Datasetpublic

deepdml/cv17-neucodec

Dataset Dataset Overview This dataset contains Common Voice speech data encoded into neural codec representations. Each sample includes: audio_path duration codes sentence language client_id The dataset is organized by language configuration and split into train, validation, and test sets when available. Dataset Statistics The following table summarizes the number of examples for each config_name and split. Dataset Statistics The… See the full description on the dataset page: https://huggingface.co/datasets/deepdml/cv17-neucodec.

sourceHugging Faceupdated 6mo agoView on Hugging Face

0likes186downloads

Dataset Card

Dataset

Dataset Overview

This dataset contains Common Voice speech data encoded into neural codec representations.

Each sample includes:

—audio_path
—duration
—codes
—sentence
—language
—client_id

The dataset is organized by language configuration and split into train, validation, and test sets when available.

Dataset Statistics

The following table summarizes the number of examples for each config_name and split.

Dataset Statistics

The following table summarizes the number of examples for each config_name, along with its corresponding language and available splits.

config_name	language	train_examples	validation_examples	test_examples
ar	Arabic	28,369	10,470	10,480
ast	Asturian	387	112	162
be	Belarusian	347,637	15,880	15,878
bg	Bulgarian	4,849	2,766	3,201
bn	Bengali	21,228	9,327	9,327
br	Breton	2,663	2,253	2,212
cs	Czech	20,144	9,009	9,067
cy	Welsh	7,960	5,371	5,379
da	Danish	3,484	2,105	2,530
de	German	—	16,183	16,183
el	Greek	1,920	1,700	1,701
es	Spanish	336,846	15,857	15,857
et	Estonian	3,157	2,653	2,653
fa	Persian	28,893	10,559	10,559
fi	Finnish	2,076	1,770	1,763
fr	French	558,054	16,159	16,159
frold	Old French	558,054	16,159	16,159
gl	Galician	25,159	9,982	9,990
ha	Hausa	1,925	582	661
hu	Hungarian	37,140	11,350	11,435
it	Italian	169,771	15,149	15,155
ja	Japanese	10,039	6,261	6,261
ka	Georgian	52,321	12,545	12,618
ko	Korean	376	330	339
lt	Lithuanian	7,253	4,436	4,753
lv	Latvian	11,364	6,752	6,752
mk	Macedonian	1,686	1,289	1,097
ml	Malayalam	1,259	764	710
mn	Mongolian	2,175	1,870	1,896
mr	Marathi	2,215	1,780	1,751
nl	Dutch	34,898	11,252	11,266
oc	Occitan	271	260	254
pl	Polish	20,729	9,230	9,230
pt	Portuguese	21,968	9,464	9,467
ro	Romanian	5,141	3,881	3,896
ru	Russian	26,377	10,203	10,203
sk	Slovak	3,258	2,588	2,647
sl	Slovenian	1,388	1,232	1,242
sr	Serbian	1,879	1,583	1,539
sv-SE	Swedish	7,744	5,210	5,259
sw	Swahili	46,494	12,251	12,253
ta	Tamil	45,587	12,095	12,074
te	Telugu	62	48	49
th	Thai	32,823	11,042	11,042
tr	Turkish	35,147	11,258	11,290
uk	Ukrainian	25,137	10,007	10,011
ur	Urdu	5,368	4,057	4,056
vi	Vietnamese	2,298	641	1,274

Notes

—Most configurations include train, validation, and test splits.
—de currently includes only validation and test splits in the dataset metadata.
—The language column provides a readable language name for each dataset configuration.

Features

—audio_path (string): path to the audio sample
—duration (float32): audio duration in seconds
—codes (sequence[int32]): neural codec token sequence
—sentence (string): transcription text
—language (string): language code
—client_id (string): speaker/client identifier

Usage

python

from datasets import load_dataset

dataset = load_dataset("deepdml/commonvoice-neucodec", "ar")
print(dataset)