def test_save_load_pretrained_additional_features(self): processor = Speech2TextProcessor( tokenizer=self.get_tokenizer(), feature_extractor=self.get_feature_extractor()) processor.save_pretrained(self.tmpdirname) tokenizer_add_kwargs = self.get_tokenizer(bos_token="(BOS)", eos_token="(EOS)") feature_extractor_add_kwargs = self.get_feature_extractor( do_normalize=False, padding_value=1.0) processor = Speech2TextProcessor.from_pretrained(self.tmpdirname, bos_token="(BOS)", eos_token="(EOS)", do_normalize=False, padding_value=1.0) self.assertEqual(processor.tokenizer.get_vocab(), tokenizer_add_kwargs.get_vocab()) self.assertIsInstance(processor.tokenizer, Speech2TextTokenizer) self.assertEqual(processor.feature_extractor.to_json_string(), feature_extractor_add_kwargs.to_json_string()) self.assertIsInstance(processor.feature_extractor, Speech2TextFeatureExtractor)
def s2t_predictions(audio_file): device = 'cuda' if torch.cuda.is_available() else 'cpu' audio_array = s2t_audio_to_array(audio_file) model = Speech2TextForConditionalGeneration.from_pretrained( "facebook/s2t-small-librispeech-asr").to(device).eval() processor = Speech2TextProcessor.from_pretrained( "facebook/s2t-small-librispeech-asr", do_upper_case=True) features = processor(audio_array, sampling_rate=16000, return_tensors="pt") input_features = features.input_features.to(device) attention_mask = features.attention_mask.to(device) gen_tokens = model.generate(input_ids=input_features) text = processor.batch_decode(gen_tokens, skip_special_tokens=True) return text
def test_save_load_pretrained_default(self): tokenizer = self.get_tokenizer() feature_extractor = self.get_feature_extractor() processor = Speech2TextProcessor(tokenizer=tokenizer, feature_extractor=feature_extractor) processor.save_pretrained(self.tmpdirname) processor = Speech2TextProcessor.from_pretrained(self.tmpdirname) self.assertEqual(processor.tokenizer.get_vocab(), tokenizer.get_vocab()) self.assertIsInstance(processor.tokenizer, Speech2TextTokenizer) self.assertEqual(processor.feature_extractor.to_json_string(), feature_extractor.to_json_string()) self.assertIsInstance(processor.feature_extractor, Speech2TextFeatureExtractor)
def default_processor(self): return Speech2TextProcessor.from_pretrained( "facebook/s2t-small-librispeech-asr")