Why this exists
Voice detection has a credibility problem that is not technical. The tools mostly work. What they will not do is tell you when they are wrong, and a verdict about a person from a system that will not state its error rate is not a finding. It is an accusation with a percentage attached.
Truthring is an AI voice detection service made by Lacewing Technologies, an independent software company in Navi Mumbai, India. It analyses a recording and returns one of three verdicts, likely synthetic, likely human or unclear, and it publishes its false positive rate next to its detection rate. The product is pre-launch; access is by request.
The number nobody publishes
Detectors advertise a headline accuracy figure. It is almost always the rate at which the tool catches synthetic clips, and it is easy to make large: flag more aggressively and it rises.
What that conceals is the price of raising it, which is the rate at which real human speech gets flagged as fake. The two move in opposite directions. A vendor showing only the flattering half is not exactly lying; they are showing you one side of a trade-off and letting you assume the other side is free.
It is not free. The people who pay are the ones being accused: the support agent whose call is flagged, the witness whose voicemail is disputed, the candidate who loses a job over an artefact of compression. For them the false positive rate is the product.
So we publish both, broken out by audio condition, and we would rather lose a comparison on the headline number than win one by hiding the number that matters. It goes on the accuracy page, with the conditions each rate was measured under and the date it was measured.
The two verdicts are not worth the same
Reading “likely synthetic” and “likely human” as mirror images is the most common mistake people make with this category. They are not doing the same kind of work.
“Likely synthetic” is the strong result, because it needs positive evidence: something in the signal that generated speech leaves behind and a human throat does not. “Likely human” is the weak one, because it is also what you get when that evidence was destroyed on the way to us. A narrowband phone codec, a clip re-recorded off a laptop speaker, aggressive noise suppression in a messaging app: any of those can strip the traces that would have justified the other verdict.
So a clean result from any voice detector, ours included, should be read as nothing found rather than nothing there. Where the signal survives and where it does not is set out under phone audio and limitations.
Four commitments
Publish the error rate
Both directions, by audio condition, with the dataset described and the date attached. If we are not prepared to publish a number, we do not quote it in marketing either.
Say where we fail
A page whose entire purpose is the conditions under which we are wrong. Re-recorded audio defeats our strongest signal. We would rather write that ourselves than have you find it out.
Version every result
A verdict names the model that produced it and carries a reference code. A result you relied on in March remains reproducible after we retrain in June.
Return “unclear” without charging
When the audio does not support an answer we say so, and the check is not billed. We do not want a financial reason to manufacture confidence.
What we are not
- Not proof. A probability with a stated method is one input to a human decision. Anyone selling certainty in this category is selling something the technology does not support.
- Not real-time. We analyse recordings after the fact. If you need to interrupt a live call, the call-centre vendors are built for that and we are not.
- Not multi-modal. Audio only, deliberately. Platforms covering images and video too will cover more ground and go less deep on this.
Where the alternatives are stronger, and where we are: comparisons.
Who builds Truthring
Truthring is a product of Lacewing Technologies, an independent software company based in Navi Mumbai, Maharashtra, India. Lacewing is operated by Dipak Ashok Bhosale as proprietor and describes itself on its own site as a software and AI services firm. It is a small operation. This page is not going to imply otherwise.
Two details are missing on purpose. We do not print a corporate form after the name, because no incorporated entity, company identification number or registry record has been confirmed, and the word Lacewing uses about itself is proprietor. Until a registration is produced and checked, “Lacewing Technologies” appears here as a trading name and nothing more. [VERIFY: registered legal entity name and form]
We also do not print a street address. Three different ones currently appear across the company’s own pages, an app store listing and business directories, and a confident address that turns out to be stale is worse on a legal page than a city and a state. [VERIFY: registered address for legal pages]
A tax registration number, GSTIN 27BSZPB5125R1ZW, is published on the TextSight site. We have not checked it against the GST portal ourselves, so it is repeated here as something to confirm rather than something confirmed. [VERIFY: confirm GSTIN against the GST portal before republishing] Lacewing’s published terms for TextSight name Indian law and Maharashtra jurisdiction; what governs your use of this site is on the terms page.
How Truthring relates to TextSight.ai
Lacewing also operates TextSight.ai, founded in 2025. TextSight is an AI text detection and writing suite. Its subject is written prose: whether a passage looks machine-generated, and what a writer or an institution does about that. TextSight also ships a working voice detector, which accepts MP3, WAV, M4A, OGG and FLAC files up to 10 MB.
The two products therefore touch at one edge, and it is worth being exact about what that means. They are siblings under one owner, not a single product wearing two names. Truthring is audio only, and it is designed around the recordings people actually receive rather than the clean samples detectors are usually graded on.
What genuinely carries across is the company and one habit. TextSight ran a false positive study on 1,180 human-written academic papers from 2018, released the dataset under CC BY 4.0, and states on its own site that it has had no independent third-party audit. That is the standard Truthring is being held to: publish the rate that embarrasses you, publish the data underneath it, and say plainly when nobody outside the company has checked your work.
What does not carry across is anything technical. There is no claim here of shared models, shared training data, shared infrastructure or transferable results. A false positive rate measured on written English tells you nothing about a rate on a compressed voice note. Every Truthring figure will be measured on Truthring, on audio, under conditions named on the page that reports it. [VERIFY: which components, if any, are shared between the two products]
What state this is in, plainly
Truthring is pre-launch. There is no live detector on this site yet, no price you can pay, no API key to issue and no customers. Where a page invites you to check a recording, that is a request for access rather than an analysis running today, and the quantitative results are forthcoming rather than withheld.
Nothing here has been certified by anybody. There is no SOC 2 report, no ISO 27001 certificate and no independent audit of the method, the dataset or the security posture. No publication has covered Truthring, no award has been given to it and no funding has been announced. You will not find a logo strip on this site, because there is nothing honest to put in one.
Competitors do not write paragraphs like that. It is the one advantage a company this size actually has: every sentence above can be checked, and being checkable is the whole proposition.
Questions about the company
Who owns Truthring?
Truthring is a product of Lacewing Technologies, an independent software company based in Navi Mumbai, Maharashtra, India, operated by Dipak Ashok Bhosale as proprietor. The registered legal entity name and form have not been confirmed and are not published here.
Is Truthring the same thing as the TextSight voice detector?
No. Both are Lacewing Technologies products. TextSight.ai is an AI text detection and writing suite that also ships a voice detector; Truthring is a separate product for audio only. No shared models, training data or infrastructure are claimed between them.
Is Truthring certified, audited or accredited?
No. There is no SOC 2 report, no ISO 27001 certificate and no independent third-party audit of the method, the evaluation data or the security posture. If that changes, the certificate and its scope will be published rather than the badge alone.
Can I use a Truthring result as proof?
No. A result is a probability with a stated method, an audio condition and a model version attached. Treat it as one input to a human decision, and verify anything consequential through a channel you already trust.
Can I buy Truthring today?
Not yet. The product is pre-launch, so there is no live detector, no billing and no API access. You can request access, and published accuracy figures will follow the first evaluation run rather than precede it.
Reviewed