MERaLiON 2 10B ASR
MERaLiON 2 10B ASR is the generation 2 checkpoint tuned for speech recognition. It is a separate weight release from the general 10B AudioLLM. It focuses on transcription in English and Southeast Asian languages.
A dedicated transcription checkpoint
The release targets speech recognition, including regional speech and mixed-language input. Its configuration still uses a 42-layer Gemma decoder. ASR specialisation changes the trained checkpoint; it is not a smaller encoder-only recognizer.MERaLiON 2 10B ASR published configurationMERaLiON 2 10B ASR official model card
Available earlier-generation weights
The downloadable checkpoint remains useful for reproducing generation 2 results. Its public license has the generation 2 conditions, including upstream terms, redistribution notices and acknowledgement. A current hosted endpoint for these exact weights is not established.MERaLiON Public Licence v3 for generation 2
Compare recognition errors directly
Practical guidance. Use this model as a transcription baseline, especially when assessing the smaller generation 3 release. Compare names, dialect words and language switches on identical clips. General audio-question accuracy does not replace a transcription error analysis.MERaLiON 2 10B ASR official model card
Specifications & access
| Detail | Value and conditions |
|---|---|
| Access | Downloadable |
| Model identifiers | MERaLiON-2-10B-ASR |
| Context window | 8,192 tokens Maximum text-decoder positions in the published configuration; audio consumes representation space and has separate duration guidance. MERaLiON 2 10B ASR published configuration |
| Languages | English, Chinese, Malay, Tamil, Indonesian, Thai and Vietnamese.MERaLiON 2 10B ASR official model card |
| Components | Whisper large-v3 speech encoder with Gemma 2 9B IT text decoder.MERaLiON 2 10B ASR published configuration |
| Recommended ASR clip length | 30 seconds, using 16 kHz mono audio.MERaLiON 2 10B ASR official model card |
Related models
- MERaLiON 2 10B
MERaLiON / A*STAR IAIC · MERaLiON 2
A general audio-understanding model that combines a localized Whisper encoder with Gemma 2 text generation.
Launch date not confirmedDownloadable - MERaLiON 3 ASR
MERaLiON / A*STAR IAIC · MERaLiON 3 ASR
A regional speech recognition release with a public 3B-class checkpoint and a separately described hosted service.
Launch date not confirmedDownloadable · API · Provider product