Spaces:
Running
Running
| import gradio as gr | |
| from transformers import Wav2Vec2ForCTC, AutoProcessor | |
| import torch | |
| import librosa | |
| import json | |
| with open('ISO_codes.json', 'r') as file: | |
| iso_codes = json.load(file) | |
| languages = list(iso_codes.keys()) | |
| model_id = "facebook/mms-1b-all" | |
| processor = AutoProcessor.from_pretrained(model_id) | |
| model = Wav2Vec2ForCTC.from_pretrained(model_id) | |
| def transcribe(audio_file_mic=None, audio_file_upload=None, language="English (eng)"): | |
| if audio_file_mic: | |
| audio_file = audio_file_mic | |
| elif audio_file_upload: | |
| audio_file = audio_file_upload | |
| else: | |
| return "Please upload an audio file or record one" | |
| # Make sure audio is 16kHz | |
| speech, sample_rate = librosa.load(audio_file) | |
| if sample_rate != 16000: | |
| speech = librosa.resample(speech, orig_sr=sample_rate, target_sr=16000) | |
| # Keep the same model in memory and simply switch out the language adapters by calling load_adapter() for the model and set_target_lang() for the tokenizer | |
| language_code = iso_codes[language] | |
| processor.tokenizer.set_target_lang(language_code) | |
| model.load_adapter(language_code) | |
| inputs = processor(speech, sampling_rate=16_000, return_tensors="pt") | |
| with torch.no_grad(): | |
| outputs = model(**inputs).logits | |
| ids = torch.argmax(outputs, dim=-1)[0] | |
| transcription = processor.decode(ids) | |
| return transcription | |
| examples = [["balinese.mp3", None, "Bali (ban)"], | |
| ["madura.mp3", None, "Madura (mad)"]] | |
| description = '''Automatic Speech Recognition with [MMS](https://ai.facebook.com/blog/multilingual-model-speech-recognition/) (Massively Multilingual Speech) by Meta. | |
| Supports [1162 languages](https://dl.fbaipublicfiles.com/mms/misc/language_coverage_mms.html). Read the paper for more details: [Scaling Speech Technology to 1,000+ Languages](https://arxiv.org/abs/2305.13516).''' | |
| iface = gr.Interface(fn=transcribe, | |
| inputs=[ | |
| gr.Audio(source="microphone", type="filepath", label="Record Audio"), | |
| gr.Audio(source="upload", type="filepath", label="Upload Audio"), | |
| gr.Dropdown(choices=languages, label="Language", value="English (eng)") | |
| ], | |
| outputs=gr.Textbox(label="Transcription"), | |
| examples=examples, | |
| description=description | |
| ) | |
| iface.launch() | |