Marathi speech-to-text
Not usableWe sent 30 Marathi test clips through this API. Whisper Large v3 made 80% word errors, which is not good enough to read as a transcript. This page shows what came back so you can judge for yourself.
Measured accuracy
| Model | Word error rate | Character error rate | Clips with no errors | Empty transcripts | Speed |
|---|---|---|---|---|---|
Whisper Large v3openai/whisper-large-v3 | 80% | 33% | 0 of 30 | 2 of 30 | 0.182× real time |
Not usable. Most words come out wrong. Do not build on it for this language. The clips are clean, single-speaker read speech, so treat this as a best case for your own recordings. How we measured, and all 83 languages.
What a transcript looks like
The clip with the median error rate from the test, reference text against what the API returned.
| Spoken | त्याच्या 78 शिफारशींपैकी प्रथम म्हणजे विरोधी हस्तक्षेपांविरूद्ध इराकच्या सीमा सुरक्षित करण्यासाठी आणि त्याच्या शेजारील राष्ट्रांसोबत राजनैतिक संबंध पुन्हा स्थापित करण्यासाठी या वर्षाच्या अखेरीस एक नवीन परराष्ट्रसंबंधाविषयक कुशलतापूर्वक पुढाकार घ्यावा. |
| Whisper Large v3 | त्याच्या सत्यांशी शिफारिशीन पैकी प्रथम मन्जे विरोधी हस्तक शेपान विरुध इराक्च्या सीमा सुरक्षित करने साथी आणि त्याच्या शेजारिल राष्ट्रान सोबत राजनाइतिक संबंध पुन्ह स्थापित करने साथी या वर्षाच्या अखेरी एक न� |
Language detection
Send language=mr. Of 10 clips sent without it, only 8 were identified as Marathi; the others were labelled English (1), Hindi (1). When detection is wrong, Whisper writes the transcript in the language it guessed.
What to do instead
We would rather you know this before you build than after. For Marathi, look at services that train specifically on the language, such as Google Cloud Speech-to-Text or Azure AI Speech, both of which publish their language lists, or a Whisper checkpoint fine-tuned on Marathi from Hugging Face run on your own hardware. If your audio mixes Marathi with a well-supported language, the well-supported parts will still transcribe. You can confirm all of this on your own audio with the free credit on a new account: transcription playground.
FAQ
How accurate is Whisper for Marathi?
On 30 clean read-speech clips it made 80% word errors. Most words come out wrong. Do not build on it for this language. Real recordings with noise, accents or several speakers will score worse.
Does language auto-detection work for Marathi?
Not reliably. Of 10 clips sent without a language, 8 were identified as Marathi; the rest were labelled English (1), Hindi (1). A wrong label means the transcript is written in the wrong language, so always send language=mr.
Can it translate Marathi speech to English?
The API returns Marathi text. For English, transcribe first and translate the text with a chat model on the same key; with accuracy this low the translation will inherit the errors.