updated audio reading module

gradio_asr_en_libri100_word_vs_bpe.py

@@ -8,7 +8,9 @@ import os
 from difflib import Differ
 from glob import glob
 from loguru import logger
-import librosa
+# import librosa
+from scipy.io import wavfile
+import scipy.signal as sps
 import gradio as gr
 
 from espnet2.bin.asr_inference import Speech2Text
@@ -43,7 +45,11 @@ logger.info('model loaded')
 def predict(wav_file):
     logger.info('wav file loaded')
     # Load audio    
-    speech, rate = librosa.load(wav_file, sr=16000)
+    # speech, rate = librosa.load(wav_file, sr=16000)
+    rate, speech_raw = wavfile.read(wav_file)
+    number_of_samples = round(len(speech_raw) * float(16000) / rate)
+    speech = sps.resample(speech_raw, number_of_samples)
+    
     # Run inference
     W = model_word(speech)[0]
     B = model_bpe(speech)[0]

requirements.txt

@@ -1,6 +1,6 @@
 espnet==0.10.6
 espnet_model_zoo==0.1.7
-gradio==2.7.5.2
+gradio
 loguru==0.6.0
 librosa
 soundfile