Call and meeting recordings for AI training
Recorded calls and meetings capture how people actually talk at work, with interruptions, accents, jargon and decisions. SourceX looks for contact-center calls, sales calls and internal meetings with transcripts and outcomes, for speech recognition, voice agents and summarization models.
Listings
| Data | Typical sources | Modality | Availability |
|---|---|---|---|
| Contact-center call recordings with transcripts | Five9, Genesys Cloud, NICE CXone | Audio and text | On request |
| Sales call recordings linked to deal outcomes | Gong, Chorus, Zoom | Audio or video, and text | On request |
| Internal meeting recordings and transcripts | Zoom, Google Meet, Microsoft Teams | Audio or video, and text | On request |
What’s included
- Call audio, often with agent and caller on separate channels
- Time-aligned transcripts with speaker turns
- Disposition codes, handle times, transfers and call outcomes
- Sales calls linked to the deal they belonged to
- Internal meetings with agendas, notes and follow-up tasks
Public datasets and licensed data
The classic speech corpora were recorded between 1990 and 2003 and aren’t customer-service calls. The one large real call-center set, CallCenterEN, is transcripts only and non-commercial. Meeting corpora are largely role-played (AMI) or public city-council sessions (MeetingBank). Licensed recordings add current audio, real business conversations and what happened after each call.
| Public dataset | Released | What it contains | Limits for enterprise use | License |
|---|---|---|---|---|
| Switchboard-1 Release 2 | Linguistic Data Consortium (LDC), 1997 | About 2,400 two-sided telephone conversations (about 260 hours) recorded in 1990–91. | Strangers talking about assigned topics, decades ago; no customer-service calls. | LDC license (paid or membership) |
| Fisher English Training Speech (Parts 1 and 2) | LDC, 2004–05 | 11,699 telephone calls, about 1,960 hours across both parts, recorded in 2003. | Conversations on assigned topics, not business calls. | LDC license (paid or membership) |
| CallCenterEN | AIxBlock, 2025 | Transcripts of 91,706 real outsourced call-center conversations (about 10,448 audio hours). | Transcripts only (audio withheld for biometric-privacy reasons) and no commercial use. | CC BY-NC 4.0 |
| AMI Meeting Corpus | AMI Consortium, 2005 | 100 hours of meeting recordings with transcripts, slides and whiteboard data. | About two-thirds are role-played design-team scenarios. | CC BY 4.0 |
| MeetingBank | UCF, Adobe Research and Emory, 2023 | 1,366 city-council meetings (2014–2022) from six US cities. | Public government meetings, not internal company meetings; non-commercial license. | CC BY-NC-SA 4.0 |
Preparation and rights
Preparation requirements are agreed with each supplier before any work begins. For this kind of data they typically include:
- Checking recording notices and consent, including whether they cover sharing recordings with a third party for AI training; rules differ by state and country
- Reviewing audio under biometric privacy laws: voiceprints are biometric identifiers under Illinois’ Biometric Information Privacy Act, and California’s privacy law treats voice recordings that a voiceprint can be extracted from as biometric information
- Redacting names, card numbers and account details from transcripts and silencing them in audio
Every dataset has a documented owner and confirmed licensing rights, and is delivered only after an executed agreement and the supplier’s authorization. See data governance on sourcex.si.
More on sourcex.si
Questions
Can I license transcripts without the audio?
Yes. Audio raises extra consent and biometric questions, so a package can be scoped as transcripts with metadata only.
Which call systems do recordings come from?
Contact-center platforms such as Five9, Genesys Cloud, NICE CXone, Talkdesk and Amazon Connect, sales tools such as Gong, and meeting tools such as Zoom, Google Meet and Microsoft Teams.
Need call and meeting recordings?
Describe what you need: domain, volume, history, format and licensing terms. SourceX looks for companies that hold it and can license it.