talkbank-tools

Batchalign3 is the TalkBank audio and machine-learning pipeline for creating and enriching CHAT transcripts. It handles transcription, forced alignment, neural morphotagging, translation and utterance segmentation. The repository hosts the `batchalign3` command-line tool, its Python package, a PyO3 bridge, a dashboard web UI and an experimental desktop shell. The CHAT format itself — grammar, specification, tree-sitter parser, data model, validation, the `chatter` CLI and the CLAN command reference — lives in the separate TalkBank `chatter` repository. This project consumes chatter's published crates rather than being a CHAT-format toolchain itself. Installation is via GitHub releases only; there is no PyPI package. Shell and PowerShell one-liners fetch an installer that bootstraps `uv` if absent, installs `batchalign3` into an isolated environment with a uv-managed Python (3.13 by default, overridable via `BATCHALIGN3_PYTHON`), and upgrades on re-run. The first install downloads large ML dependencies such as PyTorch, Stanza and Whisper. Double-click `.command` and `.bat` wrappers under `installers/` run the same installer. Typical commands include `batchalign3 transcribe recordings/ -o transcripts/ --lang eng`, `batchalign3 align corpus/ -o aligned/` and `batchalign3 morphotag corpus/ -o tagged/`. A full CLI reference is provided in the mdBook. Supported platforms are macOS (Apple Silicon and Intel), Linux (x86_64 and aarch64) and Windows (x86_64), with Python 3.13 or 3.14. The documentation notes several GB of disk for ML models downloaded on first use, 8 GB RAM minimum and 16 GB recommended, and FFmpeg needed only for some media formats. Documentation is an mdBook under `book/`, covering installation, quickstart, the Batchalign3 user guide, architecture and developer guidance, buildable with `make book` and servable with `make book-serve`. The repository map includes the `batchalign/` Python package, Rust crates for the pipeline (ASR, forced alignment, morphotagging, jobs/runner, store, local server control plane, dashboard API), Batchalign-specific CHAT transforms, the PyO3 bridge and wheel build surface, shared types, a React dashboard frontend, a Tauri desktop shell, installers, the mdBook and repo-level reference notes. CHAT core crates are consumed from the public `chatter` repository via git-tag dependencies. Development tasks are exposed through a Makefile: `make check`, `make test`, `make verify`, `make batchalign-ci-rust`, `make batchalign-ci-python`, `make book`, `make ci-local` and `make ci-full`, with lower-level helpers via `cargo run -q -p xtask -- help`. The project is BSD-3-Clause licensed, copyright Carnegie Mellon University.