Up to 20,000 Hours of Real Job Interview Audio for AI Training (50+ Accents)
Audio, Speech & Acoustic Datasets
Tags and Keywords
$55,300
About
Up to 20,000 Hours (Growing Daily) of Fully-Consented Real Online Job Interview Audio in English (+ Non-English available) | 50+ Country Accents & Diverse Speakers (Africa, SEA, South Asia, LatAm + More) | AI Training Data | Audio (Extracted from Video Interviews) + Timestamp-Aligned Transcripts | Question/Context Descriptions | Global Coverage
A large-scale dataset of up to 20,000 hours (growing daily) of real online job interview audio in English, featuring natural, non-scripted interview responses from a highly diverse speaker base across many regions and accent groups — including Africa, Southeast Asia, South Asia, Latin America, North America, Europe, and more.
All participants explicitly opt in (fully consented) for their data to be used for AI training and shared under controlled licensing.
This is a living dataset: new recordings are added daily, so total hours increase over time (current total: 18,538 hours as of 26 August 2026).
Each clip is primarily single-speaker (candidate-focused), making it highly valuable for training models on authentic, real-world monologue speech in interview conditions.
Unlike staged or scripted recordings, this dataset captures authentic interview behavior — spontaneous phrasing, pauses, disfluencies, turn-taking, and real-world device variability. Recording conditions vary naturally across devices, rooms and connections, which improves model robustness.
Key Features
1) Accent Diversity at Scale (Underrepresented Accents Included)
Designed for real-world robustness with broad English accent coverage:
- Diverse regional accents and speech patterns
- Variation in pronunciation, rhythm, speech speed, and vocabulary
- Strong coverage of accents often missing from public datasets (Africa, SEA, South Asia, LatAm)
2) Real Interview Audio (Non-Scripted, Natural Speech)
All sessions come from genuine interview-style Q&A, capturing:
- Natural disfluencies (hesitations, self-corrections, fillers)
- Realistic interview pacing and tone
- Authentic response structure under real interview conditions
3) AI-Ready Packaging (Audio + Transcript + Context)
Each session can include synchronized assets such as:
- Audio extracted from the original online interview capture
- Timestamp-aligned transcripts (segment/sentence level)
- Question prompts, verbatim, on 85% of answers; question type/category where the employer configured one (20%)
- Technical and quality metadata (duration, device/channel signals)
Supports tasks including:
- Speech recognition, accent robustness and speaker analytics
- Lip/audio alignment research
- Speech modeling
- Multimodal conversational understanding
4) Real-World Capture Conditions
Audio reflects realistic online interview environments:
- Mobile and desktop capture
- Consumer-grade device cameras and microphones
- Mostly indoor environments with natural lighting/background variation
- VoIP-style audio characteristics and device differences
5) Fully-Consented & Commercial-Ready
- Explicit opt-in consent for AI training and controlled dataset sharing
- Packaged for smooth integration into ML pipelines and enterprise procurement workflows
6) Continuously Expanding Library (Daily Updates)
- New recordings added every day
- Monthly intake: 3,000–4,200 new hours (June 2026: 4,165 h; July 2026: 4,220 h)
- Custom collection to brief: language, region, job family or question set — capacity on request
- Current total: 18,538 hours (26 August 2026)
- Updated releases available upon request
Use Cases
- Speech recognition and speaker modelling
- Audiovisual speech recognition and robustness benchmarking
- ASR / speech-to-text with diverse accents
- Multimodal evaluation under real capture conditions
- Accent robustness testing across regions
Delivery Format (Typical)
- Audio files (extracted from the original interview capture)
- Timestamp-aligned transcripts
- Question/context descriptors + metadata schema
- Documentation (data card, release manifest, usage notes)
TRUSTLoading...
$55,300
Download Dataset in AUDIO Format
Recommended Datasets
Loading recommendations...