Skip to content
tl;dr of AmberScript
Amberscript is a Netherlands-based transcription tool that turns audio and video files you already have into text and subtitles, with strong European data residency and the option to escalate any file to a human transcriber.
Its automatic tier is fast and easy to edit, but it works best as a draft rather than a finished transcript. I tested it against an official UK parliamentary transcript alongside tl;dv.
AmberScript identified all seven speakers correctly but trailed on word accuracy and entity recognition, misreading names like Jingye on every pass and changing them between runs.
Pricing is by audio hours, from €19 a month, with auto-renewal and no refunds, worth reading first.
Choose Amberscript to transcribe single files under EU data rules. Choose tl;dv to record and transcribe live meetings with AI notes on top.
AmberScript is a tool that takes audio and video files and turns them into text, and while that is clear from the marketing, many reviews of AmberScript don’t actually stop to see if the text is any good.
So, we decided to put AmberScript to the test against tl;dv and see how the two tools fared. tl;dv is incredibly strong on transcribing languages such as Japanese, Swedish, German, French, and Spanish, but how did it stand up against a tool transcribing in English?
I found a UK parliamentary clip, a committee session on British Steel, which, while maybe not the most exciting of topics, was a detailed, technical conversation with many different English accents and plenty of policy jargon to put both tools through their paces.
Before I get to the results, let’s take a look at AmberScript, what it is, where it started, and dig beneath the surface.
What Is AmberScript?
AmberScript is a Netherlands-based transcription service. It takes any audio and video files you may have and creates detailed, accurate transcripts of the text, can create subtitles, and more.
That’s the core job, and while it isn’t a meeting recorder or AI meeting assistant, it’ll often be put together with these same tools.
In the signup, it becomes clear that AmberScript offers two separate services. There is an automatic tier that uses AI and speech recognition to create a transcript in minutes, with a claimed accuracy of around 90% on clean English audio. There is also the human option that matches you with a professional human transcriber who then takes your file and creates a deliverable within 24 hours to five days. The accuracy of this is claimed to be at 99%.
I tested the automatic transcription because while the human transcription is a very useful tool, it’s not something that we offer at tl;dv.
The company, being based in Europe, means that they are understandably strong when it comes to privacy and security. The company stores files and processes in Frankfurt, meaning that no data leaves Europe. The company holds several accreditations, including ISO 27001 and ISO 9001, and is GDPR-compliant.
AmberScript Features
The features of AmberScript are fairly simple. They offer users the chance to upload and transcribe files. They offer automatic transcription in over 90 languages and human transcription in 18.
The way that the tool operates is an in-browser editor, where, after you upload your files, you can edit drafts, with the audio available to check against. The audio is synced to the text, allowing users to click and jump to that particular point in the recording.
There is the ability to add words and specific jargon to a custom dictionary. Exports are available in a range of formats, including Word, text, SRT, and VTT, meaning that you can generate subtitles natively.
On the AI plans, there is speaker separation, meaning that AmberScript can find and isolate speakers, labeling them with placeholders. There is no automatic speaker naming, so this needs to be done manually by the user.
One area that I found when I was researching the tool was that there is a separate subtitling product and a data-annotation business that builds training datasets to order. While this isn’t relevant specifically to the testing that we undertook, it does appear to have led to some slightly confusing mixed messaging regarding training data on AI.
How I Tested AmberScript’s Accuracy
In order to test the capabilities of AmberScript’s accuracy I needed to run tests where there was a Ground Truth source of a baseline transcript, or supporting documentation.
For this, I selected to use clips from the UK Parliament Public Accounts Committee session on the government’s intervention in British Steel, held on 22 June 2026. This meant that while I am a native English speaker myself, the test wasn’t based purely upon my own listening skills and gave me a solid reference to work with. You can read the official committee transcript here.
The basis of the test was a 10-minute clip with seven speakers. Both clips were uploaded directly to AmberScript using its interface and to tl;dv’s interface as well. I left every default setting on AmberScript as it was, except for the “smoothing” feature. As tl;dv offers filler word tracking, this meant that the two transcripts would both be as raw as they were spoken, offering the best view of speaker recognition rather than a tidied-up version.
There was one constraint that shaped the testing, and ended up being a problem later on: AmberScript’s free trial caps at 10 minutes, which was exactly one run, and no way to check a second attempt would score differently. So I purchased additional credits in order to run more tests, taking my time availability to an hour, and run three runs of the same transcript in total. As a slight tangent, I did enjoy the “Drop It Like It’s Hot” Easter Egg when I was uploading files. It was a smidge of humanity in quite a corporate-looking interface. I did have to “checkout” each time I uploaded a file, which felt a little bit nerve-racking, and I did think I may have accidentally charged my credit card several times during testing.
Since I had the baseline transcript, I was able to gauge the word error rate (WER). This is a standard measure of how much a transcript has deviated from the original references. There was, however, one caveat to this, and something that I’ve found across all the testing that we have done over the many types of transcripts available. Getting a “true” raw transcript, based on the verbatim spoken words of a meeting, is difficult. Most transcripts for official government and business meetings that are available are cleaned up. To circumnavigate this restriction, I stripped filler and expanded contractions on both sides before scoring; the same rule applied to each tool. The absolute numbers still run high for both and the differences are still fairly clear in the results.
AmberScript Accuracy: The Results
After testing both tools, I used an LLM to score the outputs across the runs. While a true WER wasn’t feasible, given that the transcript was cleaned, I used an LLM to score both outputs against that reference for a directional error rate instead. What this means is that these are what I would deem a gap rather than a true lab measurement. If the transcript were a clean verbatim reference, we would expect a much lower percentage score from both, so bear that in mind.
AmberScript’s automatic tier averaged around a 39% word error rate across the runs of the clip. tl;dv, for comparison, scored 31.6% on the same file. Both of these numbers are inflated by the cleaned reference rather than being an absolute grade.
tl;dv was able to pull away on some very specific parts of the transcript, naming and terms.
AmberScript was unable to consistently recognize entities. For example, the steel company’s owner is called Jingye. tl;dv was able to identify this clearly and consistently, whereas AmberScript missed it on every pass but also changed it throughout the runs. It was rendered as Zengi, Geneva, Gina, and at one point “junior”. It also turned Sarah Green into Sarah Graham, and “steel strategy” into “street strategy” and “deal strategy”. This is the kind of inconsistency and error that will be picked up immediately.
Across the three runs, I found that AmberScript’s overall accuracy was steady across the board, but the proper nouns really struggled. This instability was not scattered noise but is clearly a bit of a weak spot in the transcription. A more cynical mind than mine would think that it might be a way of upselling the more expensive human transcription service, but that’s probably not the case. If you are able to load the custom dictionary before you begin your transcription, this will likely vanish instantly.
Both tools handled numbers and figures well and were able to identify the correct values of weight. There was a little issue with formatting when it came to AmberScript, writing numbers like “6 to 700” and “4052,” whereas tl;dv was able to display these as solid, clean, separate figures. Nothing that would alter the overall meaning and cause any issues.
Another real bonus across both tools is that there was a distinct lack of hallucination or invented sentences. I’ve tested a number of tools across the years, and hallucination and strange looping cycles can happen fairly frequently. Every error on both tools was genuinely a misheard phrase rather than the AI trying to slot something neat into place. Of the two types of error, this is the one that will cause less hassle.
Accuracy-wise, overall it was fairly solid, but the naming issue is something that could be a real issue or factor in day-to-day life, and would require a bit of legwork inputting into the custom dictionary beforehand.
Where AmberScript Wins
There was one area where AmberScript did slightly better than tl;dv: speaker labels. On my seven-speaker clip, AmberScript identified all seven voices. tl;dv slightly slipped slightly on one block.
The audio I provided was crystal clear, professionally recorded, and of good quality. It was people speaking for clear long stretches, and that is not how most recordings sound. Digging around online and looking at reviews of the tool, multiple AmberScript reviewers report the opposite of what I found. The tool collapses multiple voices into one the minute the audio gets a bit messy, deeply accented or people start to speak over each other. One reviewer had four speakers and got a single label for all of them.
Perhaps I got lucky and just picked the “right” kind of clip, but that’s something to be aware of. On chaotic real-world audio, treat that strength as unproven until you test your own files. The free ten minutes are enough to find out before you pay.
AmberScript Pricing
This is where things can get expensive. AmberScript pricing is billed by audio hours and not by seats. This is great if you are doing file-by-file. For five hours a month, you are looking at €19, ten hours is €29, and it goes up to €49 for 25. There is also the option to buy credits at around €10 per hour, dropping to roughly €9 as you buy more.
The free trial gives you 10 minutes of automatic transcription, and while I did pay to upgrade my credits, it’s not a lot of time to base a purchasing decision on.
On looking a little deeper, I found that the subscription plans are prepaid and auto-renewing. The monthly plans are marked “cancel anytime” on the pricing page, and the small print confirms a subscription renews automatically unless you cancel before the next renewal date. Cancelling stops the next renewal, but you won’t get a refund for the period you’re already in. Unused hours also do not roll over and expire each month.
One other thing is that they don’t offer refunds for what you deem to be disappointing transcription quality, and they will judge the quality of the audio you submit at their own discretion. Caveat emptor.
To be frank, none of this is particularly unusual or shady for this sector and none of it is hidden. But a casual buyer may be caught out on this. So this may be a case of being conservative and then topping up if you need to.
AmberScript Alternatives
If you are looking for just transcription, then there are other alternatives to AmberScript out there that could be worth looking at. For human-grade accuracy, Rev covers the same automatic-plus-human options. For anything media-heavy, Sonix and Trint are other tools in the market.
If you are looking for something to accurately record meetings and transcribe in multiple languages, then tl;dv is still your best option.
AmberScript vs tl;dv
While I’ve tested both of these tools out based on uploading audio, and they often get compared against each other, AmberScript and tl;dv are two totally different tools with different functions.
AmberScript is for files you already have. You record something, upload it, get it transcribed.
tl;dv does so much more. It is built around the meeting itself.
It records and transcribes meetings live, can be run with a bot in the meeting or even bot-free using the desktop recorder. And once that meeting happens, you have a transcription, as well as AI notes, speaker-level insights, and even a native MCP server that pulls meeting data into tools like Claude and ChatGPT. tl;dv also runs a free plan that includes recording and transcription, plus its mobile app when you’re away from your desk.
On this particular experiment, transcription from an audio file, even though it’s AmberScript’s entire premise, tl;dv came out with a better result when it came to word accuracy and entity recognition.
Security-wise, tl;dv is a European tool, is GDPR-compliant, holds ISO 27001 certification, and stores data in Europe.
The one thing I did find when researching into whether or not AmberScript trains AI on your data (something that tl;dv does not do) was a slight discrepancy between its marketing line and its privacy policy. The AmberScript G2 profile claims it uses the data its users generate to train its speech recognition engines. The privacy policy states plainly that generated audio and transcripts are never used to train language models, and that this is the default for all customers.
Those can’t both be the whole story. My read is that the marketing line describes AmberScript’s separate paid data-annotation service, where it builds training datasets on commissioned audio, and that language leaked into the general profile copy. On the product you actually sign up for, both AmberScript and tl;dv keep your uploads out of model training.
How tl;dv and Amberscript compare across the jobs a team actually needs.
| Capability | tl;dv | Amberscript |
|---|---|---|
| Live meeting recording (bot or bot-free) | Yes | No |
| Upload and transcribe existing files | Yes | Yes |
| Word accuracy (this test, WER) | 31.6% | 39% |
| Entity and name recognition | Consistent | Missed / varied |
| AI meeting notes and summaries | Yes | No |
| Speaker intelligence in live meetings | Yes | Positional labels only |
| Native MCP server (Claude, ChatGPT) | Yes | No |
| CRM enrichment and 5000+ integrations | Yes | No |
| Free plan | Unlimited recording & transcription | 10-min trial only |
| GDPR compliant, ISO 27001, EU data | Yes | Yes |
| Human transcription option | No | Yes |
My call is if you want to upload files and get them transcribed and then perhaps get a human to check them over for an additional cost, AmberScript may be suitable. For a tool that you can upload files to transcribe, while also running meetings, enriching your CRM system, using the MCP to gather detailed insight from your data and action them, coach your sales team and much more, then tl;dv is the better pick.
Is AmberScript A Good Transcription Tool?
AmberScript is a very capable upload-and-transcribe tool that offers great security and data residency in Europe. It’s fairly simple to use, easy to upload and offers a wide range of formats that you can upload to it. The automatic tier is fast and is easy to edit, with the ability to playback your recording at varying speeds, touch a word to jump to the section and more. It struggles with entity detection and the way that you “charge” your account for uploads can feel a little bit archaic. But if your main concern is transcription and subtitling options, it’s a solid bet.
Personally, for me, I would use tl;dv to do this based on my own experience of how easy and accurate the tool is to upload files to, but also because of the wide range of features it offers.
If you would like to give tl;dv a try today and upload your own files to test, then feel free to try our free tier.
FAQs About AmberScript
How accurate is AmberScript?
Amberscript’s automatic transcription works as an internal draft but needs editing before publishing. Against an official UK parliamentary transcript I measured a 39% word error rate on formal audio, versus a cleaned reference. Amberscript markets 90% on clean audio, 99% human-edited.
Can I cancel AmberScript anytime?
Yes. Amberscript’s monthly plans are marked “cancel anytime” on the pricing page. Subscriptions are prepaid and auto-renew, so cancelling stops the next renewal but won’t refund the current period. Unused hours expire rather than rolling over.
How much does AmberScript cost?
Is AmberScript's AI or human transcription better?
AmberScript vs tl;dv: which should I use?
Use Amberscript to transcribe files you already have, use tl;dv to record and transcribe meetings live. Amberscript suits interviews, research, and subtitles. tl;dv covers transcription and adds live meeting capture, summaries, and AI notes. Amberscript edged speaker labels in my test.
