Speech data at $60 to $95 an hour: what an owner must show

A $60 to $95 hour includes more than a recording

SpeechData.ai's current price page advertises US$60 to US$95 per audio hour for its catalog. The package includes two-person conversations recorded on separate channels, time-aligned transcripts checked by people, speaker metadata, a commercial license and speaker consent documentation. These are vendor list prices, checked October 27, 2026, before any negotiated discount.

An archive with audio alone is not the same deliverable. Nor does having transcripts establish that their timing, speaker labels, quality or permission scope meet the proposed use. Those differences need to be described before a ready-made dataset price can serve as a useful reference.

The hour also needs a definition. The provider defines one hour as one hour of recorded conversation. Two speaker channels do not turn a one-hour conversation into two saleable hours under that definition. Its buyer guide also asks buyers to distinguish total audio duration from speech-only duration. Agree the counting rule before applying a rate.

The same 500 hours can carry different catalog prices

The provider publishes a budget example covering three Arabic varieties. Each portion contains 500 hours, which makes the price difference easy to see without changing volume or mixing suppliers.

Source: SpeechData.ai's public pricing page, checked October 27, 2026. These are catalog budget examples before volume or multi-language discounts, not observed orders, paid amounts or RightMark supply. All prices are US dollars.

At the displayed prices, the Gulf Arabic portion costs $12,500 more than either other portion for the same 500 hours. That is a difference between these catalog listings. It does not prove that one dialect always commands a premium, or that the same difference applies to an owner's archive.

For an owner, the practical conclusion is straightforward: “Arabic audio” leaves out a field that affects which product and price a reviewer can compare. Describe the actual language variety and how it was identified. More detail makes the comparison meaningful; it does not automatically raise the price.

Use the Price Index as a dated reference, then check the quote

RightMark's Q4 2026 Price Index was published and updated on October 26. Its ready-made and low-resource speech ranges are unchanged from Q3. Those rows still cite a June 2026 guide; the linked guide was subsequently updated on October 17. The provider's current catalog describes a more specific package.

These ranges describe different scopes and dates; the index rows retain their published per-hour unit, so confirm the counting rule before comparing them. They are not evidence that the market rose or fell. The current provider also quotes custom collection per project; it does not give every existing archive a custom-collection rate. A price for recruiting speakers and creating new recordings cannot simply be applied to recordings already held.

The owner should use the index to orient the conversation, then ask for a scoped comparison: the same language, hour definition, included work and license. No verified buyer offer, accepted volume or completed payment underpins this note. There is also no supported deduction schedule from which to calculate an owner's net proceeds.

Count matching hours before you calculate dollars

Consider an explicitly hypothetical archive of 1,000 hours. Suppose 200 hours match the language, recording and documentation requirements being discussed. At the catalog's $65 rate, the comparison would be 200 × $65 = $13,000, not 1,000 × $65 = $65,000.

The calculation is an illustration of scope, not an appraisal. The 200 hours are assumed, not measured; $65 is one provider's packaged-data list price, not an offer for those hypothetical files. No rejection rate, expected payout or probability of sale can be inferred from it.

The remaining 800 hours may fit another brief, need more documentation or have unresolved restrictions. They are not automatically worthless. The useful first task is to separate what can be described against the present brief from what requires a different discussion.

What the current buyer record actually asks for

One Arabic-speech request recorded in RightMark's admin asks for hours by dialect, audio format, sampling rate, speaker configuration, the metadata schema and a representative sample. The request is still marked open, and the buyer relationship is in discussion. Its latest stored note is dated October 20; we reread it on October 27. No target volume, price or deadline is recorded, and no agreement or completed purchase is established.

That is useful demand evidence, but it is not a quote. The $65 and $90 catalog rates above come from a separate public supplier. They cannot be attributed to this buyer, multiplied by the buyer's presumed order size, or presented as what RightMark will pay an owner.

The practical first response is a dialect-level inventory and a permitted sample. A large collection in a different language does not fill a missing dialect. Nor should an owner pay for extensive preparation before the requested scope and review process are clear.

A refreshed estimate still needs qualification

The audio submission inspected for this note remains at the submitted stage. Its intake records permission to record; subsequent staff notes still call for permission evidence and sample review before a qualified scope can be priced. No supporting documents or completed verification checks were attached at our check.

Its stored estimate was recalculated against Q4 on October 26 without changing the displayed amount. That newer calculation date does not establish new accepted hours or completed consent review. An intake answer, an engine estimate, a reviewed sample and a buyer offer are different forms of evidence.

For an owner in this position, divide the archive into material whose proposed use can be supported and material that still needs review. Recording permission alone does not demonstrate that the proposed AI-training use is covered. RightMark's public speech-lot criteria require AI-training permission linked to the files, with the relevant rights review before scoping. The next step is to resolve that evidence and inspect a permitted sample, then price the supported subset. This case supplies no measured rejection rate or general rule for what a speech archive is worth.

Give the reviewer six answers they can check

A strong first response can be a small scope sheet. It should let someone trace the proposed hours to the corresponding files, metadata and permission evidence, without handing over the entire archive.

The first three fields and the sample request also appear in the admin buyer request checked October 27; its metadata-schema request informs the fifth field. Transcript and permission guidance additionally draw on the public provider package and RightMark speech-lot criteria. This combined checklist is preparation guidance, not a universal acceptance standard or legal clearance.

Start with a defined subset and make missing information visible. If dialect labels have not been checked, say so. If recording conditions vary, show that variation. If permissions need review, begin with the scope discussion before sharing recordings. A polished sample should not imply that every file in the archive has the same characteristics.

A better appraisal starts with a comparable product

The best immediate outcome is a narrower, more useful commercial question: “Here are the hours we can substantiate, here is what accompanies them, and here are the terms we need reviewed. Which route fits?”

That helps an owner decide whether to organize existing evidence, commission specific preparation work, seek another brief or wait. It does not justify paying for a full-archive transcription project on the assumption that the catalog rate will repay it. First establish what the proposed buyer actually needs and what the quoted scope includes.

RightMark publishes these notes to make that gap between an archive and a priced deliverable easier to understand. The numbers are useful when they help you prepare a comparable offer, rather than a larger unsupported total.