README.md

---
license: mit
task_categories:
- automatic-speech-recognition
- voice-activity-detection
language:
- en
pretty_name: TurnMaster
size_categories:
- 100K<n<1M
tags:
- endpointing
---
# Download
```bash
# Install huggingface_hub if needed
pip install huggingface_hub


# Download dataset
hf download turnmaster/TurnMaster --repo-type dataset --local-dir ./TurnMaster

cd TurnMaster
```

# Installation

```bash
pip install -r requirements.txt
```


# Project structure

```text
TurnMaster/
├── audio/
├── metadata/
├── processing/
│   ├── augment_taskmaster.py
│   ├── extract_shards.py
│   ├── curate_dataset.py
├── requirements.txt
└── README.md
```


# Extract shards


```bash
# Extract to custom directory
python TurnMaster_processing/extract_shards.py --sharded-dir ./ --output-dir Turnmaster_raw
```

# OPTIONAL: remove shards

```bash
Remove extracted shards (unecessary for the latter pre-processing)
rm -rf audio/
```

### OPTIONAL: create a curated version of the dataset. You can also skip this step and reuse the standart curated version of TurnMaster

```bash
# Extract from cached download
python extract_shards.py --sharded-dir ~/.cache/huggingface/hub/datasets--turnmaster--TurnMaster/snapshots/main --output-dir extracted
```

# Build dataset

### Base (no silence)
```bash
python TurnMaster_processing/augment_taskmaster.py \
 --aug-dataset-save-dir YOUR/SAVE/DIR \
 --taskmaster-root-dir DIR/TO/EXTRACTED/DATASET \
 --manifest-dir DIR/TO/EXTRACTED/DATASET \
 --silence-max-start 0.0 \
 --silence-min-end 0.2 \
 --silence-max-end 2.0 \
 --normalize-text \
 --custom-dollar-normalisation \
 --keep-punct \
 --lam 0.0 \
 ```

### Base silence insertion
```bash
python TurnMaster_processing/augment_taskmaster.py \
 --aug-dataset-save-dir YOUR/SAVE/DIR \
 --taskmaster-root-dir DIR/TO/EXTRACTED/DATASET \
 --manifest-dir DIR/TO/MANIFEST/EXTRACTED/DATASET \
 --silence-max-start 0.0 \
 --silence-min-end 0.2 \
 --silence-max-end 2.0 \
 --normalize-text \
 --custom-dollar-normalisation \
 --keep-punct \
 --silence-insertion \
 --lam 0.0 \
 --max-initial-silence-duration 1.5 \
 --silence-insertion-max-s 2.0 \
 --pause-per-second-ratio 1000.0
 ```

 ```bibtex
@misc{turnmaster2026,
  title        = {TurnMaster: A Synthetic Spoken Dialogue Benchmark for Voicebot End-Of-Turn Detection},
  author       = {Anonymous},
  year         = {2026},
  publisher    = {Hugging Face},
  url          = {https://huggingface.co/datasets/turnmaster/TurnMaster}
}
 ```