---
language: mar
license: apache-2.0
tags:
- audio
- automatic-speech-recognition
- speech
- marathi
datasets:
- openslr
model-index:
- name: Marathi ASR
  results:
  - task:
      type: automatic-speech-recognition
      name: Speech Recognition
    dataset:
      name: Marathi OpenSLR Dataset
      type: openslr
    metrics:
      - name: Word Error Rate
        type: wer
        value: "Your WER here"  # Replace with your model's WER
---

# Marathi ASR Model

This is a fine-tuned Wav2Vec2-BERT model for Automatic Speech Recognition (ASR) in Marathi language.

## Model Details

- **Model Type:** Wav2Vec2-BERT for CTC
- **Language:** Marathi
- **Training Dataset:** OpenSLR Marathi Dataset
- **Last Updated:** April 16, 2025

## Usage

```python
from transformers import Wav2Vec2BertProcessor, Wav2Vec2BertForCTC
import torchaudio
import torch

# Load model and processor
processor = Wav2Vec2BertProcessor.from_pretrained("hriteshMaikap/marathi-asr-model")
model = Wav2Vec2BertForCTC.from_pretrained("hriteshMaikap/marathi-asr-model")

# Load audio
waveform, sample_rate = torchaudio.load("audio.wav")
# Resample if needed
if sample_rate != 16000:
    resampler = torchaudio.transforms.Resample(sample_rate, 16000)
    waveform = resampler(waveform)
    sample_rate = 16000
# Convert to mono if needed
if waveform.shape[0] > 1:
    waveform = torch.mean(waveform, dim=0, keepdim=True)
# Convert to numpy
speech_array = waveform.squeeze().numpy()

# Transcribe
inputs = processor(speech_array, sampling_rate=16000, return_tensors="pt")
with torch.no_grad():
    logits = model(inputs.input_features).logits
predicted_ids = torch.argmax(logits, dim=-1)
transcription = processor.decode(predicted_ids[0])

print(transcription)

Step	Training Loss	Validation Loss	Wer
300	0.211100	0.220232	0.183333
600	0.086900	0.172057	0.113889