Alxalexandru/romanian-tts-stt

Romanian TTS/STT Application

A Gradio web application for Romanian language speech processing:

• —Speech to Text (STT): Transcribe Romanian audio using Whisper
• —Speaker Diarization: Identify different speakers in audio
• —Text to Speech (TTS): Generate Romanian speech from text

Features

Speech to Text

Upload an audio file or record using your microphone to transcribe Romanian speech. Uses the gigant/whisper-medium-romanian model with 4.73% WER.

Speaker Diarization

Identify who spoke when in multi-speaker audio. Combines speaker detection with transcription to produce labeled output. Uses pyannote/speaker-diarization-3.1.

Text to Speech

Convert Romanian text to natural-sounding speech. Uses facebook/mms-tts-ron for fast CPU inference.

Setup

Local Development

1. 1.Install dependencies:

pip install -r requirements.txt

1. 1.Set your Hugging Face token (required for diarization):

export HF_TOKEN=your_token_here

1. 1.Run the application:

python app.py

Hugging Face Spaces

1. 1.Create a new Space with Gradio SDK
2. 2.Upload all files
3. 3.Add HF_TOKEN as a secret in Space settings
4. 4.The app will start automatically

Models

Requirements

• —Python 3.9+
• —See requirements.txt for full dependency list
• —Hugging Face token with access to pyannote models (for diarization)