# Sell Multilingual Call Center Recordings for AI Training

> Why AI buyers want two-channel call recordings in languages other than English, published speech data prices, consent for reuse and recording law.

Canonical: https://origindatapartners.com/guides/sell-multilingual-call-recordings-ai
By: Origin Data Partners
Status: Published
Published: 2026-11-12
Updated: 2026-11-12
Sources checked: 2026-11-12

## Quick answer

Call centers and BPOs with two-channel (stereo) recordings in languages other than English hold scarce speech data: most speech technology covers only about one hundred languages. Buyers prefer agent and customer on separate channels, with transcripts and call outcomes. Published resale prices are low per hour, from $25 an hour for a Hindi support-call set. The deciding question is consent: callers agreed to recording, rarely to licensing for AI.

## Key points

- Two-channel audio keeps agent and customer apart, which is what speech buyers want.
- Languages other than English are where public speech data runs thin.
- Published dataset prices are per hour or per package and are not what a call center is paid.
- Consent to record is not consent to license; check each country's law and your client contracts.
- BPO recordings usually belong to the client whose customers are calling.

## Can a call center sell multilingual call recordings for AI training?

Sometimes, and the demand is real. Speech AI is thin outside a few languages: the authors of the Massively Multilingual Speech (MMS) paper wrote that "current speech technology is restricted to about one hundred languages which is a small fraction of the over 7,000 languages spoken around the world" ([MMS paper](https://arxiv.org/abs/2305.13516)). Versos scores speech and voice data at "demand index 60 (4 / 3 / 3)" in its November 2026 tracker ([Versos](https://www.versos.ai/blog/ai-training-data-demand-tracker)).

Real customer-service calls in Hindi, Tamil, Tagalog, Spanish or German carry what read speech cannot: accents, code-switching, interruptions, hold music and line noise. The obstacle is rights, covered below. Start with the [general call recordings guide](https://origindatapartners.com/guides/sell-call-recordings-ai-training) for the US consent and payment-card rules.


## Why buyers ask for two-channel (stereo) audio

Two-channel or stereo recording puts the agent on one channel and the customer on the other. It lets a buyer transcribe each speaker cleanly, measure overlap and train voice agents on one side of the conversation. Sellers advertise it: Defined.ai lists "274 hours of non-simulated, channel-separated telephony conversations in Hindi" and describes it as a "channel-separated customer service dataset with PII-redacted transcriptions" ([Defined.ai](https://defined.ai/datasets/hindi-call-center-speech-dataset)). By contrast, a large Nexdata call set is listed as "Format: 8kHz 16bit, wav, mono channel" ([Datarade](https://datarade.ai/data-products/unscripted-call-center-telephony-speech-data-20-000-hours-nexdata)).

| What buyers want | Why | Lowers value |
| --- | --- | --- |
| Agent and customer on separate channels | Clean per-speaker transcription | Mixed mono audio |
| Real, unscripted calls | Natural speech and noise | Scripted or role-played calls |
| Transcripts aligned to audio, personal details redacted | Ready to train | Untranscribed audio |
| Call reason and outcome labels | Teaches resolution, not only words | No record of how the call ended |
| Language, dialect and region tags | Buyers fill specific language gaps | Unlabeled language mix |
| Original sample rate, lossless files | No quality loss | Heavily compressed audio |


## Published price signals for conversational speech

There is little public pricing for a company's own call archive. The visible figures are list prices for finished data sets sold by data vendors, which include their cost of collection, cleaning and transcription. They show the market, not what a call center is paid.

| Listing | Published price | Source |
| --- | --- | --- |
| Hindi Audio Dataset [Support Calls] | "Starts at $25 / hour" | [Datarade](https://datarade.ai/data-products/unscripted-call-center-telephony-speech-data-20-000-hours-nexdata) |
| Nexdata unscripted call center telephony speech, 20,000 hours, many languages | "starts at USD20,000 per purchase" | [Datarade](https://datarade.ai/data-products/unscripted-call-center-telephony-speech-data-20-000-hours-nexdata) |
| Defined.ai channel-separated Hindi call center set, 274 hours | Quote on request | [Defined.ai](https://defined.ai/datasets/hindi-call-center-speech-dataset) |
| Consumer app paying people for their own calls (2025) | $0.15 to $0.30 per minute | [Origin call recordings guide](https://origindatapartners.com/guides/sell-call-recordings-ai-training) |

Company data programs differ on language. micro1 says: "At this time, we are primarily looking for companies that operate in English and maintain their documentation, workflows, and knowledge bases in English." ([micro1](https://company.micro1.ai/register/?referralCode=24823479-4339-4311-a157-2600b8cf72cf&utm_source=referral&utm_medium=share&utm_campaign=job_referral)). In Origin's view, speech-data buyers are often the better fit for non-English audio.


## Consent that covers reuse

Most calls open with a notice about recording for quality and training. That notice was written for your own staff, not for licensing audio to an AI developer. Buyers ask for proof: Defined.ai requires that "All data must be ethically sourced, including proper consent, documentation, and the right to commercialize the data for AI training." ([Defined.ai](https://defined.ai/partnership-programs)).

Where GDPR applies, personal data must be "collected for specified, explicit and legitimate purposes and not further processed in a manner that is incompatible with those purposes" ([GDPR Article 5](https://gdpr-info.eu/art-5-gdpr/)). A Business Insider source framed the same question for employees: does consent "actually extend to the sale of the data?" ([Business Insider](https://www.businessinsider.com/ai-training-employee-data-emails-slack-dms-2026-10)). Agents are recorded too, so their consent matters as well as the callers'.

- Check what your recording notice said in each language and each year.
- For new calls, consider a separate, explicit notice and opt-in for AI training use.
- Remove calls with card numbers, health details, minors or disputes.
- Redact names, account numbers, addresses and phone numbers from transcripts and, where possible, from audio.


## BPOs: whose recordings are they?

A BPO usually records calls on behalf of a client brand. The client's contract normally decides who owns the recordings and whether the BPO may reuse them, and many contracts forbid it. Recordings from in-house lines, internal help desks or your own brand are the cleaner starting point. Ask each client in writing before including their calls.


## Recording-law basics by country

Recording and privacy rules differ by country and travel with the caller, not only with your site. Origin's country pages summarize the local rules for call centers:

| Country | Origin page |
| --- | --- |
| India | [Call centers in India](https://origindatapartners.com/sell-company-data/india/call-centers) |
| Philippines | [Call centers in the Philippines](https://origindatapartners.com/sell-company-data/philippines/call-centers) |
| Mexico | [Call centers in Mexico](https://origindatapartners.com/sell-company-data/mexico/call-centers) |
| Germany | [Call centers in Germany](https://origindatapartners.com/sell-company-data/germany/call-centers) |
| United States | [Selling call recordings for AI training](https://origindatapartners.com/guides/sell-call-recordings-ai-training) |


## How to prepare

Related: [de-identifying company data](https://origindatapartners.com/guides/de-identify-company-data-before-licensing), [GDPR and CCPA for data licensing](https://origindatapartners.com/guides/data-licensing-gdpr-ccpa) and [which training data holds its value](https://origindatapartners.com/guides/ai-training-data-that-holds-value).

- Count hours by language, year, line of business and channel format (stereo or mono).
- Note which recordings belong to which client, and what each contract says.
- Collect the recording notices used, by language and date.
- Check the share of calls already transcribed and the sample rate of the files.
- Tag call reasons and outcomes from your CRM or ticketing system.


## Next step

This page states what each rule says and where to read it. It is not legal advice: your counsel decides how the rules apply to your records.

Does your company hold records like these? [See what your data is worth](https://origindatapartners.com/check-fit).

Origin Data Partners works with companies that own data and helps get it bought. We are independent of the services on this page, and a buyer may pay us when a deal closes ([how Origin is paid](https://origindatapartners.com/referral-disclosure)). [See what your data is worth](https://origindatapartners.com/check-fit).


## Questions and answers

### Why do AI companies want call recordings in languages other than English?

Because speech data is thin outside a few languages. The authors of the MMS paper wrote that speech technology covers about one hundred of more than 7,000 languages. Real calls add accents, code-switching and line noise that read speech lacks.

### What is a two-channel or stereo call recording?

A recording with the agent on one channel and the customer on the other. Buyers prefer it because each speaker can be transcribed and used separately.

### How much is call center audio worth?

Vendor list prices start at $25 an hour for a Hindi support-call set on Datarade, and a 20,000-hour multilingual set starts at USD 20,000. These are resale prices for finished data sets, not what a call center is paid.

### Does "this call may be recorded for training" allow selling the audio?

Usually not on its own. That notice is generally read as training your own staff. Licensing to an AI developer is a further use; under GDPR it must be compatible with the original purpose. Your counsel decides.

### Can a BPO sell its clients' call recordings?

Only if the client contract allows it, which is rare. Recordings usually belong to or are controlled by the client brand. Ask each client in writing.

## Sources

- [arXiv 2305.13516: Scaling speech technology to 1,000+ languages (MMS)](https://arxiv.org/abs/2305.13516)
- [Versos: AI training data demand tracker (November 2026)](https://www.versos.ai/blog/ai-training-data-demand-tracker)
- [Defined.ai: Hindi call center speech dataset](https://defined.ai/datasets/hindi-call-center-speech-dataset)
- [Datarade: unscripted call center telephony speech data, 20,000 hours (Nexdata)](https://datarade.ai/data-products/unscripted-call-center-telephony-speech-data-20-000-hours-nexdata)
- [Defined.ai: partnership programs](https://defined.ai/partnership-programs)
- [GDPR Article 5: principles relating to processing of personal data](https://gdpr-info.eu/art-5-gdpr/)
- [micro1: Company Data Partnerships](https://company.micro1.ai/register/?referralCode=24823479-4339-4311-a157-2600b8cf72cf&utm_source=referral&utm_medium=share&utm_campaign=job_referral)
- [Business Insider: AI labs are fueling a wild market for your Slack DMs (October 2026)](https://www.businessinsider.com/ai-training-employee-data-emails-slack-dms-2026-10)

## Related reading

- [Can a company sell customer call recordings for AI training?](https://origindatapartners.com/guides/sell-call-recordings-ai-training)
- [/sell-company-data/india/call-centers](https://origindatapartners.com/sell-company-data/india/call-centers)
- [/sell-company-data/philippines/call-centers](https://origindatapartners.com/sell-company-data/philippines/call-centers)
- [/sell-company-data/mexico/call-centers](https://origindatapartners.com/sell-company-data/mexico/call-centers)
- [/sell-company-data/germany/call-centers](https://origindatapartners.com/sell-company-data/germany/call-centers)
- [How to de-identify company data before licensing it](https://origindatapartners.com/guides/de-identify-company-data-before-licensing)
- [Which AI Training Data Holds Its Value in 2026?](https://origindatapartners.com/guides/ai-training-data-that-holds-value)
- [Employee Screen Recordings for AI Training: A Company Guide](https://origindatapartners.com/guides/employee-screen-recordings-for-ai-training)
- [/case-studies](https://origindatapartners.com/case-studies)
- [/refer](https://origindatapartners.com/refer)
- [/case-studies/call-center-recordings-905000](https://origindatapartners.com/case-studies/call-center-recordings-905000)