| import soundfile as sf | |
| import torch | |
| from transformers import Wav2Vec2ForCTC, Wav2Vec2Processor | |
| import gradio as gr | |
| import sox | |
| import subprocess | |
| def read_file_and_process(wav_file): | |
| filename = wav_file.split('.')[0] | |
| filename_16k = filename + "16k.wav" | |
| resampler(wav_file, filename_16k) | |
| speech, _ = sf.read(filename_16k) | |
| inputs = processor(speech, sampling_rate=16_000, return_tensors="pt", padding=True) | |
| return inputs | |
| def resampler(input_file_path, output_file_path): | |
| command = ( | |
| f"ffmpeg -hide_banner -loglevel panic -i {input_file_path} -ar 16000 -ac 1 -bits_per_raw_sample 16 -vn " | |
| f"{output_file_path}" | |
| ) | |
| subprocess.call(command, shell=True) | |
| def parse_transcription(logits): | |
| predicted_ids = torch.argmax(logits, dim=-1) | |
| transcription = processor.decode(predicted_ids[0], skip_special_tokens=True) | |
| return transcription | |
| def parse(wav_file): | |
| input_values = read_file_and_process(wav_file) | |
| with torch.no_grad(): | |
| logits = model(**input_values).logits | |
| return parse_transcription(logits) | |
| model_id = "SeyedAli/Persian-Speech-Transcription-Wav2Vec2-V1" | |
| processor = Wav2Vec2Processor.from_pretrained(model_id) | |
| model = Wav2Vec2ForCTC.from_pretrained(model_id) | |
| input_ = gr.Audio() | |
| txtbox = gr.Textbox( | |
| label="persian text output:", | |
| lines=5 | |
| ) | |
| title = "Speech-to-Text (persian)" | |
| description = "Upload a prsian audio, and let AI do the hard work of transcribing." | |
| article = "<p style='text-align: center'><a href='https://arxiv.org/abs/2104.06678'>Large-Scale Self- and Semi-Supervised Learning for Speech Translation</a></p>" | |
| demo = gr.Interface(fn=parse, inputs = input_, outputs=txtbox, title=title, description=description, article = article, | |
| streaming=True, interactive=True, | |
| analytics_enabled=False, show_tips=False, enable_queue=True) | |
| demo.launch(share=True) |