textMLCommons/peoples_speechspeech-recognitionasraudioenglishmlcommonscc-byspeech-to-textml-training

People's Speech English ASR Corpus

Free

Open dataset

Sample structure: 100 / 100

2 download links issued

Seller: DataBazaar

Already have an account? Log in

Agent? Connect your account →

Source, license and coverage

Supplier documentation. These claims are separate from the automated sample score. A listing edit date is not a data freshness date.

License: cc-by-2.0,cc-by-2.5,cc-by-3.0,cc-by-4.0,cc-by-sa-3.0,cc-by-sa-4.0
Source / creator: MLCommons/peoples_speech
Collection method: Audio was sourced primarily from openly-licensed material on the public web (e.g., government proceedings, archive.org content, and other CC-licensed audio). Transcriptions were obtained through a combination of crowdsourced human annotation and machine-generated alignment, then filtered. MLCommons normalizes audio into a consistent format and ships per-clip license metadata so downstream users can filter by license compatibility (e.g., CC-BY only vs. including share-alike).
Coverage start: Not documented
Coverage end: Not documented
Data last updated: Not documented
Update schedule: Not documented

Source documentation ↗

Embedded binary payloads are explicitly replaced with byte-length descriptors; the preview preserves accompanying text and metadata. Monolingual English only — not suitable for multilingual ASR training. Speaker demographic distribution skews toward sources available on the public web (e.g., over-representation of US/UK accents and formal/public-speaking registers; under-representation of conversational speech and some accents). A portion of transcripts are machine-generated and contain alignment/recognition errors. Share-alike clips (CC-BY-SA) require downstream model/data outputs to honor the SA term if those clips are used in training — filter by the per-clip license field if that matters for your use case.

Sample structure score: 100 / 100

This automated check describes the inspected sample, not factual accuracy, legal rights, representativeness, or the quality of the entire dataset. It is not a customer rating.

Assessed 10 sample records (JSON) on 2026-10-09. All records in the provided sample were checked.

Check	Points	Evidence
Populated cells	50 / 50	40 of 40 top-level cells contain a value. Null, absent and blank values count as missing; zero and false count as populated.
Consistent value types	30 / 30	40 of 40 populated cells match their column's most common observed type. Types are inferred, not checked against real-world truth.
Consistent record shape	20 / 20	10 of 10 records have the expected fields. CSV/TSV use the header width; JSON uses the union of observed keys.

Field-level findings and improvements

Check missing cells and mixed types below. Document intentional missing values or mixed types in your field descriptions. Do not fill legitimate unknowns with invented values just to increase this score.

Field	Missing cells	Most common type	Other populated types
id	0 / 10	string	0 / 10
audio	0 / 10	object	0 / 10
duration_ms	0 / 10	number	0 / 10
text	0 / 10	string	0 / 10

How the score is calculated, its limitations, and how to correct an assessment →

About this data

Large open English automatic speech recognition corpus of transcribed speech, designed for training and evaluating speech-to-text systems. Built by MLCommons under CC-BY/CC-BY-SA licenses.

Retrieve with your agent or Python

Create an account and configure DATABAZAAR_API_KEY. This example retrieves free or already purchased data; it never makes a purchase. For a multi-file dataset, choose a file index from the manifest.

Download the Python example

python3 retrieve-dataset.py 9d1a3d0b-3e65-49fc-ae00-66184bfc5989 --output dataset.bin

Full supplier documentation

## Overview The People's Speech Dataset is a large-scale English speech recognition corpus published by MLCommons, containing 30,000+ hours of transcribed audio from a diverse set of speakers. Data is distributed as parquet files with audio + text modalities, sized in the 1M–10M sample range. It is intended for training and benchmarking automatic speech recognition (ASR) systems and is openly licensed under CC-BY and CC-BY-SA variants. ## Schema - audio — audio (wav/flac bytes + sampling_rate) — the speech clip - text — string — human or machine transcription of the audio - duration_ms — int — clip duration in milliseconds - id — string — unique clip identifier - license — string — per-clip license tag (one of the CC-BY / CC-BY-SA variants) ## Sources - MLCommons/peoples_speech on Hugging Face — https://huggingface.co/datasets/MLCommons/peoples_speech — licenses: CC-BY-2.0, CC-BY-2.5, CC-BY-3.0, CC-BY-4.0, CC-BY-SA-3.0, CC-BY-SA-4.0 - Reference paper: Galvez et al., "The People's Speech: A Large-Scale Diverse English Speech Recognition Dataset for Commercial Usage" (arXiv:2111.09344) ## Methodology Audio was sourced primarily from openly-licensed material on the public web (e.g., government proceedings, archive.org content, and other CC-licensed audio). Transcriptions were obtained through a combination of crowdsourced human annotation and machine-generated alignment, then filtered. MLCommons normalizes audio into a consistent format and ships per-clip license metadata so downstream users can filter by license compatibility (e.g., CC-BY only vs. including share-alike). ## Known gaps & limitations Monolingual English only — not suitable for multilingual ASR training. Speaker demographic distribution skews toward sources available on the public web (e.g., over-representation of US/UK accents and formal/public-speaking registers; under-representation of conversational speech and some accents). A portion of transcripts are machine-generated and contain alignment/recognition errors. Share-alike clips (CC-BY-SA) require downstream model/data outputs to honor the SA term if those clips are used in training — filter by the per-clip license field if that matters for your use case. ## Intended use & out-of-scope - IS for: training and fine-tuning English ASR models, speech foundation model pretraining, ASR benchmarking, robustness/noise evaluation. - NOT for: multilingual ASR, speaker identification (no verified speaker labels), or any use that ignores per-clip license terms — CC-BY requires attribution and CC-BY-SA imposes share-alike on derivatives. _Federated dataset: 4,494 parquet shards, 1977.81 GB total. Queries and downloads stream through the DataBazaar API._ Original supplier listing: People's Speech — 30,000+ Hours of Transcribed English Speech (MLCommons) One of the world's largest open English ASR corpora: 30,000+ hours of transcribed speech under CC-BY/CC-BY-SA. Built by MLCommons for training and evaluating speech-to-text systems.

Schema

Name	Type	Description
id	VARCHAR	Unique identifier for the audio clip derived from source filename and timestamp
audio	STRUCT(bytes BLOB, path VARCHAR)	FLAC-encoded audio bytes and file path reference for the speech sample
duration_ms	INTEGER	Length of audio clip in milliseconds
text	VARCHAR	Human or machine-generated English transcription of the speech content

Sample Data

Preview a sample of the data before downloading.

Public sample only. Sign in to retrieve the full dataset, including free datasets.

For AI Agents

Via MCP Server

# 1. Add to your agent's MCP config (claude_desktop_config.json or similar):
{
  "mcpServers": {
    "databazaar": { "command": "npx", "args": ["databazaar-mcp"] }
  }
}

# 2. Your agent can then call:
search_datasets({ query: "People's Speech English ASR Co" })
// Found: 9d1a3d0b-3e65-49fc-ae00-66184bfc5989
get_download_url({ dataset_id: "9d1a3d0b-3e65-49fc-ae00-66184bfc5989" })  // free — sign in with MCP OAuth first

Via REST API

# Free dataset — sign in or use your account API key:
curl https://api.databazaar.io/datasets/9d1a3d0b-3e65-49fc-ae00-66184bfc5989/download-url -H "Authorization: Bearer $DATABAZAAR_API_KEY"