The Hindi Automatic Speech Recognition (ASR) model is designed to transcribe spoken Hindi into text from 16kHz mono-channel audio files. It leverages a Conformer-Large architecture with 120 million parameters, providing robust and efficient speech recognition. The model utilizes a hybrid CTC-RNNT decoder for accurate and reliable transcription. With 17 conformer blocks and a model dimension of 512, it effectively captures linguistic and acoustic variations, making it well-suited for various speech-to-text applications in Hindi.
MIT
Sector Agnostic
AI4Bharat
N.A.
N.A.
N.A.
N.A.
N.A.
N.A.
N.A.
N.A.
Open
21/02/25 13:21:45
N.A.
0
© 2026 - Copyright AIKosh. All rights reserved.