| import torch | |
| import torchaudio | |
| from transformers import AutoProcessor, AutoModelForSpeechSeq2Seq | |
| class Inference: | |
| def __init__(self): | |
| self.processor = AutoProcessor.from_pretrained(".") | |
| self.model = AutoModelForSpeechSeq2Seq.from_pretrained(".") | |
| self.model.eval() | |
| def __call__(self, inputs): | |
| audio_path = inputs.get("inputs") | |
| if not audio_path: | |
| return {"error": "No audio provided."} | |
| waveform, sample_rate = torchaudio.load(audio_path) | |
| if sample_rate != 16000: | |
| waveform = torchaudio.transforms.Resample(sample_rate, 16000)(waveform) | |
| inputs = self.processor(waveform.squeeze(), sampling_rate=16000, return_tensors="pt") | |
| with torch.no_grad(): | |
| generated_ids = self.model.generate(inputs["input_features"]) | |
| text = self.processor.batch_decode(generated_ids, skip_special_tokens=True)[0] | |
| return {"text": text} | |