# VoiceBank+DEMAND noisy speech database (16 kHz)

## Facts

- Provider: Cassia Valentini-Botinhao (University of Edinburgh CSTR); 16 kHz conversion by JacobLinCool
- License: CC-BY-4.0 (https://datashare.ed.ac.uk/handle/10283/2791)
- Size: 9 files
- Format: Parquet (in sampled files)
- Stored at: Hugging Face, https://huggingface.co/datasets/JacobLinCool/VoiceBank-DEMAND-16k
- Access: public, readable without an account
- DOI: https://doi.org/10.7488/ds/2117
- Category: Speech & Audio
- Source: https://datashare.ed.ac.uk/handle/10283/2791
- Page: https://scigantic.com/archive/c4b66476-b44e-507b-a092-05aae40cb607
- Keywords: Speech & Audio, text-to-speech, benchmark, Parquet, Hugging Face

## About

VoiceBank+DEMAND is the noisy speech database for training speech enhancement algorithms and TTS models created by Cassia Valentini-Botinhao at the University of Edinburgh. It is a clean and noisy parallel speech database, designed to train and test speech enhancement methods operating at 48 kHz, in which read speech from Voice Bank speakers is paired with noisy versions mixed with environmental noise such as recordings from the DEMAND noise database; the original release offers 28-speaker and 56-speaker training sets and a separate test set. It is the standard benchmark for single-channel speech enhancement and denoising, reported with PESQ, STOI, CSIG, CBAK and COVL (SEGAN, MetricGAN, CMGAN and many others). This repository resamples the corpus to 16 kHz and stores it as six Parquet files under data/ (one test file and five train shards) with paired clean and noisy audio (measured 2026-11-14; the Hub reports about 2.3 GB). Load with datasets.load_dataset('JacobLinCool/VoiceBank-DEMAND-16k') or pandas. The original database on Edinburgh DataShare is licensed under CC BY 4.0, which governs this conversion; cite Valentini-Botinhao et al., 'Investigating RNN-based speech enhancement methods for noise-robust text-to-speech', SSW 2016.

## Load the data

```python
from datasets import load_dataset
ds = load_dataset("JacobLinCool/VoiceBank-DEMAND-16k")
print(ds)
```

## Data structure

File types in sampled files: Parquet

Example file: `data/test-00000-of-00001.parquet`

## Open it

- In a hosted notebook with the data mounted: https://scigantic.com/archive/c4b66476-b44e-507b-a092-05aae40cb607
- Full record as JSON (schema card included): https://api.scigantic.com/api/archives/c4b66476-b44e-507b-a092-05aae40cb607
- Over MCP: `get_schema_card` with id `c4b66476-b44e-507b-a092-05aae40cb607` at https://api.scigantic.com/mcp

## Related topics

- [Parquet datasets](https://scigantic.com/topics/parquet)
- [Benchmark datasets](https://scigantic.com/topics/benchmark)
- [Speech & Audio datasets](https://scigantic.com/topics/speech-and-audio)
- [Text-to-speech datasets](https://scigantic.com/topics/text-to-speech)
