Spaces:

---

ImPolymath

/

demorrha

like 0

Paused

Changes in the new version - 1.3.9

1. Pages and Navigation:
   • The app.py file has been updated to incorporate multi-page navigation, featuring two main sections: "Configuration" and "Translator". The configuration part now includes a new page (pages/configuration_ui_lang.py) for selecting the interface language, while the main page (pages/main.py) handles translation and user interaction.
   • The application structure has been clarified by separating interface configuration (language selection, STT/TTS settings) from the message processing on the main page.

2. User Interface Enhancements:
   • In pages/main.py, the main_page function has been improved to support multiple input methods: text input (using st.chat_input), audio input (with st.audio_input and a recording widget), and file uploads (managed via st.file_uploader).
   • Distinct tabs (text_input, audio_input, file_upload_input) have been created to organize these different input modes.
   • The audio and text-to-speech parameters are encapsulated within dialog boxes (st.dialog) for configuring STT and TTS settings.
   • Fine session management is implemented using st.session_state to hold and manipulate session information, such as selected languages, uploaded files, transcriptions, and generated responses.

3. Language Management and Translation:
   • Language selection and management are now centralized: the var_app.py file defines a list of supported languages and their corresponding emojis, while pages/configuration_ui_lang.py offers an interface to set the interface language. Translations are loaded from a JSON file (ui_lang_support.json).
   • The init_langs_for_processing function (in pages/main.py) and functions in core/core.py initialize a processing mode for translation by configuring both the system prompt and the operational prompt.

4. Audio Processing and Transcription:
   • In core/speech_to_text.py, the module now includes key functions like huggingface_endpoints_stt and transcribe_audio, which handle transcription via Hugging Face endpoints and segment audio files that exceed the 25 MB limit.
   • Audio segmentation logic is integrated (based on file size and audio properties) to facilitate processing of lengthy audio files.
   • An option for voice isolation (using the isolate_audio function) is provided before transcription, based on the user’s choice.

5. Assistant Integration and Streaming Responses:
   • In core/core.py and core/demorrha.py, the DemorrhaAssistant class has been enhanced to manage the translation assistant. This class handles the creation, search, and update of the assistant via the OpenAI API.
   • A vectorization logic is introduced to improve search and provide personalization according to style or context.
   • The use_assistant method employs a streaming thread that gradually returns the assistant’s response. This response is collected by a generator and displayed in real time within the Streamlit interface.

6. Various Enhancements and File Management:
   • Utility functions such as save_attachment, hash_file, and callback_change_edited_text have been introduced for handling and saving files (both text and audio).
   • The README.md file has been updated to reflect the current state and new functionalities, including multi-language support, audio input options, and streaming responses.
   • The requirements.txt file now specifies the exact dependencies (notably streamlit, openai, pydub, python-dotenv, and elevenlabs), ensuring consistency in the development environment.

Overall, the latest merges have significantly improved Demorrha's application architecture by better separating responsibilities (interface configuration, input processing, transcription, translation, text-to-speech, and assistant management). The user interface is now more intuitive with dedicated tabs and dynamic settings management, while robust integration with OpenAI and Hugging Face APIs ensures reliable real-time audio and text processing.