A private Bulgarian transcription workspace with browser recording, Whisper processing, transcript/email attachments and companion calendar/task tools.
Scribe · Bulgarian Transcription↗
Scribe — browser interface for Bulgarian audio transcription.
A CLOSER LOOK
See the workflow.
Illustrated workflow · sample interface. The application requires authorised access.
01 / 03
SCRIBEWORKFLOW ILLUSTRATION
01 / BROWSER
A voice, captured.
Microphone → browser audio
Illustrative explanation · no live connection
01Record in the browser
SCRIBEWORKFLOW ILLUSTRATION
02 / PRIVATE SERVER
Audio becomes language.
FFmpeg16 kHz · mono→WhisperBulgarian
FastAPI · Python · PyTorch
Illustrative explanation · no live connection
02Process on a private server
SCRIBEWORKFLOW ILLUSTRATION
03 / TRANSCRIPT
Ready for the next task.
Репетицията започва в десет часа. Подгответе програмата за концерта.
Example text · not transcribed from a recording
Illustrative explanation · no live connection
03Use the transcript
An authorised user records audio with recording and processing feedback.
The FastAPI/Whisper service prepares Bulgarian transcription. This portfolio walkthrough does not record or transmit audio.
Transcript history and staff-directed email connect the result to everyday workflows.
About the project
Scribe is a web application at scribe.vso.bg that connects to a private server for AI audio transcription into Bulgarian text.
The project provides a browser-based entry point to a privately hosted transcription workflow.
INSIDE THE PROJECT
Features & capabilities
Explore the tools and workflows built into this project.
EXPLORE BY CATEGORY
01CAPABILITIES01 / 09
Browser recording and Bulgarian transcription
Start/stop microphone recording from the browser, with an audio-level waveform and recording/processing/ready feedback.
Send recordings to a FastAPI backend for Bulgarian-focused Whisper transcription; the API also accepts a language parameter.
Microphone use requires permission and a supported browser, with HTTPS required for production. Transcription needs a reachable backend; it does not run offline in the browser.
02CAPABILITIES02 / 09
Private audio processing
PyTorch selects CUDA when available and otherwise uses CPU. The Whisper model is configurable, with large-v3 as the inspected default.
FFmpeg prepares 16 kHz mono audio with loudness normalization before recognition.
Decode settings such as temperature, beam search and word timestamps can be configured on the server. Audio and transcription records are stored with database metadata.
03CAPABILITIES03 / 09
History, playback and downloads
Review recent recordings and the stored transcription history, including timestamps and email-delivery fields.
Open transcript text, replay associated audio in the voice interface and download transcript/audio files through backend endpoints.
Calendar events and tasks can retain source-audio/transcript references for later review.
04CAPABILITIES04 / 09
Email messages and attachments
Bulgarian spoken commands can route a message to a named employee or all active employees.
The parser extracts recipient/content information and removes command wording from the message body.
SMTP delivery supports text, generated PDF and audio attachments. ReportLab prepares transcript PDFs, and responses/history record email results and recipients.
05CAPABILITIES05 / 09
Bulgarian voice commands
Recognize intents for sending messages, reading notes/tasks, adding calendar events or tasks, querying events/tasks, deleting events and reassigning tasks.
Extract employees, participants, dates, periods and spoken time ranges, including today/tomorrow, weeks/months and named dates.
These operational commands use the implemented command parser; they are not a general-purpose conversational chatbot. Event deletion has confirmation and multiple-match handling in the voice interface.
06CAPABILITIES06 / 09
Calendar workspace
A Bulgarian FullCalendar interface provides month/week/day views, event creation/editing, date/time, location, participants and all-day events.
Drag or resize events to adjust their schedule. The backend supports event lookup, update and deletion, with associated voice/audio references where supplied.
The companion calendar module includes account- and creator-based permission checks for editing/deleting events.
07CAPABILITIES07 / 09
Task management
A three-column board separates pending, in-progress and completed tasks, with employee assignment, deadlines and overdue indicators.
Four priorities, filters, status editing, task creation/update/deletion and voice reassignment support day-to-day work.
The companion task module checks administrator/creator/assignee permissions and can restrict an assignee to status-only changes. Destructive task actions have confirmation in the interface.
08CAPABILITIES08 / 09
Daily notes and staff directory
Browse dated notes through month/day navigation, see days with notes and create, manually save or delete note text.
Query notes and tasks by employee/time period through the voice workspace and administrator text-query tools.
Employee records hold names, email addresses and active status for command matching and distribution, with management API operations and directory controls.
09CAPABILITIES09 / 09
Private web architecture
PHP login/session interfaces and MySQL storage accompany Python/FastAPI routes protected by an API key.
The voice and calendar modules include administrator/user distinctions and employee associations. Responsive Bulgarian interfaces provide modals, status messages and audio controls.
Model, database, mail and server settings are configurable for the private deployment.