Founding offer — up to 50% off until December 31. Lock in your price and keep it for as long as you stay subscribed

How accurate is TranscriptGo? Our public test, language by language

Every number on this page comes from the live service, measured on a public speech test that anyone can download and repeat. We show where we are strong and where we are not.

30 free minutes every month · no credit card neededSee pricing

Words transcribed exactly right

20 recordings per language from Google's FLEURS test set, scored word by word against the sentence the speaker read. The right column is Whisper large-v3, the open speech model that many transcription services are built on, run on the same files.

Words transcribed exactly rightTranscriptGoWhisper large-v3
SpanishTranscriptGo: 98.6%Whisper large-v3: 97.9%
FrenchTranscriptGo: 95.3%Whisper large-v3: 94.1%
ArabicTranscriptGo: 94.2%Whisper large-v3: 91.2%
HindiTranscriptGo: 93.6%Whisper large-v3: 73.4%
UrduTranscriptGo: 80.9%, always in Urdu scriptWhisper large-v3: 77.0%
BanglaTranscriptGo: 87.3% letter for letter, about 97% with the right meaningWhisper large-v3: 28.7%

Measured on 19 November 2026 on the live TranscriptGo service: the average of two runs, because each run differs by about one point. Whisper large-v3 is the official open model from OpenAI (github.com/openai/whisper), run with full decoding on the same files. Clean, read speech is the best case: noisy rooms, phone audio and people talking over each other score lower on every tool.

1

Pick your language

Choose the spoken language when you upload. Each language runs on the speech engine that scored best for it in this test.

2

Upload your own file

The 30 free minutes every month are enough to test a real recording, with no credit card.

3

Check it against the audio

Click any line to hear that moment, then download Word, TXT, SRT or VTT.

How we tested

The recordings

FLEURS is a free speech dataset published by Google. Each recording is one person reading one sentence, and the dataset includes the exact sentence they read. We took the first 20 recordings from the test set of each language: Spanish (Latin America), French, Arabic (Egypt), Hindi, Urdu (Pakistan) and Bangla (India). That is about 400 to 600 words per language, from different speakers.

How a word is counted

A word counts as right only if it matches the reference exactly. Before comparing, we remove punctuation, treat digits in any script as the same digits, and for Arabic and Urdu remove the optional vowel marks (harakat), which is the standard way Arabic speech recognition is scored. Everything else counts: a misspelled name, a missing word, "1" written as "one".

Why Bangla shows two numbers

Bangla has many accepted spellings for the same word, and the reference sentences mix them. ভাল and ভালো, সরকারী and সরকারি, গভীর ভাবে and গভীরভাবে are all correct, but a strict count marks every one as an error. We read every difference by hand: 12 of 408 words actually changed the meaning, so about 97% of words were right. The strict letter-for-letter count is 87.3%.

What the remaining errors are

Most are rare names spelled by ear (a surname like Carpanedo, a place like Pohutukawa), numbers written as words instead of digits, and short words such as de and del. A human typist who has not seen the names written would miss many of the same ones.

What to expect on your own recordings

These are clean recordings of one person reading. Your results depend mostly on the audio: a voice close to the microphone with little background noise comes close to these numbers, while echo, music, phone calls and crosstalk lower them on every service. The best test is your own file: you get 30 free minutes every month.

Repeat it yourself

FLEURS is free to download on Hugging Face (google/fleurs). Upload the same recordings to any service you are considering and compare the text with the sentences in the dataset. If you get different numbers from ours, we would like to hear about it.

Everything you need, included

  • A different speech engine for each language, chosen by test
  • Measured on a public dataset anyone can repeat
  • Urdu always delivered in Urdu script
  • 99+ languages, translation into 99+
  • Word, TXT, SRT and VTT downloads
  • 30 free minutes every month

Frequently asked questions

How accurate is TranscriptGo?

On Google's public FLEURS test, 98.6% of Spanish words, 95.3% of French, 94.2% of Arabic and 93.6% of Hindi words came out exactly right. Bangla is about 97% right in meaning. Clean, read speech is the best case; noisy recordings score lower.

Is TranscriptGo more accurate than Whisper?

In every language we tested, yes. The biggest gaps are in Bangla (about 97% against 28.7%) and Hindi (93.6% against 73.4%). In Spanish the two are close: 98.6% against 97.9%.

Why not claim 99% accuracy like other services?

Because we measured it. A number without a public test behind it cannot be checked. Ours can: the dataset, the method and the date are on this page.

Does TranscriptGo label speakers?

No. TranscriptGo does not label speakers. Every line has a timestamp you can click to hear it.

How can I test it on my own audio?

Create a free account and upload a recording. You get 30 free minutes every month with no credit card.

Try it on your own recording

30 free minutes every month. Upload a file and see the transcript in minutes.

Transcribe now, free