Datasets:
metadata
language:
- cv
license: cc0-1.0
task_categories:
- automatic-speech-recognition
- text-to-speech
pretty_name: Chuvash Voice
dataset_info:
features:
- name: audio
dtype: audio
- name: path
dtype: string
- name: sentence
dtype: string
- name: locale
dtype: string
- name: client_id
dtype: string
splits:
- name: train
num_bytes: 1343571989.56
num_examples: 29860
download_size: 1346925000
dataset_size: 1343571989.56
configs:
- config_name: default
data_files:
- split: train
path: data/train-*
How to use
We recommend using our dataset in conjunction with the Common Voice Corpus. We have attempted to maintain a consistent structure.
from datasets import load_dataset, DatasetDict, concatenate_datasets, Audio
comm_voice = DatasetDict()
comm_voice["train"] = load_dataset("mozilla-foundation/common_voice_17_0", "cv", split="train+validation", use_auth_token=True)
comm_voice["test"] = load_dataset("mozilla-foundation/common_voice_17_0", "cv", split="test", use_auth_token=True)
comm_voice = comm_voice.remove_columns(["accent", "age", "down_votes", "gender", "segment", "up_votes", "variant"])
comm_voice = comm_voice.cast_column("audio", Audio(sampling_rate=16000))
print(comm_voice)
print(comm_voice["train"][0])
chuvash_voice = DatasetDict()
chuvash_voice = load_dataset("alexantonov/chuvash_voice")
chuvash_voice = chuvash_voice.cast_column("audio", Audio(sampling_rate=16000))
print(chuvash_voice)
print(chuvash_voice["train"][0])
common_voice = DatasetDict({"train": concatenate_datasets([comm_voice["train"], chuvash_voice["train"]]), "test": comm_voice["test"]})
print(common_voice)
Text to Speech
Most of the corpus is a unique voice (client_id='177'). Therefore, the corpus can also be used for synthesis tasks.