Fact · Languages and accents
1languages
AuroraAI Research's model card says AuroraASR-1M was trained on the full LibriSpeech dataset in English only, so the model's count is one language.
Quote found word for word on the live source, last checked October 23, 2026
In the source's own words
“Trained on the full LibriSpeech dataset for 379 epochs, English only, 16 kHz mono input”
AuroraAI-Research/AuroraASR-1M · Hugging Face, AuroraAI Research, Hugging Face (model card by AuroraAI Research), 2026, Model card, training description.
Open the original- What it measures
- The model card describes a conformer model with CTC decoding and a 256-piece BPE vocabulary. Its dataset tags list English, and its benchmarks use the LibriSpeech test sets.
- What it does not show
- The card gives no list of supported languages, so the count of one comes from its statement that training was English only. The card says nothing about other languages, so this fact claims no figure for them.
In the data
A row of How Many Languages Speech Recognition Supports, as AuroraASR-1M, English only: 1 languages.
- System
- AuroraASR-1M
- Maker
- AuroraAI Research
- Kind
- Open model
- What is counted
- Languages
- On device
- Not stated
- As of
- 2026-10
- Note
- 1 means the card names English as the only training language.
Source
Used to calculate
- In the languages supported by speech recognition systems dataset, the 75 published figures for “Open model” have a median of 25 languages, ranging from 1 to 1,672.
- The languages supported by speech recognition systems dataset holds 280 data points, each a published figure traced to its source.
- The languages supported by speech recognition systems dataset draws on 233 distinct sources.