File size: 1,120 Bytes
2859a48 |
1 2 3 4 5 6 7 8 9 10 11 12 13 14 15 16 17 18 19 20 21 22 23 24 25 26 27 28 29 30 31 32 33 34 |
import librosa
from transformers import Wav2Vec2ForCTC, AutoProcessor
import torch
ASR_SAMPLING_RATE = 16_000
MODEL_ID = "facebook/mms-1b-all"
processor = AutoProcessor.from_pretrained(MODEL_ID)
model = Wav2Vec2ForCTC.from_pretrained(MODEL_ID)
def transcribe(audio_source=None, microphone=None, file_upload=None):
audio_fp = file_upload if "upload" in str(audio_source or "").lower() else microphone
if audio_fp is None:
return "ERROR: You have to either use the microphone or upload an audio file"
audio_samples = librosa.load(audio_fp, sr=ASR_SAMPLING_RATE, mono=True)[0]
processor.tokenizer.set_target_lang("fao") # Set Faroese language
model.load_adapter("fao")
inputs = processor(audio_samples, sampling_rate=ASR_SAMPLING_RATE, return_tensors="pt")
# Set device
device = torch.device("cuda" if torch.cuda.is_available() else "cpu")
model.to(device)
inputs = inputs.to(device)
with torch.no_grad():
outputs = model(**inputs).logits
ids = torch.argmax(outputs, dim=-1)[0]
transcription = processor.decode(ids)
return transcription
|