# kimang18/implementing-encodec-khmer architecture

> Architecture diagram of the GitHub repository kimang18/implementing-encodec-khmer, generated by GitDiagram from its file tree, README and a sample of source files (last updated 2026-11-21). It is an AI-made overview: check details against the source.

- Interactive diagram: https://gitdiagram.com/kimang18/implementing-encodec-khmer
- Repository: https://github.com/kimang18/implementing-encodec-khmer

## Overview

This repository trains an EnCodec neural audio-compression model for speech tokenization. The main workflow is to configure a dataset and model, load and segment audio, encode and quantize it, then train against reconstruction and discriminator losses. The model also exposes tokenization and audio reconstruction capabilities. Edges reflect the README and shown imports/calls; some Lightning-managed training handoffs are not visible in the excerpts and are left unasserted.

Main entry points: [`train.py`](https://github.com/kimang18/implementing-encodec-khmer/blob/main/train.py) (Training entry), [`src/modules/encodec.py`](https://github.com/kimang18/implementing-encodec-khmer/blob/main/src/modules/encodec.py) (EnCodec model), [`src/modules/seanet.py`](https://github.com/kimang18/implementing-encodec-khmer/blob/main/src/modules/seanet.py) (SEANet encoder and decoder), [`src/dataset.py`](https://github.com/kimang18/implementing-encodec-khmer/blob/main/src/dataset.py) (Audio data module)

## Diagram (Mermaid)

```mermaid
flowchart TD

subgraph group_workflow["Training workflow"]
  node_config["Training config<br/>[config.yaml]"]
  node_entry["Training entry<br/>[train.py]"]
  node_data["Audio data module<br/>[dataset.py]"]
  node_audioio["Audio file helpers<br/>[utils.py]"]
end

subgraph group_codec["Codec model"]
  node_codecmodel["EnCodec model<br/>[encodec.py]"]
  node_tokenizer["Residual vector quantizer<br/>[quantizer.py]"]
  node_rvqvariant["Standard RVQ variant<br/>[standard_rvq.py]"]
end

subgraph group_networks["Network components"]
  node_encoderdecoder["SEANet encoder and decoder<br/>[seanet.py]"]
  node_conv["Convolution layers<br/>[conv.py]"]
  node_lstm["Sequence LSTM<br/>[lstm.py]"]
  node_discriminator["STFT discriminator<br/>[discriminator.py]"]
end

subgraph group_training["Training support"]
  node_trainmodule["Training module<br/>[train.py]"]
  node_balancer["Loss balancer<br/>[balancer.py]"]
  node_outputs["Training logs and checkpoints<br/>[train.py]"]
end

node_user(("Researcher"))
node_hfdata[("Hugging Face dataset")]

node_user -->|"starts training"| node_entry
node_user -->|"sets options"| node_config
node_entry -->|"loads config"| node_config
node_entry -->|"creates data module"| node_data
node_data -->|"loads audio splits"| node_hfdata
node_entry -->|"creates generator"| node_codecmodel
node_entry -->|"creates discriminator"| node_discriminator
node_entry -->|"assembles training"| node_trainmodule
node_trainmodule -->|"balances losses"| node_balancer
node_trainmodule -->|"logs and saves"| node_outputs
node_codecmodel -->|"constructs encoder and decoder"| node_encoderdecoder
node_codecmodel -->|"quantizes latents"| node_tokenizer
node_encoderdecoder -->|"uses convolutions"| node_conv
node_encoderdecoder -->|"uses sequence layers"| node_lstm
node_discriminator -->|"uses 2D convolutions"| node_conv

click node_config "https://github.com/kimang18/implementing-encodec-khmer/blob/main/configs/config.yaml"
click node_entry "https://github.com/kimang18/implementing-encodec-khmer/blob/main/train.py"
click node_data "https://github.com/kimang18/implementing-encodec-khmer/blob/main/src/dataset.py"
click node_trainmodule "https://github.com/kimang18/implementing-encodec-khmer/blob/main/train.py"
click node_balancer "https://github.com/kimang18/implementing-encodec-khmer/blob/main/src/balancer.py"
click node_outputs "https://github.com/kimang18/implementing-encodec-khmer/blob/main/train.py"
click node_codecmodel "https://github.com/kimang18/implementing-encodec-khmer/blob/main/src/modules/encodec.py"
click node_tokenizer "https://github.com/kimang18/implementing-encodec-khmer/blob/main/src/modules/quantizer.py"
click node_encoderdecoder "https://github.com/kimang18/implementing-encodec-khmer/blob/main/src/modules/seanet.py"
click node_conv "https://github.com/kimang18/implementing-encodec-khmer/blob/main/src/modules/conv.py"
click node_lstm "https://github.com/kimang18/implementing-encodec-khmer/blob/main/src/modules/lstm.py"
click node_discriminator "https://github.com/kimang18/implementing-encodec-khmer/blob/main/src/modules/discriminator.py"
click node_rvqvariant "https://github.com/kimang18/implementing-encodec-khmer/blob/main/src/modules/standard_rvq.py"
click node_audioio "https://github.com/kimang18/implementing-encodec-khmer/blob/main/src/utils.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_config,node_entry,node_data,node_audioio toneBlue
class node_codecmodel,node_tokenizer,node_rvqvariant,node_hfdata toneAmber
class node_encoderdecoder,node_conv,node_lstm,node_discriminator toneMint
class node_trainmodule,node_balancer,node_outputs toneRose
class node_user toneIndigo
```

## Components

### Training workflow

- [Training config](https://github.com/kimang18/implementing-encodec-khmer/blob/main/configs/config.yaml): `configs/config.yaml`
- [Training entry](https://github.com/kimang18/implementing-encodec-khmer/blob/main/train.py): `train.py`
- [Audio data module](https://github.com/kimang18/implementing-encodec-khmer/blob/main/src/dataset.py): `src/dataset.py`
- [Audio file helpers](https://github.com/kimang18/implementing-encodec-khmer/blob/main/src/utils.py): `src/utils.py`

### Codec model

- [EnCodec model](https://github.com/kimang18/implementing-encodec-khmer/blob/main/src/modules/encodec.py): `src/modules/encodec.py`
- [Residual vector quantizer](https://github.com/kimang18/implementing-encodec-khmer/blob/main/src/modules/quantizer.py): `src/modules/quantizer.py`
- [Standard RVQ variant](https://github.com/kimang18/implementing-encodec-khmer/blob/main/src/modules/standard_rvq.py): `src/modules/standard_rvq.py`

### Network components

- [SEANet encoder and decoder](https://github.com/kimang18/implementing-encodec-khmer/blob/main/src/modules/seanet.py): `src/modules/seanet.py`
- [Convolution layers](https://github.com/kimang18/implementing-encodec-khmer/blob/main/src/modules/conv.py): `src/modules/conv.py`
- [Sequence LSTM](https://github.com/kimang18/implementing-encodec-khmer/blob/main/src/modules/lstm.py): `src/modules/lstm.py`
- [STFT discriminator](https://github.com/kimang18/implementing-encodec-khmer/blob/main/src/modules/discriminator.py): `src/modules/discriminator.py`

### Training support

- [Training module](https://github.com/kimang18/implementing-encodec-khmer/blob/main/train.py): `train.py`
- [Loss balancer](https://github.com/kimang18/implementing-encodec-khmer/blob/main/src/balancer.py): `src/balancer.py`
- [Training logs and checkpoints](https://github.com/kimang18/implementing-encodec-khmer/blob/main/train.py): `train.py`

### Other components

- **Researcher**
- **Hugging Face dataset**

## Connections

- Researcher → Training entry: starts training (evidence: `README.md`)
- Researcher → Training config: sets options (evidence: `README.md`)
- Training entry → Training config: loads config (evidence: `train.py`)
- Training entry → Audio data module: creates data module (evidence: `train.py`)
- Audio data module → Hugging Face dataset: loads audio splits (evidence: `src/dataset.py`)
- Training entry → EnCodec model: creates generator (evidence: `train.py`)
- Training entry → STFT discriminator: creates discriminator (evidence: `train.py`)
- Training entry → Training module: assembles training (evidence: `train.py`)
- Training module → Loss balancer: balances losses (evidence: `train.py`)
- Training module → Training logs and checkpoints: logs and saves (evidence: `train.py`)
- EnCodec model → SEANet encoder and decoder: constructs encoder and decoder (evidence: `src/modules/encodec.py`)
- EnCodec model → Residual vector quantizer: quantizes latents (evidence: `src/modules/encodec.py`)
- SEANet encoder and decoder → Convolution layers: uses convolutions (evidence: `src/modules/seanet.py`)
- SEANet encoder and decoder → Sequence LSTM: uses sequence layers (evidence: `src/modules/seanet.py`)
- STFT discriminator → Convolution layers: uses 2D convolutions (evidence: `src/modules/discriminator.py`)
