diff --git a/sidecar/pyproject.toml b/sidecar/pyproject.toml index 1df90d92..71c60c94 100644 --- a/sidecar/pyproject.toml +++ b/sidecar/pyproject.toml @@ -18,8 +18,8 @@ dependencies = [ "httpx>=0.27", # HTTP client for the llama.cpp OpenAI-compatible server (§7) # P2 (T2, A6 lesson 5: PINNED). ⚠ human: verify current releases at install # time. chatterbox-tts/torch must NOT be added here (isolated env asset only). - "kokoro-onnx==0.5.0", # A4: the onnx build — NEVER the torch `kokoro` package - "onnxruntime==1.28.0", # kokoro-onnx backend (or onnxruntime-gpu, same pin) + "kokoro-onnx==0.6.1", # A4: the onnx build — NEVER the torch `kokoro` package + "onnxruntime==1.29.0", # kokoro-onnx backend (or onnxruntime-gpu, same pin) "edge-tts==7.2.8", # hosted TTS engine (network only at runtime) # F3: PRIMARY system-RAM source for the device advisor (cross-platform), with # an OS-stdlib fallback (Win32 GlobalMemoryStatusEx / POSIX sysconf) behind the @@ -59,7 +59,7 @@ reframe-gpu = [ "safetensors==0.8.0", # verify-before-load safetensors weight reader (B4/I2) "opencv-python-headless==4.14.0.94", # frame I/O + face-crop ops (headless: no GUI libs) "python_speech_features==0.6", # MFCC features for the ASD audio frontend - "scipy==1.18.0", # face-track interpolation + median filter + wav read + "scipy==1.18.1", # face-track interpolation + median filter + wav read # DIARIZE stage (VAD + ECAPA-TDNN) — the speaker turns the director fuses with # visual ASD. Uses the ``speechbrain.inference`` API (requires >= 1.0.0). PINNED to # 1.1.0: loads ``speechbrain/vad-crdnn-libriparty`` + ``spkrec-ecapa-voxceleb``, and @@ -70,7 +70,7 @@ reframe-gpu = [ # bare ``revision=``). See ``pinned_fetch_kwargs``. # # HISTORY — this comment used to say "PINNED to 1.0.3 ... 1.1.0 is DELIBERATELY - # AVOIDED", six lines above a line that reads ``speechbrain==1.1.0``. The prose was + # AVOIDED", six lines above a line that reads ``speechbrain==1.1.1``. The prose was # the stale half: the codebase was later adapted TO 1.1.0 (the FetchConfig probe and # the two-shape ``pinned_fetch_kwargs``) and the note was never updated, leaving the # file contradicting itself and three docs repeating the dead value. @@ -84,7 +84,7 @@ reframe-gpu = [ # # speechbrain missing at runtime still fails LOUD (typed # ``DiarizeBackendUnavailableError``), never a silent ModuleNotFoundError. - "speechbrain==1.1.0", + "speechbrain==1.1.1", # speechbrain 1.0.x calls ``hf_hub_download(use_auth_token=...)``, an argument # REMOVED in huggingface_hub 1.0. Cap below 1.0 so the diarize models actually # fetch/load (validated at 0.36.2, the newest 0.x). huggingface_hub is otherwise a diff --git a/sidecar/requirements.lock.txt b/sidecar/requirements.lock.txt index 04891a17..f042c1a4 100644 --- a/sidecar/requirements.lock.txt +++ b/sidecar/requirements.lock.txt @@ -25,7 +25,7 @@ certifi==2026.7.22 # edge-tts # httpcore # httpx -click==8.4.2 +click==8.5.0 # via # huggingface-hub # scenedetect @@ -41,7 +41,7 @@ colorlog==6.12.0 # via kokoro-onnx csvw==4.1.0 # via segments -ctranslate2==4.8.1 +ctranslate2==4.8.2 # via faster-whisper dlinfo==2.0.0 # via phonemizer-fork @@ -51,7 +51,7 @@ espeakng-loader==0.2.4 # via kokoro-onnx faster-whisper==1.2.1 # via media-studio-sidecar (sidecar/pyproject.toml) -filelock==3.32.2 +filelock==3.32.5 # via huggingface-hub flatbuffers==25.12.19 # via onnxruntime @@ -71,26 +71,26 @@ httpx==0.28.1 # via # media-studio-sidecar (sidecar/pyproject.toml) # huggingface-hub -huggingface-hub==1.26.0 +huggingface-hub==1.29.0 # via # faster-whisper # tokenizers humanfriendly==10.0 # via coloredlogs -idna==3.18 +idna==3.19 # via # anyio # httpx # yarl isodate==0.7.2 # via csvw -joblib==1.5.3 +joblib==1.6.0 # via phonemizer-fork jsonschema==4.26.0 # via csvw jsonschema-specifications==2025.9.1 # via jsonschema -kokoro-onnx==0.5.0 +kokoro-onnx==0.6.1 # via media-studio-sidecar (sidecar/pyproject.toml) language-tags==1.3.1 # via csvw @@ -104,14 +104,14 @@ multidict==6.7.1 # via # aiohttp # yarl -numpy==2.5.1 +numpy==2.5.2 # via # ctranslate2 # kokoro-onnx # onnxruntime # opencv-python # scenedetect -onnxruntime==1.28.0 +onnxruntime==1.29.0 # via # media-studio-sidecar (sidecar/pyproject.toml) # faster-whisper @@ -124,17 +124,17 @@ packaging==26.3 # onnxruntime phonemizer-fork==3.3.2 # via kokoro-onnx -platformdirs==4.11.0 +platformdirs==4.11.7 # via scenedetect propcache==0.5.2 # via # aiohttp # yarl -protobuf==7.35.1 +protobuf==7.36.1 # via onnxruntime psutil==7.2.2 # via media-studio-sidecar (F3 — primary system-RAM source) -pygments==2.20.0 +pygments==2.21.0 # via rich pyparsing==3.3.2 # via rdflib @@ -152,7 +152,7 @@ referencing==0.37.0 # via # jsonschema # jsonschema-specifications -regex==2026.7.19 +regex==2026.9.3 # via segments rfc3986==1.5.0 # via csvw @@ -187,7 +187,7 @@ tqdm==4.70.0 # faster-whisper # huggingface-hub # scenedetect -typer==0.27.1 +typer==0.27.2 # via huggingface-hub typing-extensions==4.16.0 # via diff --git a/sidecar/runtime_setup/requirements-sidecar.txt b/sidecar/runtime_setup/requirements-sidecar.txt index ca858570..03b0dd57 100644 --- a/sidecar/runtime_setup/requirements-sidecar.txt +++ b/sidecar/runtime_setup/requirements-sidecar.txt @@ -22,29 +22,29 @@ # lock is validated by media_studio.assets.manager.validate_hashed_lock first. faster-whisper==1.2.1 -ctranslate2==4.8.1 +ctranslate2==4.8.2 scenedetect==0.7.1 opencv-python==4.14.0.94 httpx==0.28.1 -numpy==2.5.1 +numpy==2.5.2 av==17.1.0 -onnxruntime==1.28.0 -huggingface_hub==1.26.0 +onnxruntime==1.29.0 +huggingface_hub==1.29.0 tokenizers==0.23.1 # TTS (kokoro default engine — the onnx build, NEVER the torch pip package; A4) # CONTRACT-NOTE: kokoro-onnx is not in the dev venv yet (T2 lane); 0.4.9 is the # pinned choice per the T5 brief ("latest pinned by exact version you choose"). -kokoro-onnx==0.5.0 +kokoro-onnx==0.6.1 # CUDA runtime wheels for ctranslate2/onnxruntime GPU paths (pins = dev venv) nvidia-cublas-cu12==12.9.2.10 -nvidia-cudnn-cu12==9.24.0.43 +nvidia-cudnn-cu12==9.25.1.1 # DELIBERATELY NOT HERE: mediapipe. The claudeshorts reframe engine's optional # MediaPipe backend uses the LEGACY Solutions API (`mp.solutions.face_detection`), # which only exists in mediapipe <=0.10.21 — and those wheels declare `numpy<2`, -# a hard conflict with the `numpy==2.5.1` above (required by ctranslate2 / av / +# a hard conflict with the `numpy==2.5.2` above (required by ctranslate2 / av / # onnxruntime / kokoro-onnx). mediapipe >=0.10.31 (numpy-2 clean) REMOVED the # Solutions API entirely, so pinning it would install a mediapipe whose # `mp.solutions` is missing -> the backend crashes at job time -> a SILENT drop