Datasetpublic

danielrosehill/ASR-WPM-And-Background-Noise-Eval

ASR WPM and Background Noise Evaluation Dataset A dataset of annotated audio recordings for evaluating how different factors affect Whisper (and other ASR/STT systems) transcription accuracy. Purpose This dataset provides controlled audio samples with annotations to evaluate ASR performance across: Speaking pace (fast, normal, slow, mumbled, whispered, weird voices) Background noise (cafe, music, conversations in various languages, traffic, sirens, etc.)… See the full description on the dataset page: https://huggingface.co/datasets/danielrosehill/ASR-WPM-And-Background-Noise-Eval.

sourceHugging Facemitupdated 10mo agoView on Hugging Face

1likes38downloads

Dataset Card

ASR WPM and Background Noise Evaluation Dataset

A dataset of annotated audio recordings for evaluating how different factors affect Whisper (and other ASR/STT systems) transcription accuracy.

Purpose

This dataset provides controlled audio samples with annotations to evaluate ASR performance across:

—Speaking pace (fast, normal, slow, mumbled, whispered, weird voices)
—Background noise (cafe, music, conversations in various languages, traffic, sirens, etc.)
—Microphone distance (close, normal, far)

Dataset Structure

Each sample includes:

—A WAV audio file (16kHz mono)
—Metadata with annotations describing recording conditions

Features

Feature	Type	Description
`id`	string	4-character hex identifier
`audio`	audio	Path to WAV file
`sample`	string	Text sample identifier
`sample_file`	string	Source text filename
`word_count`	int	Number of words in the sample
`duration_seconds`	float	Recording duration
`recorded_at`	string	Timestamp (YYYYMMDD_HHMMSS)
`annotations.pace`	string	Speaking pace category
`annotations.mic_distance`	string	Microphone distance
`annotations.background_noise`	string	Background noise type
`annotations.notes`	string	Additional notes
`equipment.microphone`	string	Recording device
`equipment.sample_rate`	int	Audio sample rate (16000)
`equipment.channels`	int	Audio channels (1 = mono)

Annotation Categories

Speaking Pace:

—fast - As fast as possible
—quick - Quicker than normal
—normal - Normal/conversational
—slow - Deliberately slow
—whispered - Whispered speech
—loud - Louder than normal
—weird_voices - Altered/unusual voice patterns

Microphone Distance:

—close - Less than 6 inches
—normal - 6-12 inches
—far - Greater than 12 inches

Background Noise:

—none - Silence
—cafe - Coffee shop ambience
—music - Background music (various genres)
—convo_same - Same-language conversation
—convo_other - Other-language conversation (Spanish, Arabic, Korean, Japanese, Mandarin, Cantonese, Irish English)
—convo_mixed - Mixed language babble
—transit - Airport/transportation sounds
—honking - Traffic/horns
—siren - Emergency vehicle sirens
—dogs - Dog barking
—baby - Baby sounds

Audio Specifications

—Format: WAV
—Sample Rate: 16kHz
—Channels: Mono
—Equipment: Samson Q2U USB Microphone

Usage

python

from datasets import load_dataset

dataset = load_dataset("danielrosehill/ASR-WPM-And-Background-Noise-Eval")

# Access audio and metadata
for sample in dataset["train"]:
    audio = sample["audio"]
    pace = sample["annotations"]["pace"]
    noise = sample["annotations"]["background_noise"]

Use Cases

—Benchmarking ASR/STT models under varying conditions
—Evaluating robustness to background noise
—Testing speech recognition at different speaking rates
—Comparing transcription accuracy across challenging audio scenarios

Source

Recording tools and methodology: Whisper-WPM-Eval

License

MIT