Spaces:
Runtime error
Runtime error
File size: 2,522 Bytes
11fc755 83014ea 11fc755 2800d66 11fc755 beaef93 2feeb15 e71fef9 11fc755 |
1 2 3 4 5 6 7 8 9 10 11 12 13 14 15 16 17 18 19 20 21 22 23 24 25 26 27 28 29 30 31 32 33 34 35 36 37 38 39 40 41 42 43 44 45 46 47 48 49 50 51 52 53 54 55 56 57 58 59 60 61 62 63 64 65 66 67 68 |
#References: 1. https://www.kdnuggets.com/2021/03/speech-text-wav2vec.html
#2. https://www.youtube.com/watch?v=4CoVcsxZphE
#3. https://www.analyticsvidhya.com/blog/2021/02/hugging-face-introduces-the-first-automatic-speech-recognition-model-wav2vec2/
#Importing all the necessary packages
import nltk
import librosa
import torch
import gradio as gr
from transformers import Wav2Vec2Tokenizer, Wav2Vec2ForCTC
nltk.download("punkt")
#Loading the model and the tokenizer
model_name = "bofenghuang/asr-wav2vec2-ctc-french" #"wasertech/wav2vec2-cv-fr-9"
tokenizer = Wav2Vec2Tokenizer.from_pretrained(model_name)
model = Wav2Vec2ForCTC.from_pretrained(model_name)
def load_data(input_file):
""" Function for resampling to ensure that the speech input is sampled at 16KHz.
"""
#read the file
speech, sample_rate = librosa.load(input_file)
#make it 1-D
if len(speech.shape) > 1:
speech = speech[:,0] + speech[:,1]
#Resampling at 16KHz since wav2vec2-base-960h is pretrained and fine-tuned on speech audio sampled at 16 KHz.
if sample_rate !=16000:
speech = librosa.resample(speech, orig_sr=sample_rate, target_sr=16000)
return speech
def correct_casing(input_sentence):
""" This function is for correcting the casing of the generated transcribed text
"""
sentences = nltk.sent_tokenize(input_sentence)
return (' '.join([s.replace(s[0],s[0].capitalize(),1) for s in sentences]))
def asr_transcript(input_file):
"""This function generates transcripts for the provided audio input
"""
speech = load_data(input_file)
#Tokenize
input_values = tokenizer(speech, return_tensors="pt").input_values
#Take logits
logits = model(input_values).logits
#Take argmax
predicted_ids = torch.argmax(logits, dim=-1)
#Get the words from predicted word ids
transcription = tokenizer.decode(predicted_ids[0])
#Output is all upper case
transcription = correct_casing(transcription.lower())
return transcription
gr.Interface(asr_transcript,
inputs = gr.inputs.Audio(source="microphone", type="filepath", optional=True, label="Démarrer l'enregistrement"),
outputs = gr.outputs.Textbox(label="Transcription"),
title="🎙️ Parlez, on vous écoute !",
description = "Enregistrez un audio ou utilisez les examples pour interagir avec notre dernier modèle.",
examples = [["wav/1.wav"], ["wav/2.wav"], ["wav/3.wav"], ["wav/4.wav"], ["wav/5.wav"]], theme="grass").launch()
|