You can deploy this model and start earning money today!

Prashant2024/
asr-hi-openvino-f16

Name: asr-hi-openvino-f16
Availability: InStock

automatic-speech-recognition

Views

No views yet

Hindi ASR (OpenVINO IR)

A Conformer-CTC acoustic model for Hindi speech recognition (AI4Bharat IndicConformer lineage), converted to OpenVINO Intermediate Representation (FP16) for CPU inference.

Licensed Apache-2.0, per AI4Bharat/NeMo, the base training codebase/checkpoint source for this model.

Files

File	Description
`hi.xml`	OpenVINO IR model topology
`hi.bin`	OpenVINO IR model weights (FP16)
`hi.json`	Token vocabulary - a JSON array where each token's index is its model output id. Word-boundary marker is `▁` (U+2581, SentencePiece convention).

Blank token id (for CTC decoding) is len(vocab), i.e. one past the last entry in hi.json - it is not itself listed in the file.

Model details

Architecture: Conformer encoder + CTC head (token classification over the vocabulary above)
Input: log-mel spectrogram, 80 mel bins
Frontend config (must match at inference time):
- Sample rate: 16 kHz, mono
- n_fft=512, hop_length=160, win_length=400, hann window
- Pre-emphasis coefficient: 0.97
- Per-feature (per mel-bin) mean/std normalization after log compression
Outputs: per-frame logits over vocab_size + 1 (vocabulary + blank)
Encoder time-subsampling: ~4x (mel frames -> output frames)

Usage

python

import numpy as np
import openvino as ov

core = ov.Core()
model = core.read_model("hi.xml")
compiled = core.compile_model(model, "CPU")
infer_request = compiled.create_infer_request()

# mel_batch: float16 [1, 80, T] log-mel spectrogram built with the frontend
# config above; mel_length: int64 [1] = T
infer_request.infer(inputs={"audio_signal": mel_batch, "length": mel_length})
logits = infer_request.get_output_tensor(0).data  # [1, T', vocab_size + 1]

ids = np.argmax(logits[0], axis=-1)

Decode with standard CTC greedy decoding: collapse consecutive duplicate ids, then drop the blank id (len(vocab)), then map remaining ids through hi.json and replace ▁ with a space.

Intended use

Real-time and offline Hindi dictation / transcription. Not evaluated for noisy, far-field, or multi-speaker audio.

Prashant2024/asr-hi-openvino-f16