CoolFace
Apppublic

hugging-apps/gigaam-multilingual-asr

sourceHugging Faceupdated 2mo agoView on Hugging Face
2likes
App README

GigaAM Multilingual ASR

This Space demonstrates the `ai-sage/GigaAM-Multilingual` Conformer-based ASR models.

Supported languages

Russian, English, Kazakh, Kyrgyz, and Uzbek.

Model variants

  • —Large CTC (600M) — best quality, 24-layer Conformer encoder
  • —CTC (220M) — faster, 16-layer Conformer encoder

Both variants use character-wise CTC decoding with greedy decoding.

Usage

Upload an audio file or record from your microphone, then click Transcribe. Audio is automatically resampled to 16 kHz mono.