GPT-4o Transcribe — Mongolian Speech-to-Text Benchmark
Real results from the GPT-4o Transcribe Batch API (enhanced operating point, language mn) evaluated across 4 Mongolian datasets and 265 audio samples. Every number below is measured — not marketing. WER, speed, and pricing are shown as-is, brutally honest.
265/265 samples transcribed · 100% success rate
Lower is better · across 265 samples
Pricing
GPT-4o Transcribe list pricing for batch transcription. No discounts, no negotiated rates applied — the raw per-minute rate.
Pricing source: GPT-4o Transcribe public pricing. Duudlaga Flow is shown for context only — this page isolates GPT-4o Transcribe so the number is not padded by our own product.
WER & CER by dataset
Word and character error rates per dataset. Lower is better — and these are the real GPT-4o Transcribe numbers, which are weak on Common Voice 24.
Dataset summary
Aggregate accuracy, speed, and timing for each dataset.
| Dataset | Source | Samples | Success | WER | CER | Accuracy | Speed | Audio (s) | Proc (s) |
|---|---|---|---|---|---|---|---|---|---|
| Common Voice 24 (MN) | Hugging Face → | 59 | 59/59 | 61.3% | 31.1% | 38.7% | 2.73× | 330.6s | 121.2s |
| Shunya Labs Mongolian Speech | Hugging Face → | 60 | 60/60 | 41.7% | 19.4% | 58.3% | 1.67× | 645.1s | 386.8s |
| Common Voice 20 (MN) | Hugging Face → | 54 | 54/54 | 65.9% | 36.4% | 34.1% | 2.01× | 269.8s | 133.9s |
| Modern Voice | — | 92 | 92/92 | 47.4% | 25.5% | 52.6% | 2.59× | 603.8s | 233.4s |
WER vs speed — per sample
Each dot is one audio sample. The sweet spot is the bottom-left (low error, fast). GPT-4o Transcribe sits high on error for many Common Voice 24 samples.
Per-sample results
Ground truth shown verbatim in the Expected column. The result column highlights only the words GPT-4o Transcribe got wrong, in red — no strikethrough/swap gymnastics, just the mistakes.
| # | Audio | Sample | Dataset | Expected (ground truth) | GPT-4o Transcribe result | WER | CER | I/D/S |
|---|---|---|---|---|---|---|---|---|
| 1 | btsee_0001 | Common Voice 24 (MN) | Гэхдээ амьсгал хураахаасаа өмнө танд мэдэж байгаагаа хэлье. | Гэхдээ амьсгал хураахаасаа өмнө танд мэдэж яваагаа хэлье. | 12.5% | 6.9% | 0/0/1 | |
| 2 | btsee_0002 | Common Voice 24 (MN) | Надад заяасан аз жаргал гэдэг ердөө гуравхан сарын хугацаатай байсан гэж үү? | Надад сая санагдаж байгаа гэхэд ердөө гурван сарийн хугацаатай байсан л гэж үү? | 58.3% | 22.7% | 1/0/6 | |
| 3 | btsee_0003 | Common Voice 24 (MN) | Одоо бид өөрсдөө өвчин эмгэгээсээ салахыг хичээцгээе. | Одоо би дуртай хүнтэйгээ салахыг хүсч байгаа. | 85.7% | 51.9% | 0/0/6 | |
| 4 | btsee_0004 | Common Voice 24 (MN) | Би бол голдуу хээрээр гэр, хэцээр дэр хийж явдаг хүн. | Би бас голдуу хэрээр гэр, хэцэр дээр хийж явдаг хүн. | 40.0% | 9.8% | 0/0/4 | |
| 5 | btsee_0005 | Common Voice 24 (MN) | Хан хурмаст уурлаж, Болдоггүй Бор өвгөнийг хор луугаараа ниргүүлэхээр явуулжээ. | Хаварын сүүлчээс барлагийн хоёр зоогоор нэргүүл шарив. | 100.0% | 66.2% | 0/3/7 | |
| 6 | btsee_0006 | Common Voice 24 (MN) | Алив наашаа ороод ир гээд гэртээ оров. | Айсин наашаараад ир, гэж гэртээ оров. | 57.1% | 24.3% | 0/1/3 | |
| 7 | btsee_0007 | Common Voice 24 (MN) | Өө өндөр дээдэс таны тухайд би баталж чадахгүй. | Оо, өндөртэд таныг тохойод би баталж чадахгүй. | 62.5% | 23.9% | 0/1/4 | |
| 8 | btsee_0008 | Common Voice 24 (MN) | Харин гурав дахь удаагаас эхлэн хүмүүсийг сонирхож эхлэв. | Харин 33 дахь удаагаас эхлэн хүмүүсийг сонирхож эхлэв. | 12.5% | 8.9% | 0/0/1 | |
| 9 | btsee_0009 | Common Voice 24 (MN) | Та нар очингуутаа шөл л өгч үз. | Та нар орчлонгоудаа шүүлт өгч үз. | 42.9% | 30.0% | 0/1/2 | |
| 10 | btsee_0010 | Common Voice 24 (MN) | Ерөөсөө литр үйлдвэрээсээ салаагүй явсан юм чинь. | Ирүүсөөл үлдүүрээсээ салаагүй явсна юм чинь. | 57.1% | 25.0% | 0/1/3 |
Methodology
How these numbers were produced.
Provider: GPT-4o Transcribe (OpenAI gpt-4o-transcribe hosted transcription (Mongolian prompt hint; 'mn' is not an accepted language code).).
Endpoint https://api.openai.com/v1/audio/transcriptions. Language mn. Operating point gpt-4o-transcribe. Diarization none.
Datasets: Common Voice 24 (MN), Shunya Labs Mongolian Speech, Common Voice 20 (MN), Modern Voice — 265 samples, 1849.3s of audio total.
- Common Voice 24 (MN)https://huggingface.co/datasets/btsee/common-voices-24-mn
- Shunya Labs Mongolian Speechhttps://huggingface.co/datasets/shunyalabs/mongolian-speech-dataset
- Common Voice 20 (MN)https://huggingface.co/datasets/warmestman/common-voice-20-mn-normalized
- Modern Voice—
Metrics: WER and CER are computed with a standard word/character Levenshtein alignment, normalized for case and punctuation. Accuracy = 100 − WER. Speed factor = audio duration ÷ processing time (× realtime). All requests are real GPT-4o Transcribe Batch API calls, not cached or simulated.
Diff highlighting: The result column aligns to the ground truth and colors every substitution and insertion red. Deletions (words missing from the result) are not shown in the result column — the Expected column already holds the full ground truth as-is.