# butspeechfit/diarizen architecture

> Architecture diagram of the GitHub repository butspeechfit/diarizen, generated by GitDiagram from its file tree, README and a sample of source files (last updated 2026-11-16). It is an AI-made overview: check details against the source.

- Interactive diagram: https://gitdiagram.com/butspeechfit/diarizen
- Repository: https://github.com/butspeechfit/diarizen

## Overview

DiariZen provides pretrained speaker diarization inference plus recipes for self-supervised training, multi-channel operation, and structured pruning/distillation. The principal workflow loads a model hub, accepts an audio file, runs diarization, and can write RTTM output. Pyannote supplies the shared audio/model inference and task framework. Training and pruning links are documented at capability level; unsampled implementations are represented by their exact directories, and uncertain runtime calls are omitted.

Main entry points: [`diarizen/trainer_dual_opt.py`](https://github.com/butspeechfit/diarizen/blob/main/diarizen/trainer_dual_opt.py) (Dual optimizer trainer), [`diarizen/pipelines/inference.py`](https://github.com/butspeechfit/diarizen/blob/main/diarizen/pipelines/inference.py) (DiariZen pipeline), [`diarizen/trainer_single_opt.py`](https://github.com/butspeechfit/diarizen/blob/main/diarizen/trainer_single_opt.py) (Single optimizer trainer), [`pyannote-audio/pyannote/audio/core/task.py`](https://github.com/butspeechfit/diarizen/blob/main/pyannote-audio/pyannote/audio/core/task.py) (Task framework)

## Diagram (Mermaid)

```mermaid
flowchart TD

subgraph group_inference["Diarization inference"]
  node_diarizen_pipeline["DiariZen pipeline<br/>[inference.py]"]
  node_clustering["VBx clustering<br/>[VBx.py]"]
end

subgraph group_models["Model architecture"]
  node_eend["WavLM diarization"]
  node_ssl_model["Prunable SSL model<br/>[model.py]"]
  node_ssl_components["SSL components<br/>[components.py]"]
  node_conformer["Conformer blocks<br/>[conformer.py]"]
  node_distill_model["Distillation model"]
end

subgraph group_training["Training and pruning"]
  node_train_recipes["Training recipes"]
  node_mc_recipes["Multi-channel recipes"]
  node_trainer_single["Single optimizer trainer"]
  node_trainer_dual["Dual optimizer trainer"]
  node_trainer_prune["Distill-prune trainer"]
  node_pruning_recipes["Pruning recipes"]
  node_checkpoint["Checkpoint utilities<br/>[ckpt_utils.py]"]
  node_scheduler["Noam updater<br/>[noam_updater.py]"]
  node_optimization["Optimization helpers<br/>[optimization.py]"]
  node_trainer_utils["Trainer utilities<br/>[trainer_utils.py]"]
  node_logger["Experiment logging<br/>[logger.py]"]
end

subgraph group_pyannote["Pyannote framework"]
  node_audio_io["Audio I/O<br/>[io.py]"]
  node_inference_engine["Chunk inference<br/>[inference.py]"]
  node_model_framework["Audio model<br/>[model.py]"]
  node_pipeline_framework["Pipeline framework<br/>[pipeline.py]"]
  node_diarization["Speaker diarization"]
  node_signal["Signal utilities<br/>[signal.py]"]
  node_task["Task framework<br/>[task.py]"]
  node_loss["Training losses<br/>[loss.py]"]
  node_powerset["Powerset encoding<br/>[powerset.py]"]
end

node_user(("Audio user"))

node_user -->|"submits audio"| node_diarizen_pipeline
node_diarizen_pipeline -->|"loads audio"| node_audio_io
node_diarizen_pipeline -->|"runs diarization"| node_diarization
node_diarizen_pipeline -->|"uses utilities"| node_signal
node_inference_engine -->|"loads model"| node_model_framework
node_pipeline_framework -->|"manages inference"| node_inference_engine
node_diarization -.->|"uses framework"| node_pipeline_framework
node_eend -->|"uses encoder"| node_ssl_model
node_eend -->|"uses blocks"| node_conformer
node_ssl_model -->|"builds from"| node_ssl_components
node_train_recipes -.->|"trains with"| node_trainer_single
node_train_recipes -.->|"trains with"| node_trainer_dual
node_mc_recipes -.->|"trains with"| node_trainer_dual
node_pruning_recipes -.->|"trains with"| node_trainer_prune
node_trainer_prune -->|"optimizes"| node_distill_model
node_pruning_recipes -.->|"prunes model"| node_ssl_model
node_trainer_single -->|"uses optimizer"| node_optimization
node_trainer_single -->|"uses utilities"| node_trainer_utils
node_trainer_single -->|"logs progress"| node_logger
node_trainer_single -->|"uses scheduler"| node_scheduler
node_trainer_dual -->|"uses optimizer"| node_optimization
node_trainer_dual -->|"uses utilities"| node_trainer_utils
node_trainer_dual -->|"logs progress"| node_logger
node_trainer_dual -->|"uses scheduler"| node_scheduler
node_task -->|"uses losses"| node_loss
node_task -->|"uses encoding"| node_powerset
node_model_framework -->|"configures task"| node_task
node_diarizen_pipeline -->|"loads weights"| node_checkpoint

click node_diarizen_pipeline "https://github.com/butspeechfit/diarizen/blob/main/diarizen/pipelines/inference.py"
click node_audio_io "https://github.com/butspeechfit/diarizen/blob/main/pyannote-audio/pyannote/audio/core/io.py"
click node_inference_engine "https://github.com/butspeechfit/diarizen/blob/main/pyannote-audio/pyannote/audio/core/inference.py"
click node_model_framework "https://github.com/butspeechfit/diarizen/blob/main/pyannote-audio/pyannote/audio/core/model.py"
click node_pipeline_framework "https://github.com/butspeechfit/diarizen/blob/main/pyannote-audio/pyannote/audio/core/pipeline.py"
click node_diarization "https://github.com/butspeechfit/diarizen/blob/main/pyannote-audio/pyannote/audio/pipelines/speaker_diarization.py"
click node_clustering "https://github.com/butspeechfit/diarizen/blob/main/diarizen/clustering/VBx.py"
click node_signal "https://github.com/butspeechfit/diarizen/blob/main/pyannote-audio/pyannote/audio/utils/signal.py"
click node_eend "https://github.com/butspeechfit/diarizen/blob/main/diarizen/models/eend/model_wavlm_conformer_mc.py"
click node_ssl_model "https://github.com/butspeechfit/diarizen/blob/main/diarizen/models/module/wav2vec2/model.py"
click node_ssl_components "https://github.com/butspeechfit/diarizen/blob/main/diarizen/models/module/wav2vec2/components.py"
click node_conformer "https://github.com/butspeechfit/diarizen/blob/main/diarizen/models/module/conformer.py"
click node_train_recipes "https://github.com/butspeechfit/diarizen/tree/main/recipes/diar_ssl"
click node_mc_recipes "https://github.com/butspeechfit/diarizen/tree/main/recipes/diar_ssl_mc"
click node_trainer_single "https://github.com/butspeechfit/diarizen/blob/main/diarizen/trainer_single_opt.py"
click node_trainer_dual "https://github.com/butspeechfit/diarizen/blob/main/diarizen/trainer_dual_opt.py"
click node_trainer_prune "https://github.com/butspeechfit/diarizen/blob/main/diarizen/trainer_distill_prune.py"
click node_distill_model "https://github.com/butspeechfit/diarizen/blob/main/diarizen/models/pruning/model_distill_prune.py"
click node_pruning_recipes "https://github.com/butspeechfit/diarizen/tree/main/recipes/diar_ssl_pruning"
click node_checkpoint "https://github.com/butspeechfit/diarizen/blob/main/diarizen/ckpt_utils.py"
click node_task "https://github.com/butspeechfit/diarizen/blob/main/pyannote-audio/pyannote/audio/core/task.py"
click node_loss "https://github.com/butspeechfit/diarizen/blob/main/pyannote-audio/pyannote/audio/utils/loss.py"
click node_powerset "https://github.com/butspeechfit/diarizen/blob/main/pyannote-audio/pyannote/audio/utils/powerset.py"
click node_scheduler "https://github.com/butspeechfit/diarizen/blob/main/diarizen/noam_updater.py"
click node_optimization "https://github.com/butspeechfit/diarizen/blob/main/diarizen/optimization.py"
click node_trainer_utils "https://github.com/butspeechfit/diarizen/blob/main/diarizen/trainer_utils.py"
click node_logger "https://github.com/butspeechfit/diarizen/blob/main/diarizen/logger.py"

classDef toneNeutral fill:#f8fafc,stroke:#334155,stroke-width:1.5px,color:#0f172a
classDef toneBlue fill:#dbeafe,stroke:#2563eb,stroke-width:1.5px,color:#172554
classDef toneAmber fill:#fef3c7,stroke:#d97706,stroke-width:1.5px,color:#78350f
classDef toneMint fill:#dcfce7,stroke:#16a34a,stroke-width:1.5px,color:#14532d
classDef toneRose fill:#ffe4e6,stroke:#e11d48,stroke-width:1.5px,color:#881337
classDef toneIndigo fill:#e0e7ff,stroke:#4f46e5,stroke-width:1.5px,color:#312e81
classDef toneTeal fill:#ccfbf1,stroke:#0f766e,stroke-width:1.5px,color:#134e4a
class node_diarizen_pipeline,node_clustering,node_user toneBlue
class node_eend,node_ssl_model,node_ssl_components,node_conformer,node_distill_model toneAmber
class node_train_recipes,node_mc_recipes,node_trainer_single,node_trainer_dual,node_trainer_prune,node_pruning_recipes,node_checkpoint,node_scheduler,node_optimization,node_trainer_utils,node_logger toneMint
class node_audio_io,node_inference_engine,node_model_framework,node_pipeline_framework,node_diarization,node_signal,node_task,node_loss,node_powerset toneRose
```

## Components

### Diarization inference

- [DiariZen pipeline](https://github.com/butspeechfit/diarizen/blob/main/diarizen/pipelines/inference.py): `diarizen/pipelines/inference.py`
- [VBx clustering](https://github.com/butspeechfit/diarizen/blob/main/diarizen/clustering/VBx.py): `diarizen/clustering/VBx.py`

### Model architecture

- [WavLM diarization](https://github.com/butspeechfit/diarizen/blob/main/diarizen/models/eend/model_wavlm_conformer_mc.py): `diarizen/models/eend/model_wavlm_conformer_mc.py`
- [Prunable SSL model](https://github.com/butspeechfit/diarizen/blob/main/diarizen/models/module/wav2vec2/model.py): `diarizen/models/module/wav2vec2/model.py`
- [SSL components](https://github.com/butspeechfit/diarizen/blob/main/diarizen/models/module/wav2vec2/components.py): `diarizen/models/module/wav2vec2/components.py`
- [Conformer blocks](https://github.com/butspeechfit/diarizen/blob/main/diarizen/models/module/conformer.py): `diarizen/models/module/conformer.py`
- [Distillation model](https://github.com/butspeechfit/diarizen/blob/main/diarizen/models/pruning/model_distill_prune.py): `diarizen/models/pruning/model_distill_prune.py`

### Training and pruning

- [Training recipes](https://github.com/butspeechfit/diarizen/tree/main/recipes/diar_ssl): `recipes/diar_ssl`
- [Multi-channel recipes](https://github.com/butspeechfit/diarizen/tree/main/recipes/diar_ssl_mc): `recipes/diar_ssl_mc`
- [Single optimizer trainer](https://github.com/butspeechfit/diarizen/blob/main/diarizen/trainer_single_opt.py): `diarizen/trainer_single_opt.py`
- [Dual optimizer trainer](https://github.com/butspeechfit/diarizen/blob/main/diarizen/trainer_dual_opt.py): `diarizen/trainer_dual_opt.py`
- [Distill-prune trainer](https://github.com/butspeechfit/diarizen/blob/main/diarizen/trainer_distill_prune.py): `diarizen/trainer_distill_prune.py`
- [Pruning recipes](https://github.com/butspeechfit/diarizen/tree/main/recipes/diar_ssl_pruning): `recipes/diar_ssl_pruning`
- [Checkpoint utilities](https://github.com/butspeechfit/diarizen/blob/main/diarizen/ckpt_utils.py): `diarizen/ckpt_utils.py`
- [Noam updater](https://github.com/butspeechfit/diarizen/blob/main/diarizen/noam_updater.py): `diarizen/noam_updater.py`
- [Optimization helpers](https://github.com/butspeechfit/diarizen/blob/main/diarizen/optimization.py): `diarizen/optimization.py`
- [Trainer utilities](https://github.com/butspeechfit/diarizen/blob/main/diarizen/trainer_utils.py): `diarizen/trainer_utils.py`
- [Experiment logging](https://github.com/butspeechfit/diarizen/blob/main/diarizen/logger.py): `diarizen/logger.py`

### Pyannote framework

- [Audio I/O](https://github.com/butspeechfit/diarizen/blob/main/pyannote-audio/pyannote/audio/core/io.py): `pyannote-audio/pyannote/audio/core/io.py`
- [Chunk inference](https://github.com/butspeechfit/diarizen/blob/main/pyannote-audio/pyannote/audio/core/inference.py): `pyannote-audio/pyannote/audio/core/inference.py`
- [Audio model](https://github.com/butspeechfit/diarizen/blob/main/pyannote-audio/pyannote/audio/core/model.py): `pyannote-audio/pyannote/audio/core/model.py`
- [Pipeline framework](https://github.com/butspeechfit/diarizen/blob/main/pyannote-audio/pyannote/audio/core/pipeline.py): `pyannote-audio/pyannote/audio/core/pipeline.py`
- [Speaker diarization](https://github.com/butspeechfit/diarizen/blob/main/pyannote-audio/pyannote/audio/pipelines/speaker_diarization.py): `pyannote-audio/pyannote/audio/pipelines/speaker_diarization.py`
- [Signal utilities](https://github.com/butspeechfit/diarizen/blob/main/pyannote-audio/pyannote/audio/utils/signal.py): `pyannote-audio/pyannote/audio/utils/signal.py`
- [Task framework](https://github.com/butspeechfit/diarizen/blob/main/pyannote-audio/pyannote/audio/core/task.py): `pyannote-audio/pyannote/audio/core/task.py`
- [Training losses](https://github.com/butspeechfit/diarizen/blob/main/pyannote-audio/pyannote/audio/utils/loss.py): `pyannote-audio/pyannote/audio/utils/loss.py`
- [Powerset encoding](https://github.com/butspeechfit/diarizen/blob/main/pyannote-audio/pyannote/audio/utils/powerset.py): `pyannote-audio/pyannote/audio/utils/powerset.py`

### Other components

- **Audio user**

## Connections

- Audio user → DiariZen pipeline: submits audio
- DiariZen pipeline → Audio I/O: loads audio (evidence: `diarizen/pipelines/inference.py`)
- DiariZen pipeline → Speaker diarization: runs diarization (evidence: `diarizen/pipelines/inference.py`)
- DiariZen pipeline → Signal utilities: uses utilities (evidence: `diarizen/pipelines/inference.py`)
- Chunk inference → Audio model: loads model (evidence: `pyannote-audio/pyannote/audio/core/inference.py`)
- Pipeline framework → Chunk inference: manages inference (evidence: `pyannote-audio/pyannote/audio/core/pipeline.py`)
- Speaker diarization → Pipeline framework: uses framework
- WavLM diarization → Prunable SSL model: uses encoder (evidence: `diarizen/models/eend/model_wavlm_conformer_mc.py`)
- WavLM diarization → Conformer blocks: uses blocks (evidence: `diarizen/models/eend/model_wavlm_conformer_mc.py`)
- Prunable SSL model → SSL components: builds from (evidence: `diarizen/models/module/wav2vec2/model.py`)
- Training recipes → Single optimizer trainer: trains with
- Training recipes → Dual optimizer trainer: trains with
- Multi-channel recipes → Dual optimizer trainer: trains with
- Pruning recipes → Distill-prune trainer: trains with
- Distill-prune trainer → Distillation model: optimizes (evidence: `diarizen/trainer_distill_prune.py`)
- Pruning recipes → Prunable SSL model: prunes model (evidence: `recipes/diar_ssl_pruning/apply_pruning.py`)
- Single optimizer trainer → Optimization helpers: uses optimizer (evidence: `diarizen/trainer_single_opt.py`)
- Single optimizer trainer → Trainer utilities: uses utilities (evidence: `diarizen/trainer_single_opt.py`)
- Single optimizer trainer → Experiment logging: logs progress (evidence: `diarizen/trainer_single_opt.py`)
- Single optimizer trainer → Noam updater: uses scheduler (evidence: `diarizen/trainer_single_opt.py`)
- Dual optimizer trainer → Optimization helpers: uses optimizer (evidence: `diarizen/trainer_dual_opt.py`)
- Dual optimizer trainer → Trainer utilities: uses utilities (evidence: `diarizen/trainer_dual_opt.py`)
- Dual optimizer trainer → Experiment logging: logs progress (evidence: `diarizen/trainer_dual_opt.py`)
- Dual optimizer trainer → Noam updater: uses scheduler (evidence: `diarizen/trainer_dual_opt.py`)
- Task framework → Training losses: uses losses (evidence: `pyannote-audio/pyannote/audio/core/task.py`)
- Task framework → Powerset encoding: uses encoding (evidence: `pyannote-audio/pyannote/audio/core/task.py`)
- Audio model → Task framework: configures task (evidence: `pyannote-audio/pyannote/audio/core/model.py`)
- DiariZen pipeline → Checkpoint utilities: loads weights (evidence: `diarizen/pipelines/inference.py`)
