Serbian speech-to-text
FairWe sent 30 Serbian test clips through this API. Whisper Large v3 made 21% word errors, about one word in 5. $0.002 per audio minute, word timestamps, SRT and VTT subtitles, files up to 100 MB.
Measured accuracy
| Model | Word error rate | Character error rate | Clips with no errors | Empty transcripts | Speed |
|---|---|---|---|---|---|
Whisper Large v3openai/whisper-large-v3 | 21% | 12% | 2 of 30 | 0 of 30 | 0.081× real time |
Fair. Gets the content across, but needs a human pass before it is published or quoted. Whisper writes Serbian in Latin script; the test references are Cyrillic, so they were transliterated to Latin before scoring. 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 | Због подводне топологије, повратни ток је концентрисан на неколико дубљих делова и тамо се може створити брза струја до дубоке воде. |
| Whisper Large v3 | Zbog podvodne topologije povratni tok je koncentrisan na nekoliko dubih delova i tamo se može stvoriti brzo strulja do duboke vode. |
Language detection
Send language=sr. Of 10 clips sent without it, only 5 were identified as Serbian; the others were labelled Bosnian (3), English (2). When detection is wrong, Whisper writes the transcript in the language it guessed.
Quickstart
The request is the OpenAI transcription request with a different base URL. language is optional in the API but, as shown above, not optional in practice for Serbian.
curl https://api.inferenceapis.com/v1/audio/transcriptions \
-H "Authorization: Bearer $INFERENCE_API_KEY" \
-F "model=openai/whisper-large-v3" \
-F "language=sr" \
-F "file=@recording.mp3" \
-F "response_format=verbose_json"import os
from openai import OpenAI
client = OpenAI(base_url="https://api.inferenceapis.com/v1", api_key=os.environ["INFERENCE_API_KEY"])
with open("recording.mp3", "rb") as f:
r = client.audio.transcriptions.create(model="openai/whisper-large-v3", file=f, language="sr", response_format="verbose_json")
print(r.text)Subtitles and timestamps
response_format=srt or vtt returns a subtitle file with cues of up to two lines, broken at sentence ends and pauses. verbose_json with timestamp_granularities[]=word returns a start and end time for every word if you want to build your own.
Cost
$0.002 per minute of audio on either model: $0.12 for an hour, $12 for a hundred hours. Duration is measured from the file you upload. There is no daily cap on minutes and no subscription; enough to run your own Serbian test.
Notes for Serbian
- Output is in Latin script. If you need Cyrillic, transliterate the text afterwards; the mapping is one to one.
FAQ
How accurate is Whisper for Serbian?
On 30 clean read-speech clips it made 21% word errors, about one word in 5. Gets the content across, but needs a human pass before it is published or quoted. Real recordings with noise, accents or several speakers will score worse.
Does language auto-detection work for Serbian?
Not reliably. Of 10 clips sent without a language, 5 were identified as Serbian; the rest were labelled Bosnian (3), English (2). A wrong label means the transcript is written in the wrong language, so always send language=sr.
Can it translate Serbian speech to English?
The API returns Serbian text. For English, transcribe first and translate the text with a chat model on the same key.
Related languages: Croatian (11%) · Bosnian (13%) · Macedonian (16%) · Bulgarian (8.4%)