import gradio as gr import torch import time import librosa import soundfile import nemo.collections.asr as nemo_asr import tempfile import os import uuid from transformers import BlenderbotTokenizer, BlenderbotForConditionalGeneration import torch # PersistDataset ----- import os import csv import gradio as gr from gradio import inputs, outputs import huggingface_hub from huggingface_hub import Repository, hf_hub_download, upload_file from datetime import datetime # --------------------------------------------- # Dataset and Token links - change awacke1 to your own HF id, and add a HF_TOKEN copy to your repo for write permissions # This should allow you to save your results to your own Dataset hosted on HF. DATASET_REPO_URL = "https://huggingface.co/datasets/awacke1/ASRLive.csv" DATASET_REPO_ID = "awacke1/ASRLive.csv" DATA_FILENAME = "ASRLive.csv" DATA_FILE = os.path.join("data", DATA_FILENAME) HF_TOKEN = os.environ.get("HF_TOKEN") PersistToDataset = False #PersistToDataset = True # uncomment to save inference output to ASRLive.csv dataset if PersistToDataset: try: hf_hub_download( repo_id=DATASET_REPO_ID, filename=DATA_FILENAME, cache_dir=DATA_DIRNAME, force_filename=DATA_FILENAME ) except: print("file not found") repo = Repository( local_dir="data", clone_from=DATASET_REPO_URL, use_auth_token=HF_TOKEN ) def store_message(name: str, message: str): if name and message: with open(DATA_FILE, "a") as csvfile: writer = csv.DictWriter(csvfile, fieldnames=["name", "message", "time"]) writer.writerow( {"name": name.strip(), "message": message.strip(), "time": str(datetime.now())} ) # uncomment line below to begin saving - commit_url = repo.push_to_hub() ret = "" with open(DATA_FILE, "r") as csvfile: reader = csv.DictReader(csvfile) for row in reader: ret += row ret += "\r\n" return ret # main ------------------------- mname = "facebook/blenderbot-400M-distill" model = BlenderbotForConditionalGeneration.from_pretrained(mname) tokenizer = BlenderbotTokenizer.from_pretrained(mname) def take_last_tokens(inputs, note_history, history): filterTokenCount = 128 # filter last 128 tokens if inputs['input_ids'].shape[1] > filterTokenCount: inputs['input_ids'] = torch.tensor([inputs['input_ids'][0][-filterTokenCount:].tolist()]) inputs['attention_mask'] = torch.tensor([inputs['attention_mask'][0][-filterTokenCount:].tolist()]) note_history = [' '.join(note_history[0].split(' ')[2:])] history = history[1:] return inputs, note_history, history def add_note_to_history(note, note_history): note_history.append(note) note_history = ' '.join(note_history) return [note_history] SAMPLE_RATE = 16000 model = nemo_asr.models.EncDecRNNTBPEModel.from_pretrained("nvidia/stt_en_conformer_transducer_xlarge") model.change_decoding_strategy(None) model.eval() def process_audio_file(file): data, sr = librosa.load(file) if sr != SAMPLE_RATE: data = librosa.resample(data, orig_sr=sr, target_sr=SAMPLE_RATE) data = librosa.to_mono(data) return data def transcribe(audio, state = ""): if state is None: state = "" audio_data = process_audio_file(audio) with tempfile.TemporaryDirectory() as tmpdir: audio_path = os.path.join(tmpdir, f'audio_{uuid.uuid4()}.wav') soundfile.write(audio_path, audio_data, SAMPLE_RATE) transcriptions = model.transcribe([audio_path]) if type(transcriptions) == tuple and len(transcriptions) == 2: transcriptions = transcriptions[0] transcriptions = transcriptions[0] if PersistToDataset: ret = store_message(transcriptions, state) # Save to dataset - uncomment to store into a dataset - hint you will need your HF_TOKEN state = state + transcriptions + " " + ret else: state = state + transcriptions return state, state gr.Interface( fn=transcribe, inputs=[ gr.Audio(source="microphone", type='filepath', streaming=True), "state", ], outputs=[ "textbox", "state" ], layout="horizontal", theme="huggingface", title="🗣️ASR-Gradio-Live🧠💾", description=f"Live Automatic Speech Recognition (ASR).", allow_flagging='never', live=True, article=f"Result💾 Dataset: [{DATASET_REPO_URL}]({DATASET_REPO_URL})" ).launch(debug=True)