John4064's picture
Upload 4 files
ba4023a
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/Carddata.csv"
#DATASET_REPO_ID = "awacke1/Carddata.csv"
#DATA_FILENAME = "Carddata.csv"
#DATA_FILE = os.path.join("data", DATA_FILENAME)
#HF_TOKEN = os.environ.get("HF_TOKEN")
#SCRIPT = """
#<script>
#if (!window.hasBeenRun) {
# window.hasBeenRun = true;
# console.log("should only happen once");
# document.querySelector("button.submit").click();
#}
#</script>
#"""
#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()
# return ""
#iface = gr.Interface(
# store_message,
# [
# inputs.Textbox(placeholder="Your name"),
# inputs.Textbox(placeholder="Your message", lines=2),
# ],
# "html",
# css="""
# .message {background-color:cornflowerblue;color:white; padding:4px;margin:4px;border-radius:4px; }
# """,
# title="Reading/writing to a HuggingFace dataset repo from Spaces",
# description=f"This is a demo of how to do simple *shared data persistence* in a Gradio Space, backed by a dataset repo.",
# article=f"The dataset repo is [{DATASET_REPO_URL}]({DATASET_REPO_URL})",
#)
# main -------------------------
mname = "facebook/blenderbot-400M-distill"
model = BlenderbotForConditionalGeneration.from_pretrained(mname)
tokenizer = BlenderbotTokenizer.from_pretrained(mname)
def take_last_tokens(inputs, note_history, history):
"""Filter the last 128 tokens"""
if inputs['input_ids'].shape[1] > 128:
inputs['input_ids'] = torch.tensor([inputs['input_ids'][0][-128:].tolist()])
inputs['attention_mask'] = torch.tensor([inputs['attention_mask'][0][-128:].tolist()])
note_history = ['</s> <s>'.join(note_history[0].split('</s> <s>')[2:])]
history = history[1:]
return inputs, note_history, history
def add_note_to_history(note, note_history):
"""Add a note to the historical information"""
note_history.append(note)
note_history = '</s> <s>'.join(note_history)
return [note_history]
def chat(message, history):
history = history or []
if history:
history_useful = ['</s> <s>'.join([str(a[0])+'</s> <s>'+str(a[1]) for a in history])]
else:
history_useful = []
history_useful = add_note_to_history(message, history_useful)
inputs = tokenizer(history_useful, return_tensors="pt")
inputs, history_useful, history = take_last_tokens(inputs, history_useful, history)
reply_ids = model.generate(**inputs)
response = tokenizer.batch_decode(reply_ids, skip_special_tokens=True)[0]
history_useful = add_note_to_history(response, history_useful)
list_history = history_useful[0].split('</s> <s>')
history.append((list_history[-2], list_history[-1]))
# store_message(message, response) # Save to dataset - uncomment if you uncomment above to save inputs and outputs to your dataset
return history, 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)
# monochannel
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]
# store_message(transcriptions, state) # Save to dataset - uncomment to store into a dataset - hint you will need your HF_TOKEN
state = state + transcriptions + " "
return state, state
iface = gr.Interface(
fn=transcribe,
inputs=[
gr.Audio(source="microphone", type='filepath', streaming=True),
"state",
],
outputs=[
"textbox",
"state",
],
layout="horizontal",
theme="huggingface",
title="🗣️LiveSpeechRecognition🧠Memory💾",
description=f"Live Automatic Speech Recognition (ASR) with Memory💾 Dataset.",
allow_flagging='never',
live=True,
# article=f"Result Output Saved to Memory💾 Dataset: [{DATASET_REPO_URL}]({DATASET_REPO_URL})"
)
iface.launch()