fix: use langchain-text-splitters/langchain-core (langchain 1.x removed shims)

feat: add X-ASR + Qwen3-ASR backends, Gemma-4 LLMs; fix review findings

feat: Improve summarization quality and eliminate repetition

Implement multilingual summarization by only touching system prompts

Add speaker name detection feature

Fix llama-cpp-python blocking on HF Spaces due to excessive threads

Apply zh-cn to zh-tw conversion on generated titles

Refine title generation prompt and parameters

Implement document title generation and UI improvements

Convert summarization.py prompts and comments from French to English

Add langchain dependency for summarizer module

refactor: simplify and robustify summarization module

fix: handle long transcripts in summarization by chunking

re-implement in FastAPI

Fix memory exhaustion on HF Spaces free tier by reducing n_ctx and implementing chunked summarization for long transcripts

Optimize speaker diarization performance for long audio files

work on llama.cpp huge loading time

Update src/summarization.py

Update src/summarization.py

Create summarization.py