when downloading models (initial setup or upgrades)

---

Installation for GUI usage

macOS

./install_transcriber_gui_mac.sh

Double click the Transcribe_GUI.command on your Desktop to open the Transcriber GUI.

Windows

Install__GUI.bat
Transcribe_GUI.bat

Double click the Transcribe on your Desktop to open the Transcriber GUI.

---

Installation for CLI (Console) Usage

macOS

cd /path/to/project/Scripts
chmod +x *.sh
./install_transcriber_mac.sh

Windows

install_transcriber.bat

---

Usage

Before running the tool, place your audio files inside the Input folder in the project directory.

macOS

./transcribe_mac.sh
./transcribe_mac.sh -single audio.mp3
./transcribe_mac.sh -single audio.mp3 -translate
./transcribe_mac.sh -single audio.mp3 -large
./transcribe_mac.sh -archive
./transcribe_mac.sh -force
./transcribe_mac.sh -preprocess

Windows

transcribe.bat
transcribe.bat -single audio.mp3
transcribe.bat -single audio.mp3 -translate
transcribe.bat -single audio.mp3 -large
transcribe.bat -archive
transcribe.bat -force
transcribe.bat -preprocess

---

Parameters

• -translate enable translation
• -small use the small model
• -medium use the medium model
• -large use the large model
• -accurate alias for large
• -archive move processed files to Archive
• -force reprocess existing outputs
• -single <file> process one file
• -preprocess enable audio preprocessing
• --init initialize project structure
• -config show current configuration

---

Subtitle Export

The tool supports subtitle export in .srt and .vtt formats.

Examples:

./transcribe_mac.sh -single audio.mp3 -srt
./transcribe_mac.sh -single audio.mp3 -translate -srt
./transcribe_mac.sh -single audio.mp3 -vtt
transcribe.bat -single audio.mp3 -srt
transcribe.bat -single audio.mp3 -translate -srt
transcribe.bat -single audio.mp3 -vtt

When -translate and -srt are used together, both are generated:

• original transcript/subtitle
• English translation

---

Installing Additional Models Later

If you need additional models after installation, use the upgrade scripts while connected to the internet.

macOS

./upgrade_models_mac.sh
./upgrade_models_mac.sh -large
./upgrade_models_mac.sh -medium -large

Windows

upgrade_models.bat
upgrade_models.bat -large
upgrade_models.bat -medium -large

If a required model is missing, the program will show a clear instruction.

---

Performance and Reliability Improvements

Version v1.3.6 adds several internal improvements while keeping the tool simple to use:

• Full GUI implemented for Offline Transcriber – users can now interact with the app via an intuitive interface instead of command-line.
• Added “Translate without transcription” option with safe toggle logic.
• Favorite Config can now be saved and loaded correctly.
• Reset Default Config restores all settings to initial defaults.
• Checkboxes visually indicate disabled state with dimmed text for clarity.
• Automatic backend selection: CUDA/GPU is used when available; otherwise CPU is used.
• CUDA mode uses float16 and batched inference when supported.
• CPU mode stays conservative with int8.
• -translate-only can be used when you only need an English translation and do not need the original transcript, avoiding the extra transcription pass.
• --validate checks configuration without processing audio.
• --self-test runs basic internal checks.
• --benchmark reports the planned backend and audio duration without running full transcription.
• Chunk extraction now uses mono 16 kHz WAV to reduce temporary file size and I/O.

Examples:

./transcribe_mac.sh --validate
./transcribe_mac.sh --self-test
./transcribe_mac.sh --benchmark
./transcribe_mac.sh -single interview.mp3 -translate-only
transcribe.bat --validate
transcribe.bat --self-test
transcribe.bat --benchmark
transcribe.bat -single interview.mp3 -translate-only

---

Offline-Safe Model Loading

During transcription, the tool will not download models automatically. If a model is not available locally, the program stops and instructs you to run the upgrade script.

This ensures:

• predictable behavior
• full offline operation
• maximum data privacy

---

Notes for Windows / Parallels

Running inside virtual environments (e.g. Parallels) may reduce performance. This release includes workarounds for common OpenMP runtime conflicts.

---

Accuracy Disclaimer

Transcription and translation quality depend on:

• audio quality
• speaker clarity
• accent / dialect
• background noise
• speaking speed

No automated system guarantees perfect accuracy. Users should review outputs before using them in critical contexts.

---

License

MIT License

---

Author

Abbas SALAMAT Abbas.salamat@edu.donau-uni.ac.at

Suggestions, improvements, bug reports, and contributions are welcome.