fix: WER=0 bug when no segments match in evaluation

feat: integrate domain dictionaries for flower shop ASR

update plan: add domain dictionaries integration strategy

update plan, readme, add word-level alignment to transformers-whisper

feat: word-level speaker alignment in hybrid transcription

fix: add VAD fallback when Silero VAD detects too few speech segments

fix: remove redundant no_speech_prob/avg_logprob post-filters

fix: keep all gold segments for matching, filter non-speakers only in DER

fix: pin numpy<2 for onnxruntime compatibility on T4

fix: 3 critical issues — GPU decorator, gold labels, grid search params

fix: pass string path to gigaam.transcribe() instead of list

fix: pin Python 3.10 to fix torchaudio/gigaam compatibility on ZeroGPU

revert: remove torch/torchaudio from requirements (pre-installed on ZeroGPU)

fix: add torch/torchaudio with cu121 index for dedicated GPU hardware

fix: reduce GPU duration from 300s to 120s to fit ZeroGPU quota

fix: prefetch model weights before GPU task to avoid ZeroGPU abort

feat: benchmark pipeline + grid search + new Russian ASR models

feat: hybrid transcription strategy + tuning improvements

feat: automatic evaluation with Label Studio gold annotations

fix: anti-hallucination filters + speaker count control

fix: improve diarization+transcription pipeline quality

fix: copy audio to /tmp before returning to Gradio to avoid InvalidPathError

fix: add allowed_paths for HF cache

fix: use Systran CTranslate2-converted whisper models for faster-whisper

fix: revert to use_auth_token for pyannote (monkey-patch handles the rest)

fix: patch use_auth_token→token for pyannote + huggingface_hub compat

fix: return error dict from GPU worker instead of raising

fix: wrap _run_single body to expose full traceback from ZeroGPU worker

fix: full onnxruntime stub with InferenceSession and common attrs

fix: add __spec__ and submodules to onnxruntime stub

fix: mock onnxruntime stub to bypass broken C extension on ZeroGPU

fix: add onnxruntime to fix pyannote.audio import

Diagnostics: full traceback

Remove torch/torchaudio to avoid version conflict with preinstalled

Fix pyannote import: add pytorch-lightning speechbrain

Add diagnostics button to check imports

Show full traceback in error messages

Upload only dialog files, skip already uploaded

Add upload_audio.py script for local MP3 upload to Hub

Fix slow startup: lazy-load audio file list

Replace QA dashboard with diarization experiments pipeline

Fix Space README metadata

Add QA dashboard backed by private dataset repo

initial commit