OleSpeech-IV: A Large-Scale Multispeaker and Multilingual Conversational Speech Dataset with Diverse Topics

2025-09-04

AI Summary

A critical gap exists in large-scale, multi-speaker, multilingual, and topically diverse conversational speech datasets. To address this, we introduce OleSpeech-IV—the first high-quality, multilingual conversational speech dataset explicitly designed for dialogue understanding and automatic speech recognition, covering English and Arabic and sourced from mainstream public podcasts, interviews, and conference recordings. We propose an end-to-end proprietary processing pipeline integrating automatic speaker diarization, fine-grained transcription, timestamp alignment, confidence scoring, and rigorous human verification to achieve high-precision annotation of speaker roles, turn boundaries, and word-level text. We publicly release the open-source subset OleSpeech-IV-2025-EN-AR-100, comprising 100 hours of meticulously annotated audio, under a non-commercial academic use license. This dataset substantially fills a longstanding void in multilingual conversational speech benchmarks and provides a foundational resource for developing robust, speaker-aware, and linguistically diverse dialogue models.

Technology Category

• Natural Language Processing: Conversational AI/Dialog Systems
• Machine Learning: Large Multimodal Models (LMMs)
• Data Mining & Knowledge Management: Conversational Systems for Recommendation & Retrieval

Application Category

• Search and Retrieval-Augmented AI: Multilingual and cross-lingual Web search
• Web Mining and Content Analysis: Mining multimedia, multimodal, multilingual, cross-lingual Web data
• Economics, Online Markets and Human Computation: Data quality aspects of human-annotated datasets

Abstract

OleSpeech-IV dataset is a large-scale multispeaker and multilingual conversational speech dataset with diverse topics. The audio content comes from publicly-available English podcasts, talk shows, teleconferences, and other conversations. Speaker names, turns, and transcripts are human-sourced and refined by a proprietary pipeline, while additional information such as timestamps and confidence scores is derived from the pipeline. The IV denotes its position as Tier IV in the Olewave dataset series. In addition, we have open-sourced a subset, OleSpeech-IV-2025-EN-AR-100, for non-commercial research use.

Problem

• Creating a large-scale multispeaker multilingual conversational dataset
• Providing diverse topic coverage from various speech sources
• Ensuring high-quality speaker and transcript annotations

Innovation

• Large-scale multispeaker multilingual conversational dataset
• Human-sourced speaker turns transcripts refinement pipeline
• Open-sourced subset for non-commercial research use