add app

app.py

@@ -0,0 +1,73 @@
+import gradio as gr
+from transformers.file_utils import cached_path, hf_bucket_url
+import os, zipfile
+from transformers import Wav2Vec2Processor, Wav2Vec2ForCTC
+from datasets import load_dataset
+import soundfile as sf
+import torch
+import kenlm
+from pyctcdecode import Alphabet, BeamSearchDecoderCTC, LanguageModel
+
+cache_dir = './cache/'
+processor = Wav2Vec2Processor.from_pretrained("nguyenvulebinh/wav2vec2-base-vietnamese-250h", cache_dir=cache_dir)
+model = Wav2Vec2ForCTC.from_pretrained("nguyenvulebinh/wav2vec2-base-vietnamese-250h", cache_dir=cache_dir)
+lm_file = hf_bucket_url("nguyenvulebinh/wav2vec2-base-vietnamese-250h", filename='vi_lm_4grams.bin.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 + 'vi_lm_4grams.bin'
+
+
+def get_decoder_ngram_model(tokenizer, ngram_lm_path):
+    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][:-2]
+    vocab_list = vocab
+    # convert ctc blank character representation
+    vocab_list[tokenizer.pad_token_id] = ""
+    # replace special characters
+    vocab_list[tokenizer.unk_token_id] = ""
+    # vocab_list[tokenizer.bos_token_id] = ""
+    # vocab_list[tokenizer.eos_token_id] = ""
+    # convert space character representation
+    vocab_list[tokenizer.word_delimiter_token_id] = " "
+    # specify ctc blank char index, since conventially it is the last entry of the logit matrix
+    alphabet = Alphabet.build_alphabet(vocab_list, ctc_token_idx=tokenizer.pad_token_id)
+    lm_model = kenlm.Model(ngram_lm_path)
+    decoder = BeamSearchDecoderCTC(alphabet,
+                                   language_model=LanguageModel(lm_model))
+    return decoder
+
+ngram_lm_model = get_decoder_ngram_model(processor.tokenizer, lm_file)
+
+# define function to read in sound file
+def map_to_array(batch):
+    speech, sampling_rate = sf.read(batch["file"])
+    batch["speech"] = speech
+    batch["sampling_rate"] = sampling_rate
+    return batch
+
+# tokenize
+def inference(audio):
+   # read in sound file
+    # load dummy dataset and read soundfiles
+    ds = map_to_array({"file": audio})
+    # infer model
+    input_values = processor(
+          ds["speech"], 
+          sampling_rate=ds["sampling_rate"], 
+          return_tensors="pt"
+    ).input_values
+    # decode ctc output
+    pred_ids = torch.argmax(logits, dim=-1)
+    greedy_search_output = processor.decode(pred_ids)
+    beam_search_output = ngram_lm_model.decode(logits.cpu().detach().numpy(), beam_width=500)
+    return beam_search_output
+
+inputs = gr.inputs.Audio(label="Input Audio", type="file")
+outputs =  gr.outputs.Textbox(label="Output Text")
+title = "wav2vec2-base-vietnamese-250h"
+description = "Gradio demo for a wav2vec2-base-vietnamese-250h. To use it, simply upload your audio, or click one of the examples to load them. Read more at the links below. Currently supports .wav 16_000hz files"
+article = "<p style='text-align: center'><a href='https://github.com/vietai/ASR' target='_blank'> Github repo for demonstration </a> | <a href='https://huggingface.co/nguyenvulebinh/wav2vec2-base-vietnamese-250h' target='_blank'>Pretrained model</a></p>"
+examples=[['t1_0001-00010.wav', 't1_utt000000042.wav', 't2_0000006682.wav']]
+gr.Interface(inference, inputs, outputs, title=title, description=description, article=article, examples=examples).launch()

requirements.txt

@@ -0,0 +1,6 @@
+torch==1.9.0
+transformers==4.9.2 
+soundfile 
+datasets==1.11.0 
+pyctcdecode==v0.1.0
+https://github.com/kpu/kenlm/archive/master.zip