|
--- |
|
license: apache-2.0 |
|
datasets: |
|
- mozilla-foundation/common_voice_13_0 |
|
language: |
|
- sw |
|
library_name: transformers |
|
tags: |
|
- language |
|
- swahili |
|
- asr |
|
- mozilla |
|
--- |
|
|
|
## How To Use |
|
|
|
Here is a plug and play inference code |
|
```python |
|
from transformers import WhisperProcessor, WhisperForConditionalGeneration |
|
|
|
processor = WhisperProcessor.from_pretrained("eddiegulay/Whisperer_Mozilla_Sw_2000") |
|
model = WhisperForConditionalGeneration.from_pretrained("eddiegulay/Whisperer_Mozilla_Sw_2000") |
|
forced_decoder_ids = processor.get_decoder_prompt_ids(language="swahili", task="transcribe") |
|
|
|
def transcribe(audio_path): |
|
# Load the audio file |
|
audio_input, sample_rate = torchaudio.load(audio_path) |
|
target_sample_rate = 16000 |
|
audio_input = torchaudio.transforms.Resample(orig_freq=sample_rate, new_freq=target_sample_rate)(audio_input) |
|
|
|
# Preprocess the audio data |
|
input_features = processor(audio_input[0], sampling_rate=target_sample_rate, return_tensors="pt").input_features |
|
|
|
# generate token ids |
|
predicted_ids = model.generate(input_features, forced_decoder_ids=forced_decoder_ids) |
|
|
|
# Perform inference and transcribe |
|
transcription = processor.batch_decode(predicted_ids, skip_special_tokens=True) |
|
|
|
return transcription |
|
|
|
|
|
|
|
transcribe('your_audio_file.mp3') |
|
``` |
