MERaLiON SpeechEncoder 2 ASR-CTC
This recognizer fine-tunes SpeechEncoder 2 with supervised speech data and a CTC objective. It is a distinct trained ASR system described in the encoder card.
Access not confirmed
A trained recognizer beyond the base encoder
The card reports ASR evaluation and links a dedicated demonstration. It distinguishes this trained system from the base encoder and its example fine-tuning code.MERaLiON SpeechEncoder 2 official model card
The former demo does not confirm current access
The linked demo returned 404 during review. A current download, service mapping and release-specific license were not established. The live transcription endpoint instead names MERaLiON-3-3B-ASR-CTM. CTC and CTM are not treated as aliases.MERaLiON public OpenAPI specification
Reproduction needs the trained head
Practical guidance. Do not use the base encoder download as evidence that the reported recognizer is available. Reproducing its ASR results requires the trained task weights, vocabulary and decoding settings. Keep this historical system separate until those materials or a verified service route are supplied.
Specifications & access
| Detail | Value and conditions |
|---|---|
| Status | Access not confirmed |
| Model identifiers | MERaLiON-SpeechEncoder-2-ASR-CTC |
| Training relationship | Supervised adaptation of SpeechEncoder 2.MERaLiON SpeechEncoder 2 official model card |
| Base encoder | 24 layers and 1,024 hidden dimensions; these are parent-encoder specifications.MERaLiON SpeechEncoder 2 published configuration |
| Current download | No separate ASR-CTC weight repository appears in the reviewed official catalog.MERaLiON official model catalog |
Related models
- 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 - MERaLiON SpeechEncoder 2
MERaLiON / A*STAR IAIC · MERaLiON SpeechEncoder
A 630M-class multilingual speech encoder trained from scratch to provide acoustic features for downstream tasks.
Launch date not confirmedDownloadable