Packages used¶

Pin source of truth is pyproject.toml ([project.dependencies] and [project.optional-dependencies]). This page is a map, not a second pin list.

CVE / waiver policy: dependency_audit.md. Analysis model IDs: models.md. Install profiles: install_profiles_matrix.md.

Core wheel¶

From a TranscriptX git checkout (not on PyPI): pip install -e . (no extras) is the analysis core plus the Streamlit launcher entry point. It deliberately omits compiled NLP / voice / BERTopic stacks so a clean host install is not blocked by llvmlite / CUDA wheels. Exact pins: pyproject.toml [project.dependencies].

The GUI extra is separate: Streamlit is [web], not in [full]. Docker / transcriptx.sh install the GUI via requirements.txt.

Install extras¶

Extra	What it is for
`web`	Streamlit GUI plus in-repo `transcriptx-workspaces` (CCv2). Not included in `[full]`. `file:` path on git checkouts, not PyPI.
`rag`	LanceDB + httpx for the opt-in Ask page. Also in Docker `requirements.txt` and `[full]`.
`nlp` / `ner`	spaCy NER (`ner` is an alias; modules require `nlp`).
`emotion_lexical`	NRCLex lexical emotion.
`emotion_transformers`	Torch + Transformers contextual / fine-grained emotion.
`emotion`	Compatibility union of the two emotion extras.
`voice`	pyannote / openSMILE / librosa analysis audio.
`speaker_match`	SpeechBrain speaker embeddings (separate from analysis `[voice]`).
`keyphrases`	YAKE / KeyBERT (noun-chunks still run without this extra).
`bertopic`	BERTopic + hdbscan + umap-learn. Optional; see bertopic_optional_module.md.
`maps`	Folium / geopy / Playwright for NER map HTML→PNG.
`visualization`	matplotlib, seaborn, wordcloud, ebooklib (Overview EPUB).
`plotly`	Plotly.
`full`	Union of the analysis extras above plus `[rag]` (not `[web]`, `[dev]`, or `[docs]`).
`dev`	pytest, linters, pre-commit, plus matplotlib/seaborn/geopy for smoke tests.
`docs`	Sphinx, MyST, Furo.

Workspace package¶

packages/transcriptx_workspaces is the Streamlit Components v2 package for Theme C (Speaker ID, Corrections review, viewer click-drag) and Theme D (reader_workspace full-file karaoke). The [web] extra installs it from file:packages/transcriptx_workspaces on git checkouts. Docker strips that extra before building the TranscriptX wheel (image pip cannot parse the relative file: URL), installs transcriptx with --no-deps (deps already from requirements.txt), and vendors the workspaces wheel in a later Dockerfile step.

pip install -e ".[web]"

See theme_c_workspaces_ccv2.md.