CommonVoice15

Overview

Common Voice 15 is a massively multilingual speech corpus collected by Mozilla, covering 114 languages with thousands of hours of validated speech data from volunteers worldwide.

Task Description

Task Type: Automatic Speech Recognition (ASR)
Input: Audio recordings with speech in various languages
Output: Transcribed text in the corresponding language
Languages: 114 languages including English, Mandarin Chinese, French, and many more

Key Features

Crowd-sourced speech recordings with community validation
Diverse speaker demographics (age, gender, accent)
Multiple languages with varying amounts of data
CC-0 licensed for open research and commercial use
High-quality transcriptions validated by multiple listeners

Evaluation Notes

Default configuration uses test split
Primary metric: Word Error Rate (WER)
Default subsets: en (English), zh-CN (Mandarin Chinese), fr (French)
Language-specific text normalization applied during evaluation
Prompt: "Please recognize the speech and only output the recognized content"

Properties

Property	Value
Benchmark Name	`common_voice_15`
Dataset ID	lmms-lab/common_voice_15
Paper	N/A
Tags	`Audio`, `MultiLingual`, `SpeechRecognition`
Metrics	`wer`
Default Shots	0-shot
Evaluation Split	`test`

Data Statistics

Metric	Value
Total Samples	43,143
Prompt Length (Mean)	67 chars
Prompt Length (Min/Max)	67 / 67 chars

Per-Subset Statistics:

Subset	Samples	Prompt Mean	Prompt Min	Prompt Max
`en`	16,386	67	67	67
`zh-CN`	10,625	67	67	67
`fr`	16,132	67	67	67

Audio Statistics:

Metric	Value
Total Audio Files	43,143
Audio per Sample	min: 1, max: 1, mean: 1
Formats	mp3

Sample Example

Subset: en

{
  "input": [
    {
      "id": "88959854",
      "content": [
        {
          "text": "Please recognize the speech and only output the recognized content:"
        },
        {
          "audio": "[BASE64_AUDIO: mp3, ~37.0KB]",
          "format": "mp3"
        }
      ]
    }
  ],
  "target": "Joe Keaton disapproved of films, and Buster also had reservations about the medium.",
  "id": 0,
  "group_id": 0,
  "subset_key": "en",
  "metadata": {
    "locale": "en",
    "path": "/home/tiger/.cache/huggingface/datasets/downloads/extracted/f54628fae82dd952031cdea3ec9c3d600c11d606e00cb8b3fd1b6ad500d7eb23/en_test_0/common_voice_en_27710027.mp3",
    "lang_id": "en"
  }
}

Prompt Template

Prompt Template:

Please recognize the speech and only output the recognized content:

Usage

Using CLI

evalscope eval \
    --model YOUR_MODEL \
    --api-url OPENAI_API_COMPAT_URL \
    --api-key EMPTY_TOKEN \
    --datasets common_voice_15 \
    --limit 10  # Remove this line for formal evaluation

Using Python

from evalscope import run_task
from evalscope.config import TaskConfig

task_cfg = TaskConfig(
    model='YOUR_MODEL',
    api_url='OPENAI_API_COMPAT_URL',
    api_key='EMPTY_TOKEN',
    datasets=['common_voice_15'],
    dataset_args={
        'common_voice_15': {
            # subset_list: ['en', 'zh-CN', 'fr']  # optional, evaluate specific subsets
        }
    },
    limit=10,  # Remove this line for formal evaluation
)

run_task(task_cfg=task_cfg)

3.6 KiB Raw Blame History