Spaces:
Runtime error
Runtime error
File size: 3,284 Bytes
7bfa718 4104468 7bfa718 4104468 7bfa718 4104468 |
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 69 70 71 72 73 74 75 76 77 78 79 80 81 82 83 84 85 |
import gradio as gr
import torch
import zipfile
from pyctcdecode import build_ctcdecoder
from speechbrain.pretrained import EncoderASR
from transformers.file_utils import cached_path, hf_bucket_url
cache_dir = './cache/'
lm_file = hf_bucket_url(
"dragonSwing/wav2vec2-base-vn-270h", filename='4gram.zip')
lm_file = cached_path(lm_file, cache_dir=cache_dir)
with zipfile.ZipFile(lm_file, 'r') as zip_ref:
zip_ref.extractall(cache_dir)
lm_file = cache_dir + 'lm.binary'
vocab_file = cache_dir + 'vocab-260000.txt'
model = EncoderASR.from_hparams(source="dragonSwing/wav2vec2-base-vn-270h",
savedir="/content/pretrained2/"
)
def get_decoder_ngram_model(tokenizer, ngram_lm_path, vocab_path=None):
unigrams = None
if vocab_path is not None:
unigrams = []
with open(vocab_path, encoding='utf-8') as f:
for line in f:
unigrams.append(line.strip())
vocab_dict = tokenizer.get_vocab()
sort_vocab = sorted((value, key) for (key, value) in vocab_dict.items())
vocab = [x[1] for x in sort_vocab]
vocab_list = vocab
# convert ctc blank character representation
vocab_list[tokenizer.pad_token_id] = ""
# replace special characters
vocab_list[tokenizer.word_delimiter_token_id] = " "
# specify ctc blank char index, since conventially it is the last entry of the logit matrix
decoder = build_ctcdecoder(vocab_list, ngram_lm_path, unigrams=unigrams)
return decoder
ngram_lm_model = get_decoder_ngram_model(model.tokenizer, lm_file, vocab_file)
def transcribe_file(path, max_seconds=20):
waveform = model.load_audio(path)
if max_seconds > 0:
waveform = waveform[:max_seconds*16000]
batch = waveform.unsqueeze(0)
rel_length = torch.tensor([1.0])
with torch.no_grad():
logits = model(batch, rel_length)
text_batch = [ngram_lm_model.decode(
logit.detach().cpu().numpy(), beam_width=500) for logit in logits]
return text_batch[0]
def speech_recognize(file_upload, file_mic):
if file_upload is not None:
file = file_upload
elif file_mic is not None:
file = file_mic
else:
return ""
# text = model.transcribe_file(file)
text = transcribe_file(file)
return text
inputs = [gr.inputs.Audio(source="upload", type='filepath', optional=True), gr.inputs.Audio(
source="microphone", type='filepath', optional=True)]
outputs = gr.outputs.Textbox(label="Output Text")
title = "wav2vec2-base-vietnamese-270h"
description = "Gradio demo for a wav2vec2 base vietnamese speech recognition. To use it, simply upload your audio, click one of the examples to load them, or record from your own microphone. Read more at the links below. Currently supports 16_000hz audio files"
article = "<p style='text-align: center'><a href='https://huggingface.co/dragonSwing/wav2vec2-base-vn-270h' target='_blank'>Pretrained model</a></p>"
examples = [
['example1.wav', 'example1.wav'],
['example2.mp3', 'example2.mp3'],
['example3.mp3', 'example3.mp3'],
['example4.wav', 'example4.wav'],
]
gr.Interface(speech_recognize, inputs, outputs, title=title,
description=description, article=article, examples=examples).launch()
|