deploy new code

app.py

@@ -1,7 +1,35 @@
 import gradio as gr
+import nemo.collections.asr as nemo_asr
+from pydub import AudioSegment
+import pyaudioconvert as pac
+import timeit
 
+hf_model = nemo_asr.models.EncDecRNNTBPEModel.from_pretrained(
+                model_name="mbazaNLP/Kinyarwanda_nemo_stt_conformer_model")
 
-from transformers import pipeline
+def convert (audio):
+  file_name = audio.name
+  if file_name.endswith("mp3") or file_name.endswith("wav") or file_name.endswith("ogg"):
+    if file_name.endswith("mp3"):
+      sound = AudioSegment.from_mp3(audio.name)
+      sound.export(audio.name, format="wav")
+    elif file_name.endswith("ogg"):
+      sound = AudioSegment.from_ogg(audio.name)
+      sound.export(audio.name, format="wav")
+  else:
+    return False
+  pac.convert_wav_to_16bit_mono(audio.name,audio.name)
+  return True
 
-pipe = pipeline("models/mbazaNLP/Kinyarwanda_nemo_stt_conformer_model")
-gr.Interface.from_pipeline(pipe).launch()
+def transcribe(audio):
+  start = timeit.default_timer()
+  if convert(audio)== False:
+    return "The format must be mp3,wav and ogg"
+
+  files = [audio.name]
+  print(audio.name)
+  for fname, transcription in zip(files, hf_model.transcribe(paths2audio_files=files)):
+    stop = timeit.default_timer()
+    return "message"+ transcription[0]+ "\nfilename"+ audio.name+"\nTrancriptionTime"+stop-start
+
+gradio_ui.launch()

package.txt

@@ -0,0 +1,3 @@
+libsndfile1 
+ffmpeg 
+sox

requirements.txt

@@ -1 +1,4 @@
-transformers
+pydub 
+pyaudioconvert
+nemo_toolkit[asr]
+gradio