metadata
library_name: transformers
tags:
- persian
- whisper-base
- whisper
- farsi
- Neura
- NeuraSpeech
license: apache-2.0
language:
- fa
pipeline_tag: automatic-speech-recognition
Model Description
- Developed by: Neura company
- Funded by: Neura
- Model type: Whisper Base
- Language(s) (NLP): Persian
Model Architecture
This model uses a FastConformer-TDT architecture. FastConformer [1] is an optimized version of the Conformer model with 8x depthwise-separable convolutional downsampling. You may find more information on the details of FastConformer here: Fast-Conformer Model. Fast Conformer with Linearly Scalable Attention for Efficient Speech Recognition.
Uses
Check out the Google Colab demo to run NeuraSpeech ASR on a free-tier Google Colab instance: 
make sure these packages are installed:
from IPython.display import Audio, display
display(Audio('persian_audio.mp3', rate = 32_000,autoplay=True))
from transformers import WhisperProcessor, WhisperForConditionalGeneration
import librosa
# load model and processor
processor = WhisperProcessor.from_pretrained("Neurai/NeuraSpeech_WhisperBase")
model = WhisperForConditionalGeneration.from_pretrained("Neurai/NeuraSpeech_WhisperBase")
forced_decoder_ids = processor.get_decoder_prompt_ids(language="fa", task="transcribe")
array, sample_rate = librosa.load('persian_audio.mp3', sr=16000,mono=True)
sr = 16000
array = librosa.to_mono(array)
array = librosa.resample(array, orig_sr=sample_rate, target_sr=16000)
input_features = processor(array, sampling_rate=sr, return_tensors="pt").input_features
# generate token ids
predicted_ids = model.generate(input_features)
# decode token ids to text
transcription = processor.batch_decode(predicted_ids,)
transcription = processor.batch_decode(predicted_ids, skip_special_tokens=True)
print(transcription)
trascribed text :
او خواهان آزاد کردن بردگان بود
More Information
Model Card Authors
Esmaeil Zahedi, Mohsen Yazdinejad