Metadata-Version: 2.5
Name: voxint
Version: 0.17.0
Summary: From sound to intelligence: end-to-end transcription, diarization, and speaker identity with human-grade quality gates.
Project-URL: Repository, https://github.com/bengizmo/voxint
Project-URL: Issues, https://github.com/bengizmo/voxint/issues
Project-URL: Documentation, https://github.com/bengizmo/voxint/tree/main/docs
Author: Voxint contributors
License: 
                                         Apache License
                                   Version 2.0, January 2004
                                http://www.apache.org/licenses/
        
           TERMS AND CONDITIONS FOR USE, REPRODUCTION, AND DISTRIBUTION
        
           1. Definitions.
        
              "License" shall mean the terms and conditions for use, reproduction,
              and distribution as defined by Sections 1 through 9 of this document.
        
              "Licensor" shall mean the copyright owner or entity authorized by
              the copyright owner that is granting the License.
        
              "Legal Entity" shall mean the union of the acting entity and all
              other entities that control, are controlled by, or are under common
              control with that entity. For the purposes of this definition,
              "control" means (i) the power, direct or indirect, to cause the
              direction or management of such entity, whether by contract or
              otherwise, or (ii) ownership of fifty percent (50%) or more of the
              outstanding shares, or (iii) beneficial ownership of such entity.
        
              "You" (or "Your") shall mean an individual or Legal Entity
              exercising permissions granted by this License.
        
              "Source" form shall mean the preferred form for making modifications,
              including but not limited to software source code, documentation
              source, and configuration files.
        
              "Object" form shall mean any form resulting from mechanical
              transformation or translation of a Source form, including but
              not limited to compiled object code, generated documentation,
              and conversions to other media types.
        
              "Work" shall mean the work of authorship, whether in Source or
              Object form, made available under the License, as indicated by a
              copyright notice that is included in or attached to the work
              (an example is provided in the Appendix below).
        
              "Derivative Works" shall mean any work, whether in Source or Object
              form, that is based on (or derived from) the Work and for which the
              editorial revisions, annotations, elaborations, or other modifications
              represent, as a whole, an original work of authorship. For the purposes
              of this License, Derivative Works shall not include works that remain
              separable from, or merely link (or bind by name) to the interfaces of,
              the Work and Derivative Works thereof.
        
              "Contribution" shall mean any work of authorship, including
              the original version of the Work and any modifications or additions
              to that Work or Derivative Works thereof, that is intentionally
              submitted to Licensor for inclusion in the Work by the copyright owner
              or by an individual or Legal Entity authorized to submit on behalf of
              the copyright owner. For the purposes of this definition, "submitted"
              means any form of electronic, verbal, or written communication sent
              to the Licensor or its representatives, including but not limited to
              communication on electronic mailing lists, source code control systems,
              and issue tracking systems that are managed by, or on behalf of, the
              Licensor for the purpose of discussing and improving the Work, but
              excluding communication that is conspicuously marked or otherwise
              designated in writing by the copyright owner as "Not a Contribution."
        
              "Contributor" shall mean Licensor and any individual or Legal Entity
              on behalf of whom a Contribution has been received by Licensor and
              subsequently incorporated within the Work.
        
           2. Grant of Copyright License. Subject to the terms and conditions of
              this License, each Contributor hereby grants to You a perpetual,
              worldwide, non-exclusive, no-charge, royalty-free, irrevocable
              copyright license to reproduce, prepare Derivative Works of,
              publicly display, publicly perform, sublicense, and distribute the
              Work and such Derivative Works in Source or Object form.
        
           3. Grant of Patent License. Subject to the terms and conditions of
              this License, each Contributor hereby grants to You a perpetual,
              worldwide, non-exclusive, no-charge, royalty-free, irrevocable
              (except as stated in this section) patent license to make, have made,
              use, offer to sell, sell, import, and otherwise transfer the Work,
              where such license applies only to those patent claims licensable
              by such Contributor that are necessarily infringed by their
              Contribution(s) alone or by combination of their Contribution(s)
              with the Work to which such Contribution(s) was submitted. If You
              institute patent litigation against any entity (including a
              cross-claim or counterclaim in a lawsuit) alleging that the Work
              or a Contribution incorporated within the Work constitutes direct
              or contributory patent infringement, then any patent licenses
              granted to You under this License for that Work shall terminate
              as of the date such litigation is filed.
        
           4. Redistribution. You may reproduce and distribute copies of the
              Work or Derivative Works thereof in any medium, with or without
              modifications, and in Source or Object form, provided that You
              meet the following conditions:
        
              (a) You must give any other recipients of the Work or
                  Derivative Works a copy of this License; and
        
              (b) You must cause any modified files to carry prominent notices
                  stating that You changed the files; and
        
              (c) You must retain, in the Source form of any Derivative Works
                  that You distribute, all copyright, patent, trademark, and
                  attribution notices from the Source form of the Work,
                  excluding those notices that do not pertain to any part of
                  the Derivative Works; and
        
              (d) If the Work includes a "NOTICE" text file as part of its
                  distribution, then any Derivative Works that You distribute must
                  include a readable copy of the attribution notices contained
                  within such NOTICE file, excluding those notices that do not
                  pertain to any part of the Derivative Works, in at least one
                  of the following places: within a NOTICE text file distributed
                  as part of the Derivative Works; within the Source form or
                  documentation, if provided along with the Derivative Works; or,
                  within a display generated by the Derivative Works, if and
                  wherever such third-party notices normally appear. The contents
                  of the NOTICE file are for informational purposes only and
                  do not modify the License. You may add Your own attribution
                  notices within Derivative Works that You distribute, alongside
                  or as an addendum to the NOTICE text from the Work, provided
                  that such additional attribution notices cannot be construed
                  as modifying the License.
        
              You may add Your own copyright statement to Your modifications and
              may provide additional or different license terms and conditions
              for use, reproduction, or distribution of Your modifications, or
              for any such Derivative Works as a whole, provided Your use,
              reproduction, and distribution of the Work otherwise complies with
              the conditions stated in this License.
        
           5. Submission of Contributions. Unless You explicitly state otherwise,
              any Contribution intentionally submitted for inclusion in the Work
              by You to the Licensor shall be under the terms and conditions of
              this License, without any additional terms or conditions.
              Notwithstanding the above, nothing herein shall supersede or modify
              the terms of any separate license agreement you may have executed
              with Licensor regarding such Contributions.
        
           6. Trademarks. This License does not grant permission to use the trade
              names, trademarks, service marks, or product names of the Licensor,
              except as required for reasonable and customary use in describing the
              origin of the Work and reproducing the content of the NOTICE file.
        
           7. Disclaimer of Warranty. Unless required by applicable law or
              agreed to in writing, Licensor provides the Work (and each
              Contributor provides its Contributions) on an "AS IS" BASIS,
              WITHOUT WARRANTIES OR CONDITIONS OF ANY KIND, either express or
              implied, including, without limitation, any warranties or conditions
              of TITLE, NON-INFRINGEMENT, MERCHANTABILITY, or FITNESS FOR A
              PARTICULAR PURPOSE. You are solely responsible for determining the
              appropriateness of using or redistributing the Work and assume any
              risks associated with Your exercise of permissions under this License.
        
           8. Limitation of Liability. In no event and under no legal theory,
              whether in tort (including negligence), contract, or otherwise,
              unless required by applicable law (such as deliberate and grossly
              negligent acts) or agreed to in writing, shall any Contributor be
              liable to You for damages, including any direct, indirect, special,
              incidental, or consequential damages of any character arising as a
              result of this License or out of the use or inability to use the
              Work (including but not limited to damages for loss of goodwill,
              work stoppage, computer failure or malfunction, or any and all
              other commercial damages or losses), even if such Contributor
              has been advised of the possibility of such damages.
        
           9. Accepting Warranty or Additional Liability. While redistributing
              the Work or Derivative Works thereof, You may choose to offer,
              and charge a fee for, acceptance of support, warranty, indemnity,
              or other liability obligations and/or rights consistent with this
              License. However, in accepting such obligations, You may act only
              on Your own behalf and on Your sole responsibility, not on behalf
              of any other Contributor, and only if You agree to indemnify,
              defend, and hold each Contributor harmless for any liability
              incurred by, or claims asserted against, such Contributor by reason
              of your accepting any such warranty or additional liability.
        
           END OF TERMS AND CONDITIONS
        
           APPENDIX: How to apply the Apache License to your work.
        
              To apply the Apache License to your work, attach the following
              boilerplate notice, with the fields enclosed by brackets "[]"
              replaced with your own identifying information. (Don't include
              the brackets!)  The text should be enclosed in the appropriate
              comment syntax for the file format. We also recommend that a
              file or class name and description of purpose be included on the
              same "printed page" as the copyright notice for easier
              identification within third-party archives.
        
           Copyright [yyyy] [name of copyright owner]
        
           Licensed under the Apache License, Version 2.0 (the "License");
           you may not use this file except in compliance with the License.
           You may obtain a copy of the License at
        
               http://www.apache.org/licenses/LICENSE-2.0
        
           Unless required by applicable law or agreed to in writing, software
           distributed under the License is distributed on an "AS IS" BASIS,
           WITHOUT WARRANTIES OR CONDITIONS OF ANY KIND, either express or implied.
           See the License for the specific language governing permissions and
           limitations under the License.
License-File: LICENSE
License-File: NOTICE
Keywords: asr,diarization,pyannote,speaker-identification,transcription,whisper
Classifier: Development Status :: 2 - Pre-Alpha
Classifier: License :: OSI Approved :: Apache Software License
Classifier: Programming Language :: Python :: 3.11
Classifier: Programming Language :: Python :: 3.12
Classifier: Topic :: Multimedia :: Sound/Audio :: Analysis
Requires-Python: >=3.11
Requires-Dist: alembic>=1.13
Requires-Dist: celery[redis]>=5.4
Requires-Dist: fastapi>=0.115
Requires-Dist: httpx>=0.27
Requires-Dist: jinja2>=3.1
Requires-Dist: numpy>=1.26
Requires-Dist: pgvector>=0.3
Requires-Dist: psycopg[binary]>=3.2
Requires-Dist: pydantic-settings>=2.4
Requires-Dist: pydantic>=2.8
Requires-Dist: python-multipart>=0.0.9
Requires-Dist: pyyaml>=6.0
Requires-Dist: redis>=5.0
Requires-Dist: scipy>=1.13
Requires-Dist: sqlalchemy>=2.0
Requires-Dist: uvicorn[standard]>=0.30
Requires-Dist: yt-dlp>=2024.8
Provides-Extra: dev
Requires-Dist: mypy>=1.10; extra == 'dev'
Requires-Dist: pytest-asyncio>=0.23; extra == 'dev'
Requires-Dist: pytest-cov>=5.0; extra == 'dev'
Requires-Dist: pytest>=8.2; extra == 'dev'
Requires-Dist: ruff>=0.5; extra == 'dev'
Requires-Dist: scipy-stubs; extra == 'dev'
Requires-Dist: types-pyyaml; extra == 'dev'
Provides-Extra: parity
Requires-Dist: jiwer==4.0.0; extra == 'parity'
Requires-Dist: librosa==0.10.1; extra == 'parity'
Requires-Dist: more-itertools==11.1.0; extra == 'parity'
Requires-Dist: noisereduce==2.0.1; extra == 'parity'
Requires-Dist: onnxruntime==1.28.0; extra == 'parity'
Requires-Dist: pyloudnorm==0.1.0; extra == 'parity'
Requires-Dist: regex==2026.7.19; extra == 'parity'
Requires-Dist: setuptools<81,>=70; extra == 'parity'
Requires-Dist: soundfile==0.12.1; extra == 'parity'
Description-Content-Type: text/markdown

# Voxint

**Self-hosted audio intelligence: turn any recording into a speaker-attributed
transcript, then review it by hand.** Transcription, diarization, and speaker
identity, running entirely on your own hardware.

Built for individuals and small teams (researchers, journalists, educators) who
need audio work to stay local: no cloud account, no per-minute fees, no
recordings leaving the room.

Voxint takes an audio or video file and produces an enhanced transcript with
speakers attributed:

```
local file · upload · URL → acquire → preprocess
    → transcribe (Whisper) + diarize (pyannote) + embed (TitaNet)
    → LLM transcript enhancement → speaker matching → human adjudication
```

Point it at a local file (`voxint submit`), upload one through the browser, or
hand it a URL (`voxint fetch` / `POST /fetch`), which runs a yt-dlp download as
the pipeline's first stage. URL ingestion is authenticated admin egress, not a
sandbox. Fetch only trusted URLs unless you run the worker with restricted
egress — the opt-in `compose.ytdlp-egress.yaml` overlay ships that (a filtering
proxy that pins yt-dlp to vetted public addresses), or use a host egress firewall.
See [docs/operations.md](docs/operations.md#url-ingestion--egress-security).

## What it does that a bare pipeline doesn't

Most of Voxint is the orchestration around the models, the part that usually
gets left as an exercise for the reader.

- **Quality gates.** Non-speech and digital-silence triage before you spend GPU
  time, hallucination soft-tagging and stripping, chunk-completeness checks, and
  an outage-vs-data-defect taxonomy with explicit retry budgets.
- **Durable state.** A compare-and-swap run/stage state machine in Postgres. A
  crash at any stage is recoverable, and a human pause is a database row, not a
  held task.
- **Speaker identity with a paper trail.** pgvector cosine matching against a
  speaker roster that grows as you use it, a strict *named ≠ grounded*
  invariant, and machine proposals kept separate from human rulings.
- **A built-in review console.** Review queue, guarded slot workbench, and an
  immutable decision ledger, served as Jinja + htmx from the same FastAPI app.
  No Node toolchain.
- **Operable from the browser.** A keyset-paged `/runs` history browser (with a
  per-stage attempt ledger), bounded file upload, and yt-dlp URL ingestion, all
  from the same app. Submission is durable-first: a broker outage leaves the run
  queued for the recovery sweep rather than dropping it. You can cancel a live
  run (cooperative, exact-revision CAS), soft-archive a terminal one (reversibly
  hidden, ledger intact), and delete its derived audio to reclaim disk without
  touching the shared original.
- **Measurement harnesses.** Speaker-attribution scoring you can run as CLIs:
  name-accuracy against ground truth (McNemar / bootstrap / Wilson), acoustic
  agreement verdicts, and verdict-level ensemble fusion (worked example under
  [`examples/`](examples/README.md)). These score *who spoke*, not *what was
  transcribed*. ASR accuracy and WER measurement are out of scope for now.

## The review console

Machine proposals stay separate from human rulings. The review queue lists
completed runs with voices still needing a decision, and the slot workbench
shows each voice's evidence (grounded cosine matches, unverified LLM-heard
names, transcript previews) with assign / enroll / exclude / unknown actions.
Synthetic demo data pictured.

![Adjudication queue](docs/images/review-queue.png)

![Slot workbench](docs/images/slot-workbench.png)

## Status

**Pre-alpha.** APIs, schema, and layout may change without notice through the
0.x series.

## Quickstart

Requires Docker Engine with the Compose plugin ≥ 2.24 (`docker compose version`;
the legacy v1 `docker-compose` binary cannot parse this stack).

> **No NVIDIA GPU? Start here too.** Voxint does not need one. The CPU tier runs
> the full pipeline on plain amd64/arm64 servers and Apple Silicon with zero GPU
> configuration. Follow the same quickstart, then use the `compose.cpu.yaml`
> overlay wherever the GPU one appears (see
> [No NVIDIA GPU? (CPU tier)](#no-nvidia-gpu-cpu-tier)). **AMD GPU?** The ROCm
> tier accelerates transcription on it (4.8x the CPU baseline; the amdgpu kernel
> driver is the only host requirement) via `compose.rocm.yaml` (see
> [AMD GPU? (ROCm tier)](#amd-gpu-rocm-tier)). **Apple Silicon Mac?** The metal
> tier runs the model services natively so diarization uses the Apple GPU (see
> [Apple Silicon Mac? (metal tier)](#apple-silicon-mac-metal-tier)).

```bash
git clone https://github.com/bengizmo/voxint.git && cd voxint
```

**Guided install (recommended for a first run):**

```bash
./scripts/install.sh
```

It asks for an admin password, a media folder, and a compute tier for the model
services (GPU / CPU / none for now). That's the whole interview. All model
weights, diarization included, are vendored into the images, so no Hugging Face
account or token is involved. The installer generates everything else (including
a random `CSRF_SECRET`), pulls the pinned release images, starts the core stack
plus your chosen tier's model services, waits for the API to report healthy, and
prints the console URL. Re-running it is safe: an existing `.env` is kept unless
you ask to regenerate it (which backs the old one up first), and your tier
choice is remembered (`VOXINT_COMPOSE_TIER`).

**Or configure by hand:**

```bash
cp .env.example .env          # then edit at least VOXINT_PASSWORD
mkdir -p media                # media mount; pre-create so it isn't root-owned
docker compose pull           # prebuilt release images from GHCR
docker compose up -d          # Postgres+pgvector, Redis, migrate, API + review UI, worker, beat
curl http://127.0.0.1:8080/healthz   # default port; matches API_PORT if you changed it
```

The default compose files run the pinned release images even from a `main`
checkout (set `VOXINT_IMAGE_TAG` in `.env` to run a different release). A
one-shot `migrate` service brings the schema to head before the API and worker
start. Seeing it report `Exited (0)` in `docker compose ps -a` is success, not a
crash. If a default port is already in use on your host, override the published
side in `.env` (`POSTGRES_PORT`, `REDIS_PORT`, `API_PORT`). Details and day-2
operations live in [docs/operations.md](docs/operations.md).

Open the console at `http://127.0.0.1:8080/` (HTTP Basic, the `VOXINT_USER` /
`VOXINT_PASSWORD` you set). On a fresh install the console holds you at a
first-run setup wizard (`/setup`): configure media folders, vocabulary, and
optional LLM enhancement in the browser, then finish into a short guided
tutorial on a bundled three-speaker sample. Full walkthrough:
[docs/onboarding.md](docs/onboarding.md).

Once onboarding is complete, browse runs at `/runs` and adjudicate at `/review`.
Feed it work by uploading a file, pointing it at a URL (`docker compose exec api
voxint fetch <url>`), or submitting a local path (`docker compose exec api
voxint submit path/to/file.mp3`, relative to `MEDIA_ROOT`).

To run the GPU model services too (one NVIDIA GPU assumed), bring up the GPU
overlay. The diarization weights are vendored into the pyannote image
(sha256-pinned from the `pyannote-models-v1` asset release), so no Hugging Face
token is needed (see `services/pyannote/README.md`).

All three services share the one GPU. Their loaded weights total roughly 3.5-4.5
GB of VRAM (whisper large-v2 int8 ~1.5 GB, pyannote ~1-2 GB, TitaNet ~1 GB);
budget ~6-8 GB in practice for Whisper's batch/decode headroom and three
separate CUDA contexts. An 8 GB card is comfortable. Per-service figures live in
each `services/*/README.md`.

Then:

```bash
docker compose -f compose.yaml -f compose.gpu.yaml pull
docker compose -f compose.yaml -f compose.gpu.yaml up -d
```

Per-service details, env tunables, and image matrices are in
`services/*/README.md`; wire contracts are in
[docs/gpu-contracts.md](docs/gpu-contracts.md).

### No NVIDIA GPU? (CPU tier)

The same three model services ship as multi-arch (amd64 + arm64) `-cpu` images:
no GPU, no NVIDIA toolkit, runs on plain servers and Apple Silicon via Docker
Desktop.

```bash
docker compose -f compose.yaml -f compose.cpu.yaml up -d
```

Set your expectations accordingly: CPU inference is orders of magnitude slower.
A long recording that takes minutes on a GPU takes hours on CPU. The overlay
sets `COMPUTE_TIER=cpu`, which scales the pipeline's timeouts and stage leases so
slow-but-healthy runs aren't reclaimed as hung. Same contracts, same embedding
space (TitaNet runs on ONNX Runtime under a measured-equivalence parity gate).
Details:
[docs/operations.md](docs/operations.md#running-without-an-nvidia-gpu-cpu-tier).

### AMD GPU? (ROCm tier)

A hybrid tier for amd64 hosts with an AMD GPU. Transcription (whisper) runs on
the GPU via the CTranslate2 ROCm build (same engine, same code path, measured
4.8x the CPU baseline on RDNA4), while diarization and speaker embedding run the
`-cpu` images. MIOpen convolutions currently fail on AMD consumer GPUs, tracked
in [#4](https://github.com/bengizmo/voxint/issues/4):

```bash
docker compose -f compose.yaml -f compose.rocm.yaml up -d
```

The host needs only the amdgpu kernel driver: no ROCm install, no container
toolkit, since the `-rocm` image carries its own ROCm runtime. The overlay sets
`COMPUTE_TIER=rocm` (GPU-speed ASR, CPU-scaled leases for the rest). Details:
[docs/operations.md](docs/operations.md#running-on-an-amd-gpu-rocm-tier).

### Apple Silicon Mac? (metal tier)

Docker Desktop has no GPU passthrough, so on a Mac the containerized tiers are
CPU-only. The metal tier keeps the core stack in Docker but runs the three model
services natively so diarization uses the Apple GPU (torch-MPS, measured ~5x
native-CPU diarization on an M1 Pro, identical outputs). Transcription stays on
the host CPU in v1, so runs remain transcribe-bound: faster than the Docker CPU
tier, not GPU-stack fast.

```bash
./scripts/install.sh                  # choose [M]
./scripts/metal/voxint-metal.sh setup # native venvs + sha-verified weights
./scripts/metal/voxint-metal.sh up    # services under launchd
```

Weights come from the same sha-pinned release assets the images use, so still no
Hugging Face account or token. Details:
[docs/operations.md](docs/operations.md#running-on-apple-silicon-metal-tier).

To run the source you checked out instead of the release images, layer the build
overlays (exactly one service owns each build; see
[docs/operations.md](docs/operations.md)):

```bash
docker compose -f compose.yaml -f compose.build.yaml build api
docker compose -f compose.yaml -f compose.build.yaml up -d
```

For development without Docker:

```bash
uv sync --extra dev
uv run pytest tests/unit
uv run uvicorn voxint.api.app:app --reload
```

The scoring harness needs none of the stack. `pip install voxint` gives you the
`voxint score` CLI (pure file-in/file-out, no database or GPU services;
speaker-attribution metrics only, no ASR/WER). See
[`examples/`](examples/README.md).

## Deployment model

Docker-compose-first on a single Linux machine with one NVIDIA GPU:

- `compose.yaml`: Postgres (+pgvector), Redis, one-shot `migrate`, API (+ review
  UI), Celery worker, Celery beat (crash-recovery sweep scheduler)
- `compose.gpu.yaml`: the GPU model services (faster-whisper, pyannote, TitaNet)
- `compose.build.yaml` / `compose.gpu.build.yaml`: build-from-source overlays for
  development

Kubernetes is not required. It may become an optional enhancement later.

## Modularity

ASR, diarizer, embedder, and LLM providers sit behind typed protocols with
versioned HTTP contracts (`/v1/transcribe`, `/v1/diarize`, `/v1/embed`). The LLM
enhancement stage speaks to any OpenAI-compatible endpoint and is optional
(`LLM_ENABLED=false` by default). Domain-specific vocabulary, name seeds, and
prompt fragments load from a swappable **domain pack**: point `DOMAIN_PACK_PATH`
at a default pack, or drop several named packs under `DOMAIN_PACKS_DIR` and select
per run/folder. Each run **freezes** the pack it used, so editing a pack never
rewrites a past run's results. A neutral meeting/podcast pack ships as the
default. See [`docs/domain-packs.md`](docs/domain-packs.md).

## License

Apache-2.0. See [LICENSE](LICENSE) and [NOTICE](NOTICE). Vendored model weights
are redistributed under their own licenses with attribution (titanet:
CC-BY-4.0; pyannote segmentation: MIT; WeSpeaker embedding: CC-BY-4.0). See the
provenance files under `services/*/models/` and the model-asset releases.
