Datasets:

jml2026
/

conversational-speech-dataset

	---
	license: cc-by-nc-4.0
	language:
	- en
	task_categories:
	- automatic-speech-recognition
	- audio-classification
	tags:
	- voice-ai
	- speech-data
	- conversational-ai
	- real-world-audio
	- crowdsourced
	- multi-speaker
	- meeting-transcription
	pretty_name: "🎙️ Silencio Conversational Speech Dataset"
	configs:
	- config_name: conversational_english
	data_files:
	- split: conversations
	path: conversational_english/conversations/**
	size_categories:
	- n<1K
	---

	# 🎙️ Silencio Network: Conversational Speech Dataset

	<p align="left">
	<img src="https://cdn-uploads.huggingface.co/production/uploads/69162b50b89e7abe20de4b5a/LWhs4p2lPFcyiVsP0tluu.png" width="40%">
	</p>

	[![Website](https://img.shields.io/badge/Website-silencioai.com-blue?style=flat-square)](https://www.silencioai.com)
	[![Contact](https://img.shields.io/badge/Contact-sofia@silencioai.com-green?style=flat-square)](mailto:sofia@silencioai.com)
	[![Data Available](https://img.shields.io/badge/Full_Corpus-100,000+_hours-orange?style=flat-square)](mailto:sofia@silencioai.com)

	---

	## Overview

	Sample conversational speech data from Silencio Network's crowdsourced voice AI platform. This dataset contains multi-speaker meeting recordings with word-level transcripts, speaker diarization, and rich demographic metadata.

	Each row represents one participant in a meeting and includes 3 audio files:

	\| Audio Column \| Description \| Format \|
	\|-------------\|-------------\|--------\|
	\| `file_name` (speaker audio) \| Individual participant's isolated recording \| WAV \|
	\| `full_meeting_single_channel` \| Full meeting mixed to single channel \| MP3 \|
	\| `full_meeting_multi_channel` \| Full meeting with separate speaker channels \| WAV \|

	Plus word-level meeting transcripts with speaker turns and timestamps.

	## Dataset Summary

	\| Config \| Language \| Meetings \| Participants \| Total Audio \|
	\|--------\|----------\|----------\|--------------\|-------------\|
	\| `conversational_english` \| English \| 4 \| 8 \| 135.9 MB \|

	## 🚀 Quick Start

	```python
	from datasets import load_dataset

	# Load conversational English samples
	ds = load_dataset("jml2026/conversational-speech-dataset", "conversational_english")

	conversations = ds['conversations']

	for sample in conversations:
	speaker_audio = sample['audio'] # Individual speaker recording
	meeting_audio = sample['full_meeting_single_channel'] # Full meeting
	transcript = sample['meeting_transcript_text'] # Plain text transcript
	print(f"Speaker {sample['speaker_id']} ({sample['gender']}, {sample['dialect']})")
	print(f" Words spoken: {sample['speaker_word_count']}")
	print(f" Meeting duration: {sample['meeting_duration']}s")
	```

	## Schema

	\| Column \| Type \| Description \|
	\|--------\|------\|-------------\|
	\| `file_name` \| Audio \| Individual speaker's isolated audio recording \|
	\| `meeting_id` \| int \| Unique meeting identifier \|
	\| `speaker_id` \| string \| Deterministic UUID for the speaker \|
	\| `gender` \| string \| Speaker gender \|
	\| `ethnicity` \| string \| Speaker ethnicity \|
	\| `occupation` \| string \| Speaker occupation \|
	\| `birth_place` \| string \| Speaker birth place \|
	\| `dialect` \| string \| Speaker dialect \|
	\| `year_of_birth` \| int \| Speaker year of birth \|
	\| `years_at_birth_place` \| int \| Years lived at birth place \|
	\| `languages_data` \| string \| JSON array of languages spoken with proficiency levels \|
	\| `language` \| string \| Meeting language \|
	\| `meeting_duration` \| int \| Meeting duration in seconds \|
	\| `meeting_transcript_word_count` \| int \| Total word count of the meeting transcript \|
	\| `speaker_word_count` \| int \| Word count for this speaker \|
	\| `full_meeting_single_channel` \| Audio \| Full meeting audio mixed to single channel \|
	\| `full_meeting_multi_channel` \| Audio \| Full meeting audio with separate speaker channels \|
	\| `meeting_transcript_json` \| string \| Full meeting transcript as JSON with word-level timestamps and speaker IDs \|
	\| `meeting_transcript_text` \| string \| Plain text meeting transcript with speaker turns \|

	## About Silencio Network

	Silencio Network operates a global platform with 500,000+ contributors across 130+ countries, collecting voice data through a mobile app. The full corpus exceeds 100,000 hours of validated speech data.

	For access to the complete dataset or custom data collection, contact sofia@silencioai.com.

	## License

	This sample dataset is released under [CC BY-NC 4.0](https://creativecommons.org/licenses/by-nc/4.0/). Commercial licensing is available upon request.