metadata

task_categories:
  - text-to-speech
language:
  - en
pretty_name: CommonVoice Clones
size_categories:
  - 1K<n<10K

CommonVoice Clones

This dataset consists of recordings taken from the CommonVoice english dataset. Each voice and transcript are used as input to a voice cloner, and generate a cloned version of the voice and text.

TTS Models

We use the following high-scoring models from the TTS leaderboard:

playHT
metavoice
StyleTTSv2
XttsV2

Model Comparisons

To facilitate data exploration, check out this HF space 🤗, which allows you to listen to all clones from a given source (e.g. commonvoice).

Data Sampling

The subset of the data was chosen such that a balanced amount of male/female voices are included. We also include a suggested train, validation and test split. Unique voice IDs do not overlap between the splits, i.e. a voice used to clone a sample in train will never be found in validation and test.