VibeVoice ASR — Mongolian Speech-to-Text Benchmark
Real results from the VibeVoice ASR 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.
258/265 samples transcribed · 100% success rate
Lower is better · across 265 samples
Pricing
VibeVoice ASR list pricing for batch transcription. No discounts, no negotiated rates applied — the raw per-minute rate.
Pricing source: VibeVoice ASR public pricing. Duudlaga Flow is shown for context only — this page isolates VibeVoice ASR 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 VibeVoice ASR 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 | 57/59 | 98.9% | 59.2% | 1.1% | 0.56× | 318.8s | 574.1s |
| Shunya Labs Mongolian Speech | Hugging Face → | 60 | 59/60 | 114.2% | 58% | -14.2% | 0.63× | 630.9s | 1003.2s |
| Common Voice 20 (MN) | Hugging Face → | 54 | 52/54 | 103.8% | 62.2% | -3.8% | 0.57× | 258.2s | 452.9s |
| Modern Voice | — | 92 | 90/92 | 109.4% | 63.4% | -9.4% | 0.43× | 580.6s | 1338.1s |
WER vs speed — per sample
Each dot is one audio sample. The sweet spot is the bottom-left (low error, fast). VibeVoice ASR 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 VibeVoice ASR got wrong, in red — no strikethrough/swap gymnastics, just the mistakes.
| # | Audio | Sample | Dataset | Expected (ground truth) | VibeVoice ASR result | WER | CER | I/D/S |
|---|---|---|---|---|---|---|---|---|
| 1 | btsee_0001 | Common Voice 24 (MN) | Гэхдээ амьсгал хураахаасаа өмнө танд мэдэж байгаагаа хэлье. | Китэйм солонгос рахасаа, мөн тэнд мэдэгдэхгүй байна шүү дээ. | 112.5% | 63.8% | 1/0/8 | |
| 2 | btsee_0002 | Common Voice 24 (MN) | Надад заяасан аз жаргал гэдэг ердөө гуравхан сарын хугацаатай байсан гэж үү? | Надад зээс санагч ралхитэг юрдогорхон сарын хутагдай байсан гэж байна. | 66.7% | 42.7% | 0/2/6 | |
| 3 | btsee_0003 | Common Voice 24 (MN) | Одоо бид өөрсдөө өвчин эмгэгээсээ салахыг хичээцгээе. | Атаа титурстуучин бийгээс ард хичээсгээд. | 100.0% | 55.8% | 0/2/5 | |
| 4 | btsee_0004 | Common Voice 24 (MN) | Би бол голдуу хээрээр гэр, хэцээр дэр хийж явдаг хүн. | Би бас алтуу хэдэр гэр, эцэрг дэргий чи аптахуун. | 80.0% | 43.1% | 0/1/7 | |
| 5 | btsee_0005 | Common Voice 24 (MN) | Хан хурмаст уурлаж, Болдоггүй Бор өвгөнийг хор луугаараа ниргүүлэхээр явуулжээ. | Ханхан мэдээлэл, бүртгэл баруунд нэг хэсэг, зөөвөр нэг рус шүршгээ. | 100.0% | 72.7% | 0/0/10 | |
| 6 | btsee_0006 | Common Voice 24 (MN) | Алив наашаа ороод ир гээд гэртээ оров. | Аз жүн байш, ороудыр, гейт кэсэдэ ороудыр. | 100.0% | 62.2% | 0/0/7 | |
| 7 | btsee_0007 | Common Voice 24 (MN) | Өө өндөр дээдэс таны тухайд би баталж чадахгүй. | О, өндөртөөс танигдаж байгаа бие баталгаатай байна шүү дээ. | 112.5% | 76.1% | 1/0/8 | |
| 8 | btsee_0008 | Common Voice 24 (MN) | Харин гурав дахь удаагаас эхлэн хүмүүсийг сонирхож эхлэв. | Харин, урвдаг, удахас, яглэг, хүмүүсийг санжуулж лүү. | 75.0% | 35.7% | 0/1/5 | |
| 9 | btsee_0009 | Common Voice 24 (MN) | Та нар очингуутаа шөл л өгч үз. | Танар, очонгутаа, шүлэл, юг, чүүдз. | 100.0% | 30.0% | 0/2/5 | |
| 10 | btsee_0010 | Common Voice 24 (MN) | Ерөөсөө литр үйлдвэрээсээ салаагүй явсан юм чинь. | Ирсээд үүрдүүр сэсэлтэй өгөөд сэгчихээд. | 100.0% | 72.9% | 0/2/5 |
Methodology
How these numbers were produced.
Provider: VibeVoice ASR (Microsoft VibeVoice-ASR (7.6B, MIT) run locally — a long-form speech LLM covering 50+ languages. Needs a Mongolian context prompt: without it, language auto-detection picks the wrong script entirely.).
Endpoint local (transformers). Language mn. Operating point microsoft/VibeVoice-ASR-HF (Mongolian context prompt, transcription_only) on mps. Diarization none.
Datasets: Common Voice 24 (MN), Shunya Labs Mongolian Speech, Common Voice 20 (MN), Modern Voice — 265 samples, 1788.5s 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 VibeVoice ASR 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.