Purfview

- Overall slight quality improvement (no metrics update); - Higher stability on OOD / rare / strange / unique data; - Significant quality improvements on various known edge cases: -...

Remove "local_dir_use_symlinks" from download_model() It's deprecated since huggingface_hub v0.23.0 and produce this warning: > /opt/hostedtoolcache/Python/3.9.24/x64/lib/python3.9/site-packages/huggingface_hub/utils/_validators.py:202: UserWarning: The `local_dir_use_symlinks` argument is deprecated and ignored in `snapshot_download`. Downloading to a local directory...

Adds new VAD parameters: min_silence_at_max_speech: Minimum silence duration in ms which is used to avoid abrupt cuts when max_speech_duration_s is reached. use_max_poss_sil_at_max_speech: Whether to use the maximum possible silence at...

Update to CUDA 12.8

8

Update Windows and Linux from CUDA 12.4 to CUDA 12.8 Upgraded manylinux2014 to manylinux_2_28

Update Docker to Cuda 12.8

3

Update Docker to Cuda 12.8

Fixes inconsistent VAD defaults in BatchedInferencePipeline Bug: BatchedInferencePipeline applies inconsistent defaults for `min_silence_duration_ms`. If `vad_parameters` is not passed, the value defaults to `160`. When e.g. only `threshold` is passed, it...

Adds optional batched VAD feature

1

Optional batched VAD feature. Prompted by https://github.com/SYSTRAN/faster-whisper/issues/1388 [User wanted GPU option because it's ~5x faster in slow CPU/fast GPU env, this PR should give similar speed increase on CPU] Enabled...

VAD Fix: explicit safe context, reduced RAM usage

3

VAD Fix: explicit safe context (no view mutation). Reduced SileroVADModel RAM usage by 9% [on 4h audio]. PR doesn't affect VAD speed nor probs output in current FW VAD functionality....